diff --git a/.sf.migrating/milestones/M001-6377a4/M001-6377a4-ROADMAP.json b/.sf.migrating/milestones/M001-6377a4/M001-6377a4-ROADMAP.json
index dc0764a0c..7d4276d6e 100644
--- a/.sf.migrating/milestones/M001-6377a4/M001-6377a4-ROADMAP.json
+++ b/.sf.migrating/milestones/M001-6377a4/M001-6377a4-ROADMAP.json
@@ -1,102 +1,96 @@
 {
-  "schemaVersion": 1,
-  "origin": "db-projection",
-  "generatedAt": "2026-05-06T21:03:18.498Z",
-  "sourceDbPath": "/home/mhugo/code/singularity-forge/.sf/sf.db",
-  "milestone": {
-    "id": "M001-6377a4",
-    "title": "Consolidate Memory Systems into Unified node:sqlite Store",
-    "status": "active",
-    "vision": "Replace the three fragmented memory systems (KNOWLEDGE.md, memory-store.js/sf.db, and memory extension/sql.js) with a single unified memory store backed by node:sqlite. All memory ingestion, querying, and prompt injection flows through one canonical database table with semantic search, confidence ranking, and automatic maintenance.",
-    "dependsOn": [],
-    "successCriteria": [
-      "A single `memories` table in `sf.db` stores all project memory with categories, confidence, hit_count, and embeddings",
-      "Session transcript extraction pipeline writes directly to `sf.db` instead of separate `agent.db`",
-      "KNOWLEDGE.md is retired as a live system; existing entries are optionally migrated",
-      "Memory query returns ranked results via `getRelevantMemoriesRanked(query, limit)` using node:sqlite",
-      "System prompt injection uses the unified memory store, not multiple competing sources",
-      "All better-sqlite3 and sql.js dependencies for memory are removed"
-    ],
-    "definitionOfDone": [
-      "All three legacy memory systems are removed or deprecated",
-      "Unified memory store passes all new tests",
-      "Migration script can import existing KNOWLEDGE.md and old agent.db data",
-      "Documentation updated to describe the single memory system",
-      "No references to better-sqlite3 or sql.js in memory-related code"
-    ],
-    "requirementCoverage": "Covers: unified memory storage, semantic search, session-based learning. Partially covers: cross-session context persistence. Leaves for later: multi-project memory sharing, cloud-synced memory.",
-    "boundaryMapMarkdown": "### S01 → S02\n\nProduces:\n- Unified `memories` table schema in `sf.db`\n- `MemoryStore` class with CRUD, query, and ranking API\n- `formatMemoriesForPrompt()` for system prompt injection\n\nConsumes:\n- Existing `sf-db.js` adapter and schema initialization\n\n### S02 → S03\n\nProduces:\n- Session transcript scanner and LLM extraction pipeline\n- Direct write path to `sf.db` memories table\n- `/memory` command reading from unified store\n\nConsumes:\n- `MemoryStore` API from S01\n- `completeSimple` LLM call from pi-ai\n\n### S03 → S04\n\nProduces:\n- KNOWLEDGE.md importer\n- Legacy agent.db migration script\n- Deprecation warnings for old memory paths\n\nConsumes:\n- `MemoryStore` API from S01\n- Existing KNOWLEDGE.md and agent.db files"
-  },
-  "slices": [
-    {
-      "id": "S01",
-      "title": "Unified Memory Schema and Core API",
-      "status": "pending",
-      "risk": "high",
-      "depends": [],
-      "demo": "A test creates a memory, queries it by relevance, and formats it for prompt injection — all via node:sqlite",
-      "goal": "Create the canonical `memories` table in `sf.db` and a `MemoryStore` class that replaces memory-store.js with a node:sqlite implementation",
-      "successCriteria": "- `sf.db` has a `memories` table with: id, category, content, confidence, hit_count, source, session_id, source_unit_type, source_unit_id, created_at, updated_at, superseded_by\n- `MemoryStore` class exposes: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n- All methods gracefully degrade when DB is unavailable\n- Unit tests cover CRUD, ranking, source filtering, and embedding fallback",
-      "proofLevel": "contract",
-      "integrationClosure": "Produces the MemoryStore API that all downstream slices consume. No external wiring yet.",
-      "observabilityImpact": "No runtime signals added yet; tests verify table schema and query correctness.",
-      "isSketch": false,
-      "sketchScope": "",
-      "sequence": 1
-    },
-    {
-      "id": "S02",
-      "title": "Session Transcript Extraction Pipeline",
-      "status": "pending",
-      "risk": "medium",
-      "depends": [
-        "S01"
-      ],
-      "demo": "After running a session, the pipeline extracts memories and stores them in sf.db; `/memory view` shows the extracted knowledge",
-      "goal": "Port the two-phase extraction pipeline from the memory extension to write directly into the unified sf.db store instead of separate agent.db",
-      "successCriteria": "- Session .jsonl files are scanned and filtered for the current cwd\n- Phase 1: LLM extracts durable knowledge from transcripts, writes to sf.db memories table with source='extracted'\n- Phase 2: Consolidation deduplicates and ranks memories\n- /memory command reads from unified store\n- Pipeline runs fire-and-forget on session start without blocking dispatch\n- Rate limiting prevents store bloat (max 10 memories per session)",
-      "proofLevel": "integration",
-      "integrationClosure": "Wires the extraction pipeline into the session lifecycle and replaces the old memory extension storage path.",
-      "observabilityImpact": "Pipeline logs extraction stats (processed, errors) to the console. Memory store tracks hit_count for reinforcement.",
-      "isSketch": false,
-      "sketchScope": "",
-      "sequence": 2
-    },
-    {
-      "id": "S03",
-      "title": "Migration and Legacy Deprecation",
-      "status": "pending",
-      "risk": "low",
-      "depends": [
-        "S02"
-      ],
-      "demo": "Running a migration command imports existing KNOWLEDGE.md and agent.db memories into sf.db; old files are left as read-only backups",
-      "goal": "Provide a one-time migration path from KNOWLEDGE.md and the old agent.db into the unified store, then deprecate the legacy systems",
-      "successCriteria": "- Migration script can import KNOWLEDGE.md entries as memories with source='migrated'\n- Migration script can import agent.db stage1_outputs as memories with source='extracted'\n- Dry-run mode reports counts without modifying database\n- sf.db is automatically backed up before migration\n- knowledge-compounding.js and knowledge-injector.js are removed or no-op'd\n- Memory extension uses MemoryStore instead of sql.js\n- CHANGELOG documents the consolidation and deprecation",
-      "proofLevel": "contract",
-      "integrationClosure": "Cleans up legacy code paths and provides a safe migration for existing users.",
-      "observabilityImpact": "Migration script reports counts of imported records per source. Deprecation warnings logged for old API usage.",
-      "isSketch": false,
-      "sketchScope": "",
-      "sequence": 3
-    },
-    {
-      "id": "S04",
-      "title": "System Prompt Integration and Cleanup",
-      "status": "pending",
-      "risk": "medium",
-      "depends": [
-        "S03"
-      ],
-      "demo": "Starting a new session injects relevant ranked memories from sf.db into the system prompt; no references to old memory systems remain in the codebase",
-      "goal": "Replace all system prompt injection paths to use the unified memory store, remove dead code, and verify end-to-end",
-      "successCriteria": "Not provided.",
-      "proofLevel": "Not provided.",
-      "integrationClosure": "Not provided.",
-      "observabilityImpact": "Not provided.",
-      "isSketch": false,
-      "sketchScope": "",
-      "sequence": 4
-    }
-  ]
+	"schemaVersion": 1,
+	"origin": "db-projection",
+	"generatedAt": "2026-05-06T21:03:18.498Z",
+	"sourceDbPath": "/home/mhugo/code/singularity-forge/.sf/sf.db",
+	"milestone": {
+		"id": "M001-6377a4",
+		"title": "Consolidate Memory Systems into Unified node:sqlite Store",
+		"status": "active",
+		"vision": "Replace the three fragmented memory systems (KNOWLEDGE.md, memory-store.js/sf.db, and memory extension/sql.js) with a single unified memory store backed by node:sqlite. All memory ingestion, querying, and prompt injection flows through one canonical database table with semantic search, confidence ranking, and automatic maintenance.",
+		"dependsOn": [],
+		"successCriteria": [
+			"A single `memories` table in `sf.db` stores all project memory with categories, confidence, hit_count, and embeddings",
+			"Session transcript extraction pipeline writes directly to `sf.db` instead of separate `agent.db`",
+			"KNOWLEDGE.md is retired as a live system; existing entries are optionally migrated",
+			"Memory query returns ranked results via `getRelevantMemoriesRanked(query, limit)` using node:sqlite",
+			"System prompt injection uses the unified memory store, not multiple competing sources",
+			"All better-sqlite3 and sql.js dependencies for memory are removed"
+		],
+		"definitionOfDone": [
+			"All three legacy memory systems are removed or deprecated",
+			"Unified memory store passes all new tests",
+			"Migration script can import existing KNOWLEDGE.md and old agent.db data",
+			"Documentation updated to describe the single memory system",
+			"No references to better-sqlite3 or sql.js in memory-related code"
+		],
+		"requirementCoverage": "Covers: unified memory storage, semantic search, session-based learning. Partially covers: cross-session context persistence. Leaves for later: multi-project memory sharing, cloud-synced memory.",
+		"boundaryMapMarkdown": "### S01 → S02\n\nProduces:\n- Unified `memories` table schema in `sf.db`\n- `MemoryStore` class with CRUD, query, and ranking API\n- `formatMemoriesForPrompt()` for system prompt injection\n\nConsumes:\n- Existing `sf-db.js` adapter and schema initialization\n\n### S02 → S03\n\nProduces:\n- Session transcript scanner and LLM extraction pipeline\n- Direct write path to `sf.db` memories table\n- `/memory` command reading from unified store\n\nConsumes:\n- `MemoryStore` API from S01\n- `completeSimple` LLM call from pi-ai\n\n### S03 → S04\n\nProduces:\n- KNOWLEDGE.md importer\n- Legacy agent.db migration script\n- Deprecation warnings for old memory paths\n\nConsumes:\n- `MemoryStore` API from S01\n- Existing KNOWLEDGE.md and agent.db files"
+	},
+	"slices": [
+		{
+			"id": "S01",
+			"title": "Unified Memory Schema and Core API",
+			"status": "pending",
+			"risk": "high",
+			"depends": [],
+			"demo": "A test creates a memory, queries it by relevance, and formats it for prompt injection — all via node:sqlite",
+			"goal": "Create the canonical `memories` table in `sf.db` and a `MemoryStore` class that replaces memory-store.js with a node:sqlite implementation",
+			"successCriteria": "- `sf.db` has a `memories` table with: id, category, content, confidence, hit_count, source, session_id, source_unit_type, source_unit_id, created_at, updated_at, superseded_by\n- `MemoryStore` class exposes: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n- All methods gracefully degrade when DB is unavailable\n- Unit tests cover CRUD, ranking, source filtering, and embedding fallback",
+			"proofLevel": "contract",
+			"integrationClosure": "Produces the MemoryStore API that all downstream slices consume. No external wiring yet.",
+			"observabilityImpact": "No runtime signals added yet; tests verify table schema and query correctness.",
+			"isSketch": false,
+			"sketchScope": "",
+			"sequence": 1
+		},
+		{
+			"id": "S02",
+			"title": "Session Transcript Extraction Pipeline",
+			"status": "pending",
+			"risk": "medium",
+			"depends": ["S01"],
+			"demo": "After running a session, the pipeline extracts memories and stores them in sf.db; `/memory view` shows the extracted knowledge",
+			"goal": "Port the two-phase extraction pipeline from the memory extension to write directly into the unified sf.db store instead of separate agent.db",
+			"successCriteria": "- Session .jsonl files are scanned and filtered for the current cwd\n- Phase 1: LLM extracts durable knowledge from transcripts, writes to sf.db memories table with source='extracted'\n- Phase 2: Consolidation deduplicates and ranks memories\n- /memory command reads from unified store\n- Pipeline runs fire-and-forget on session start without blocking dispatch\n- Rate limiting prevents store bloat (max 10 memories per session)",
+			"proofLevel": "integration",
+			"integrationClosure": "Wires the extraction pipeline into the session lifecycle and replaces the old memory extension storage path.",
+			"observabilityImpact": "Pipeline logs extraction stats (processed, errors) to the console. Memory store tracks hit_count for reinforcement.",
+			"isSketch": false,
+			"sketchScope": "",
+			"sequence": 2
+		},
+		{
+			"id": "S03",
+			"title": "Migration and Legacy Deprecation",
+			"status": "pending",
+			"risk": "low",
+			"depends": ["S02"],
+			"demo": "Running a migration command imports existing KNOWLEDGE.md and agent.db memories into sf.db; old files are left as read-only backups",
+			"goal": "Provide a one-time migration path from KNOWLEDGE.md and the old agent.db into the unified store, then deprecate the legacy systems",
+			"successCriteria": "- Migration script can import KNOWLEDGE.md entries as memories with source='migrated'\n- Migration script can import agent.db stage1_outputs as memories with source='extracted'\n- Dry-run mode reports counts without modifying database\n- sf.db is automatically backed up before migration\n- knowledge-compounding.js and knowledge-injector.js are removed or no-op'd\n- Memory extension uses MemoryStore instead of sql.js\n- CHANGELOG documents the consolidation and deprecation",
+			"proofLevel": "contract",
+			"integrationClosure": "Cleans up legacy code paths and provides a safe migration for existing users.",
+			"observabilityImpact": "Migration script reports counts of imported records per source. Deprecation warnings logged for old API usage.",
+			"isSketch": false,
+			"sketchScope": "",
+			"sequence": 3
+		},
+		{
+			"id": "S04",
+			"title": "System Prompt Integration and Cleanup",
+			"status": "pending",
+			"risk": "medium",
+			"depends": ["S03"],
+			"demo": "Starting a new session injects relevant ranked memories from sf.db into the system prompt; no references to old memory systems remain in the codebase",
+			"goal": "Replace all system prompt injection paths to use the unified memory store, remove dead code, and verify end-to-end",
+			"successCriteria": "Not provided.",
+			"proofLevel": "Not provided.",
+			"integrationClosure": "Not provided.",
+			"observabilityImpact": "Not provided.",
+			"isSketch": false,
+			"sketchScope": "",
+			"sequence": 4
+		}
+	]
 }
diff --git a/.sf.migrating/runtime/sift-index-warmup.json b/.sf.migrating/runtime/sift-index-warmup.json
index ab0bbc95c..4d50f992a 100644
--- a/.sf.migrating/runtime/sift-index-warmup.json
+++ b/.sf.migrating/runtime/sift-index-warmup.json
@@ -1,28 +1,28 @@
 {
-  "schemaVersion": 3,
-  "status": "warming",
-  "startedAt": "2026-05-06T21:03:03.705Z",
-  "command": "/nix/store/vzx1mi9c0xfadmsm9dhd83d005cb1qs9-coreutils-9.8/bin/timeout",
-  "cwd": "/home/mhugo/code/singularity-forge",
-  "args": [
-    "--kill-after=10",
-    "600",
-    "/home/mhugo/.cargo/bin/sift",
-    "search",
-    "--json",
-    "--strategy",
-    "page-index-hybrid",
-    "--limit",
-    "1",
-    "--retriever-timeout-ms",
-    "30000",
-    ".",
-    "repo architecture source tests entrypoints configuration"
-  ],
-  "scope": ".",
-  "siftBinary": "/home/mhugo/.cargo/bin/sift",
-  "hardTimeoutSec": 600,
-  "searchCache": "/home/mhugo/code/singularity-forge/.sf/runtime/sift/search-cache",
-  "tmpDir": "/home/mhugo/code/singularity-forge/.sf/runtime/sift/tmp",
-  "pid": 602029
+	"schemaVersion": 3,
+	"status": "warming",
+	"startedAt": "2026-05-06T21:03:03.705Z",
+	"command": "/nix/store/vzx1mi9c0xfadmsm9dhd83d005cb1qs9-coreutils-9.8/bin/timeout",
+	"cwd": "/home/mhugo/code/singularity-forge",
+	"args": [
+		"--kill-after=10",
+		"600",
+		"/home/mhugo/.cargo/bin/sift",
+		"search",
+		"--json",
+		"--strategy",
+		"page-index-hybrid",
+		"--limit",
+		"1",
+		"--retriever-timeout-ms",
+		"30000",
+		".",
+		"repo architecture source tests entrypoints configuration"
+	],
+	"scope": ".",
+	"siftBinary": "/home/mhugo/.cargo/bin/sift",
+	"hardTimeoutSec": 600,
+	"searchCache": "/home/mhugo/code/singularity-forge/.sf/runtime/sift/search-cache",
+	"tmpDir": "/home/mhugo/code/singularity-forge/.sf/runtime/sift/tmp",
+	"pid": 602029
 }
diff --git a/.sf.migrating/runtime/uok-diagnostics.json b/.sf.migrating/runtime/uok-diagnostics.json
index 60a166a91..37a3f5c0d 100644
--- a/.sf.migrating/runtime/uok-diagnostics.json
+++ b/.sf.migrating/runtime/uok-diagnostics.json
@@ -1,19 +1,19 @@
 {
-  "schemaVersion": 1,
-  "generatedAt": "2026-05-06T21:03:15.738Z",
-  "verdict": "clear",
-  "classification": "healthy",
-  "signals": {
-    "lock": "missing",
-    "parity": "ok",
-    "ledger": "consistent",
-    "runtimeProjection": "ok",
-    "wrapper": "unknown"
-  },
-  "currentUnit": null,
-  "latestRun": null,
-  "runtimeUnits": [],
-  "issues": [],
-  "recommendations": [],
-  "reportPath": "/home/mhugo/code/singularity-forge/.sf/runtime/uok-diagnostics.json"
+	"schemaVersion": 1,
+	"generatedAt": "2026-05-06T21:03:15.738Z",
+	"verdict": "clear",
+	"classification": "healthy",
+	"signals": {
+		"lock": "missing",
+		"parity": "ok",
+		"ledger": "consistent",
+		"runtimeProjection": "ok",
+		"wrapper": "unknown"
+	},
+	"currentUnit": null,
+	"latestRun": null,
+	"runtimeUnits": [],
+	"issues": [],
+	"recommendations": [],
+	"reportPath": "/home/mhugo/code/singularity-forge/.sf/runtime/uok-diagnostics.json"
 }
diff --git a/.sf.migrating/runtime/uok-parity-report.json b/.sf.migrating/runtime/uok-parity-report.json
index 9774d9961..71449efb1 100644
--- a/.sf.migrating/runtime/uok-parity-report.json
+++ b/.sf.migrating/runtime/uok-parity-report.json
@@ -1,27 +1,27 @@
 {
-  "generatedAt": "2026-05-06T21:03:18.493Z",
-  "sourcePath": "/home/mhugo/code/singularity-forge/.sf/runtime/uok-parity.jsonl",
-  "totalEvents": 0,
-  "paths": {},
-  "statuses": {},
-  "criticalMismatches": [],
-  "historicalCriticalMismatches": [],
-  "currentErrorEvents": 0,
-  "historicalErrorEvents": 0,
-  "enterEvents": 0,
-  "exitEvents": 0,
-  "missingExitEvents": 0,
-  "legacyMissingExitEvents": 0,
-  "ledgerRunCount": 0,
-  "unmatchedRuns": [],
-  "freshUnmatchedRuns": [],
-  "historicalUnmatchedRuns": [],
-  "totalDiffs": 0,
-  "divergencesByPlane": {
-    "plan": 0,
-    "graph": 0,
-    "model-policy": 0,
-    "audit-envelope": 0,
-    "gitops": 0
-  }
+	"generatedAt": "2026-05-06T21:03:18.493Z",
+	"sourcePath": "/home/mhugo/code/singularity-forge/.sf/runtime/uok-parity.jsonl",
+	"totalEvents": 0,
+	"paths": {},
+	"statuses": {},
+	"criticalMismatches": [],
+	"historicalCriticalMismatches": [],
+	"currentErrorEvents": 0,
+	"historicalErrorEvents": 0,
+	"enterEvents": 0,
+	"exitEvents": 0,
+	"missingExitEvents": 0,
+	"legacyMissingExitEvents": 0,
+	"ledgerRunCount": 0,
+	"unmatchedRuns": [],
+	"freshUnmatchedRuns": [],
+	"historicalUnmatchedRuns": [],
+	"totalDiffs": 0,
+	"divergencesByPlane": {
+		"plan": 0,
+		"graph": 0,
+		"model-policy": 0,
+		"audit-envelope": 0,
+		"gitops": 0
+	}
 }
diff --git a/.sf.migrating/scaffold-manifest.json b/.sf.migrating/scaffold-manifest.json
index 7a940f406..d73f628fe 100644
--- a/.sf.migrating/scaffold-manifest.json
+++ b/.sf.migrating/scaffold-manifest.json
@@ -1,181 +1,181 @@
 {
-  "schemaVersion": 1,
-  "applied": [
-    {
-      "path": "src/AGENTS.md",
-      "template": "src/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:ac5f874be887aed0bd29105a7c3b7e9269b4b3c0cc7b5b1506da0994c466cf01"
-    },
-    {
-      "path": "tests/AGENTS.md",
-      "template": "tests/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:dab2b9a31c2c48e07e4913cd8744be8553f279ea516ebc138730301ae401a683"
-    },
-    {
-      "path": "docs/AGENTS.md",
-      "template": "docs/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:b35804ce78ca309cab8769719f6e0738141f1121682fbd46490419abd2c6f870"
-    },
-    {
-      "path": "docs/records/AGENTS.md",
-      "template": "docs/records/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:dc21117dfa7607d7ce4cc6ce5724658348a95e9807673ff526b9cf02e2568de0"
-    },
-    {
-      "path": "docs/RECORDS_KEEPER.md",
-      "template": "docs/RECORDS_KEEPER.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:3872de9cd72bd9129814a5e77e3b86abe76bef33f3ca34e04ae7582b4cfd066a"
-    },
-    {
-      "path": "docs/design-docs/AGENTS.md",
-      "template": "docs/design-docs/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:707d43067edcfd33dc0a11db32d383f89739e187f95da4bda6b94575a9e3d272"
-    },
-    {
-      "path": "docs/exec-plans/AGENTS.md",
-      "template": "docs/exec-plans/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:7bd9f815617cb42b1f2c85e6d557986a2f1e7fbc081fcc4e71f3d14cecefc5f0"
-    },
-    {
-      "path": "docs/generated/db-schema.md",
-      "template": "docs/generated/db-schema.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:8488a607c1a2981654a3b030600d2e10627d132ebd0c75700648a08dede93368"
-    },
-    {
-      "path": "docs/product-specs/index.md",
-      "template": "docs/product-specs/index.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:ca3477e8d74fe277a2e0b2cdb3f03c235e294015a6ece2f571a82acc7475d31c"
-    },
-    {
-      "path": "docs/product-specs/AGENTS.md",
-      "template": "docs/product-specs/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:cc7b5687e8b1ea78848ee318ca5a25346e5a6ee0dc33b54506d8f56bc16f13c4"
-    },
-    {
-      "path": "docs/references/design-system-reference-llms.txt",
-      "template": "docs/references/design-system-reference-llms.txt",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:5a5a35a3f80c8b4433ad30c1f155b1e8c7fd245ce2a3def9627daa9f40854eb3"
-    },
-    {
-      "path": "docs/references/nixpacks-llms.txt",
-      "template": "docs/references/nixpacks-llms.txt",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:22f9a8549e3ced71d0b0a912c6dcdfb2ec83a573168ee1b44ca266f1eb0307bf"
-    },
-    {
-      "path": "docs/references/uv-llms.txt",
-      "template": "docs/references/uv-llms.txt",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:e8a998667c0f830a15b68e207f6b69e6377dd7e82728833f842678f72864e9b6"
-    },
-    {
-      "path": "docs/FRONTEND.md",
-      "template": "docs/FRONTEND.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:03087953d690c9902d35297720d1482262c1610e3050084f891db3be711571ef"
-    },
-    {
-      "path": ".sf/harness/AGENTS.md",
-      "template": ".sf/harness/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:685c41e601340086b8076263a71315c66554efdaeb074bc1b907eebf879174c6"
-    },
-    {
-      "path": ".sf/harness/specs/AGENTS.md",
-      "template": ".sf/harness/specs/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:0f4fbf4111704d05744e4a4e13a9bf3eada262f0da9517c2010f0b46f4bd3c45"
-    },
-    {
-      "path": ".sf/harness/specs/bootstrap.md",
-      "template": ".sf/harness/specs/bootstrap.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:b86ba7cf2cec39a7a9f9d94f885998cfe26eebfc5b76fdd8375ef125e927e0cf"
-    },
-    {
-      "path": ".sf/harness/evals/AGENTS.md",
-      "template": ".sf/harness/evals/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:6f88bf8a2bad95d8db5985c9b3317b9edd65592c12e98bb0bff1a24ec152d768"
-    },
-    {
-      "path": ".sf/harness/graders/AGENTS.md",
-      "template": ".sf/harness/graders/AGENTS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:2db17feae1acfe62d85aafbe32d016873c3036d4d76e9dd0db478375fae0794e"
-    },
-    {
-      "path": ".sf/PRINCIPLES.md",
-      "template": ".sf/PRINCIPLES.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:9d5c50cb3d602f66468a33a4324068fab8a022fab0fd6940c371a5986af2947e"
-    },
-    {
-      "path": ".sf/TASTE.md",
-      "template": ".sf/TASTE.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:ee9ea8ade6f8434e9425a989bab9735d2068395978d61940e18679b93b1f11b0"
-    },
-    {
-      "path": ".sf/ANTI-GOALS.md",
-      "template": ".sf/ANTI-GOALS.md",
-      "version": "2.75.3",
-      "appliedAt": "2026-05-06T19:53:05.336Z",
-      "stateAtApply": "pending",
-      "contentHash": "sha256:56b4f7d20d49e4558ac5aea5e967cf3475ef98d68dfa8985ba42b3e2b0d72e25"
-    }
-  ]
+	"schemaVersion": 1,
+	"applied": [
+		{
+			"path": "src/AGENTS.md",
+			"template": "src/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:ac5f874be887aed0bd29105a7c3b7e9269b4b3c0cc7b5b1506da0994c466cf01"
+		},
+		{
+			"path": "tests/AGENTS.md",
+			"template": "tests/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:dab2b9a31c2c48e07e4913cd8744be8553f279ea516ebc138730301ae401a683"
+		},
+		{
+			"path": "docs/AGENTS.md",
+			"template": "docs/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:b35804ce78ca309cab8769719f6e0738141f1121682fbd46490419abd2c6f870"
+		},
+		{
+			"path": "docs/records/AGENTS.md",
+			"template": "docs/records/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:dc21117dfa7607d7ce4cc6ce5724658348a95e9807673ff526b9cf02e2568de0"
+		},
+		{
+			"path": "docs/RECORDS_KEEPER.md",
+			"template": "docs/RECORDS_KEEPER.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:3872de9cd72bd9129814a5e77e3b86abe76bef33f3ca34e04ae7582b4cfd066a"
+		},
+		{
+			"path": "docs/design-docs/AGENTS.md",
+			"template": "docs/design-docs/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:707d43067edcfd33dc0a11db32d383f89739e187f95da4bda6b94575a9e3d272"
+		},
+		{
+			"path": "docs/exec-plans/AGENTS.md",
+			"template": "docs/exec-plans/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:7bd9f815617cb42b1f2c85e6d557986a2f1e7fbc081fcc4e71f3d14cecefc5f0"
+		},
+		{
+			"path": "docs/generated/db-schema.md",
+			"template": "docs/generated/db-schema.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:8488a607c1a2981654a3b030600d2e10627d132ebd0c75700648a08dede93368"
+		},
+		{
+			"path": "docs/product-specs/index.md",
+			"template": "docs/product-specs/index.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:ca3477e8d74fe277a2e0b2cdb3f03c235e294015a6ece2f571a82acc7475d31c"
+		},
+		{
+			"path": "docs/product-specs/AGENTS.md",
+			"template": "docs/product-specs/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:cc7b5687e8b1ea78848ee318ca5a25346e5a6ee0dc33b54506d8f56bc16f13c4"
+		},
+		{
+			"path": "docs/references/design-system-reference-llms.txt",
+			"template": "docs/references/design-system-reference-llms.txt",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:5a5a35a3f80c8b4433ad30c1f155b1e8c7fd245ce2a3def9627daa9f40854eb3"
+		},
+		{
+			"path": "docs/references/nixpacks-llms.txt",
+			"template": "docs/references/nixpacks-llms.txt",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:22f9a8549e3ced71d0b0a912c6dcdfb2ec83a573168ee1b44ca266f1eb0307bf"
+		},
+		{
+			"path": "docs/references/uv-llms.txt",
+			"template": "docs/references/uv-llms.txt",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:e8a998667c0f830a15b68e207f6b69e6377dd7e82728833f842678f72864e9b6"
+		},
+		{
+			"path": "docs/FRONTEND.md",
+			"template": "docs/FRONTEND.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:03087953d690c9902d35297720d1482262c1610e3050084f891db3be711571ef"
+		},
+		{
+			"path": ".sf/harness/AGENTS.md",
+			"template": ".sf/harness/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:685c41e601340086b8076263a71315c66554efdaeb074bc1b907eebf879174c6"
+		},
+		{
+			"path": ".sf/harness/specs/AGENTS.md",
+			"template": ".sf/harness/specs/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:0f4fbf4111704d05744e4a4e13a9bf3eada262f0da9517c2010f0b46f4bd3c45"
+		},
+		{
+			"path": ".sf/harness/specs/bootstrap.md",
+			"template": ".sf/harness/specs/bootstrap.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:b86ba7cf2cec39a7a9f9d94f885998cfe26eebfc5b76fdd8375ef125e927e0cf"
+		},
+		{
+			"path": ".sf/harness/evals/AGENTS.md",
+			"template": ".sf/harness/evals/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:6f88bf8a2bad95d8db5985c9b3317b9edd65592c12e98bb0bff1a24ec152d768"
+		},
+		{
+			"path": ".sf/harness/graders/AGENTS.md",
+			"template": ".sf/harness/graders/AGENTS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:2db17feae1acfe62d85aafbe32d016873c3036d4d76e9dd0db478375fae0794e"
+		},
+		{
+			"path": ".sf/PRINCIPLES.md",
+			"template": ".sf/PRINCIPLES.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:9d5c50cb3d602f66468a33a4324068fab8a022fab0fd6940c371a5986af2947e"
+		},
+		{
+			"path": ".sf/TASTE.md",
+			"template": ".sf/TASTE.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:ee9ea8ade6f8434e9425a989bab9735d2068395978d61940e18679b93b1f11b0"
+		},
+		{
+			"path": ".sf/ANTI-GOALS.md",
+			"template": ".sf/ANTI-GOALS.md",
+			"version": "2.75.3",
+			"appliedAt": "2026-05-06T19:53:05.336Z",
+			"stateAtApply": "pending",
+			"contentHash": "sha256:56b4f7d20d49e4558ac5aea5e967cf3475ef98d68dfa8985ba42b3e2b0d72e25"
+		}
+	]
 }
diff --git a/.sf.migrating/state-manifest.json b/.sf.migrating/state-manifest.json
index 631acdec7..4be4c8d32 100644
--- a/.sf.migrating/state-manifest.json
+++ b/.sf.migrating/state-manifest.json
@@ -1,649 +1,619 @@
 {
-  "version": 1,
-  "origin": "db-projection",
-  "source_db_path": "/home/mhugo/code/singularity-forge/.sf/sf.db",
-  "exported_at": "2026-05-06T21:01:36.517Z",
-  "milestones": [
-    {
-      "id": "M001-6377a4",
-      "title": "Consolidate Memory Systems into Unified node:sqlite Store",
-      "status": "active",
-      "depends_on": [],
-      "created_at": "2026-05-06T20:51:08.963Z",
-      "completed_at": null,
-      "vision": "Replace the three fragmented memory systems (KNOWLEDGE.md, memory-store.js/sf.db, and memory extension/sql.js) with a single unified memory store backed by node:sqlite. All memory ingestion, querying, and prompt injection flows through one canonical database table with semantic search, confidence ranking, and automatic maintenance.",
-      "success_criteria": [
-        "A single `memories` table in `sf.db` stores all project memory with categories, confidence, hit_count, and embeddings",
-        "Session transcript extraction pipeline writes directly to `sf.db` instead of separate `agent.db`",
-        "KNOWLEDGE.md is retired as a live system; existing entries are optionally migrated",
-        "Memory query returns ranked results via `getRelevantMemoriesRanked(query, limit)` using node:sqlite",
-        "System prompt injection uses the unified memory store, not multiple competing sources",
-        "All better-sqlite3 and sql.js dependencies for memory are removed"
-      ],
-      "key_risks": [
-        {
-          "risk": "Data loss during migration from existing memory stores",
-          "whyItMatters": "Users may have valuable accumulated memories in the old systems that would be lost if we don't migrate or preserve them"
-        },
-        {
-          "risk": "Performance regression from consolidating into a single database with WAL contention",
-          "whyItMatters": "The sf.db already has a single-writer invariant; adding memory extraction writes could create lock contention during session startup"
-        },
-        {
-          "risk": "Breaking the memory extension API contract for external consumers",
-          "whyItMatters": "The memory extension is a Pi extension with hooks and commands; changing its storage backend changes observable behavior"
-        }
-      ],
-      "proof_strategy": [
-        {
-          "riskOrUnknown": "Single database performance under write load",
-          "retireIn": "S02",
-          "whatWillBeProven": "Memory extraction during session startup completes without blocking dispatch, proven by timing telemetry and zero timeout errors"
-        },
-        {
-          "riskOrUnknown": "Migration path preserves existing memories",
-          "retireIn": "S03",
-          "whatWillBeProven": "Existing KNOWLEDGE.md entries and memory extension extractions are importable into the new schema with no data loss"
-        }
-      ],
-      "verification_contract": "Unit tests for memory CRUD, ranking, and embedding fallback. Integration test for full pipeline: session scan → extraction → storage → query → prompt injection.",
-      "verification_integration": "End-to-end test: run a mock session, trigger memory extraction, verify the memory appears in sf.db and is injectable into a prompt.",
-      "verification_operational": "Verify no memory-related background processes leak or consume excessive resources. Confirm WAL checkpoint behavior under memory write load.",
-      "verification_uat": "Manual verification: run `/memory view` command and confirm it reads from sf.db. Check that system prompt contains relevant memories from previous sessions.",
-      "definition_of_done": [
-        "All three legacy memory systems are removed or deprecated",
-        "Unified memory store passes all new tests",
-        "Migration script can import existing KNOWLEDGE.md and old agent.db data",
-        "Documentation updated to describe the single memory system",
-        "No references to better-sqlite3 or sql.js in memory-related code"
-      ],
-      "requirement_coverage": "Covers: unified memory storage, semantic search, session-based learning. Partially covers: cross-session context persistence. Leaves for later: multi-project memory sharing, cloud-synced memory.",
-      "boundary_map_markdown": "### S01 → S02\n\nProduces:\n- Unified `memories` table schema in `sf.db`\n- `MemoryStore` class with CRUD, query, and ranking API\n- `formatMemoriesForPrompt()` for system prompt injection\n\nConsumes:\n- Existing `sf-db.js` adapter and schema initialization\n\n### S02 → S03\n\nProduces:\n- Session transcript scanner and LLM extraction pipeline\n- Direct write path to `sf.db` memories table\n- `/memory` command reading from unified store\n\nConsumes:\n- `MemoryStore` API from S01\n- `completeSimple` LLM call from pi-ai\n\n### S03 → S04\n\nProduces:\n- KNOWLEDGE.md importer\n- Legacy agent.db migration script\n- Deprecation warnings for old memory paths\n\nConsumes:\n- `MemoryStore` API from S01\n- Existing KNOWLEDGE.md and agent.db files",
-      "vision_meeting": null
-    }
-  ],
-  "slices": [
-    {
-      "milestone_id": "M001-6377a4",
-      "id": "S01",
-      "title": "Unified Memory Schema and Core API",
-      "status": "pending",
-      "risk": "high",
-      "depends": [],
-      "demo": "A test creates a memory, queries it by relevance, and formats it for prompt injection — all via node:sqlite",
-      "created_at": "2026-05-06T20:52:17.318Z",
-      "completed_at": null,
-      "full_summary_md": "",
-      "full_uat_md": "",
-      "goal": "Create the canonical `memories` table in `sf.db` and a `MemoryStore` class that replaces memory-store.js with a node:sqlite implementation",
-      "success_criteria": "- `sf.db` has a `memories` table with: id, category, content, confidence, hit_count, source, session_id, source_unit_type, source_unit_id, created_at, updated_at, superseded_by\n- `MemoryStore` class exposes: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n- All methods gracefully degrade when DB is unavailable\n- Unit tests cover CRUD, ranking, source filtering, and embedding fallback",
-      "proof_level": "contract",
-      "integration_closure": "Produces the MemoryStore API that all downstream slices consume. No external wiring yet.",
-      "observability_impact": "No runtime signals added yet; tests verify table schema and query correctness.",
-      "adversarial_partner": "Missing partner review.",
-      "adversarial_combatant": "Missing combatant review.",
-      "adversarial_architect": "Missing architect review.",
-      "planning_meeting": {
-        "trigger": "Need to understand current memory-store.js implementation and sf-db.js schema before designing the unified table",
-        "pm": "The memory-store.js already has a working API with categories, confidence, hit_count, superseded_by, and embeddings. We should preserve this API surface but ensure it uses node:sqlite natively (not better-sqlite3 wrappers). The sf-db.js already uses node:sqlite DatabaseSync.",
-        "researcher": "Current memory-store.js uses sf-db.js which uses node:sqlite DatabaseSync. The schema already has memories table with: seq (autoincrement), id, category, content, confidence, source_unit_type, source_unit_id, created_at, updated_at, superseded_by, hit_count. Embeddings are in a separate memory_embeddings table. Relations in memory_relations. Processed units tracked in memory_processed_units.",
-        "partner": "The existing memory-store.js is already well-designed. S01 should: (1) verify the schema is correct in sf-db.js, (2) create a clean MemoryStore class that wraps the existing functions, (3) add unit tests, (4) ensure graceful degradation. We don't need to redesign the schema — it's already good.",
-        "combatant": "The current memory-store.js exports individual functions, not a class. Creating a class is an unnecessary abstraction if the functions already work. Also, the existing schema might be missing fields needed for the session extraction pipeline (like thread_id, session_id). We should check if the memory extension's schema has fields we need to add.",
-        "architect": "Agree with combatant on checking schema compatibility. The memory extension's pipeline stores: thread_id, file_path, file_size, file_mtime, cwd, status, extraction_json. These are pipeline metadata, not memory content. The actual memories are extracted JSON with category/content/confidence. So the existing memories table schema should work. However, we might want to add a `source` column to distinguish manual capture_thought vs auto-extracted vs KNOWLEDGE.md migrated entries.",
-        "moderator": "Decision: Keep the existing memories table schema (it's already correct). Add a `source` column to distinguish ingestion paths. Create a MemoryStore class that wraps the existing function-based API for cleaner consumption by the pipeline. Write unit tests. Do NOT redesign the schema from scratch.",
-        "recommendedRoute": "planning",
-        "confidenceSummary": "High confidence. The existing memory-store.js and sf-db.js are already well-designed. We just need to add a source column, create a class wrapper, and add tests."
-      },
-      "sequence": 1,
-      "replan_triggered_at": null
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "id": "S02",
-      "title": "Session Transcript Extraction Pipeline",
-      "status": "pending",
-      "risk": "medium",
-      "depends": [
-        "S01"
-      ],
-      "demo": "After running a session, the pipeline extracts memories and stores them in sf.db; `/memory view` shows the extracted knowledge",
-      "created_at": "2026-05-06T20:52:17.319Z",
-      "completed_at": null,
-      "full_summary_md": "",
-      "full_uat_md": "",
-      "goal": "Port the two-phase extraction pipeline from the memory extension to write directly into the unified sf.db store instead of separate agent.db",
-      "success_criteria": "- Session .jsonl files are scanned and filtered for the current cwd\n- Phase 1: LLM extracts durable knowledge from transcripts, writes to sf.db memories table with source='extracted'\n- Phase 2: Consolidation deduplicates and ranks memories\n- /memory command reads from unified store\n- Pipeline runs fire-and-forget on session start without blocking dispatch\n- Rate limiting prevents store bloat (max 10 memories per session)",
-      "proof_level": "integration",
-      "integration_closure": "Wires the extraction pipeline into the session lifecycle and replaces the old memory extension storage path.",
-      "observability_impact": "Pipeline logs extraction stats (processed, errors) to the console. Memory store tracks hit_count for reinforcement.",
-      "adversarial_partner": "Missing partner review.",
-      "adversarial_combatant": "Missing combatant review.",
-      "adversarial_architect": "Missing architect review.",
-      "planning_meeting": {
-        "trigger": "Need to port the session transcript extraction pipeline from the memory extension (which uses sql.js/agent.db) to use the unified MemoryStore (sf.db with node:sqlite)",
-        "pm": "The memory extension's pipeline is well-designed but writes to a separate database. We should port it to use the unified MemoryStore while preserving the two-phase architecture (extract → consolidate). The /memory command should read from sf.db.",
-        "researcher": "The memory extension pipeline has: scanSessionFiles, filterSessionContent, runPhase1 (LLM extraction), runPhase2 (consolidation). It uses sql.js MemoryStorage class. We need to replace MemoryStorage with MemoryStore and write to sf.db instead of agent.db.",
-        "partner": "The pipeline should be preserved as-is but write to the unified store. Phase 1 extracts memories from session .jsonl files via LLM. Phase 2 consolidates and deduplicates. Both should write to sf.db memories table with source='extracted'.",
-        "combatant": "The pipeline might create too many memories and overwhelm the store. We need rate limiting and deduplication. Also, running LLM extraction on every session startup could be expensive and slow.",
-        "architect": "Agree on rate limiting. The pipeline should: (1) use the existing session scanner, (2) call LLM for extraction, (3) write to MemoryStore with source='extracted', (4) run consolidation as a separate phase. We should add a processed_units check to avoid re-processing the same sessions.",
-        "moderator": "Decision: Port the pipeline to use MemoryStore. Add session_id tracking to avoid re-processing. Keep the two-phase architecture. Add rate limiting (max memories per session). Run extraction asynchronously without blocking dispatch.",
-        "recommendedRoute": "planning",
-        "confidenceSummary": "Medium-high confidence. The pipeline code is well-understood. The main risk is performance and store bloat, which we mitigate with rate limiting and deduplication."
-      },
-      "sequence": 2,
-      "replan_triggered_at": null
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "id": "S03",
-      "title": "Migration and Legacy Deprecation",
-      "status": "pending",
-      "risk": "low",
-      "depends": [
-        "S02"
-      ],
-      "demo": "Running a migration command imports existing KNOWLEDGE.md and agent.db memories into sf.db; old files are left as read-only backups",
-      "created_at": "2026-05-06T20:52:17.319Z",
-      "completed_at": null,
-      "full_summary_md": "",
-      "full_uat_md": "",
-      "goal": "Provide a one-time migration path from KNOWLEDGE.md and the old agent.db into the unified store, then deprecate the legacy systems",
-      "success_criteria": "- Migration script can import KNOWLEDGE.md entries as memories with source='migrated'\n- Migration script can import agent.db stage1_outputs as memories with source='extracted'\n- Dry-run mode reports counts without modifying database\n- sf.db is automatically backed up before migration\n- knowledge-compounding.js and knowledge-injector.js are removed or no-op'd\n- Memory extension uses MemoryStore instead of sql.js\n- CHANGELOG documents the consolidation and deprecation",
-      "proof_level": "contract",
-      "integration_closure": "Cleans up legacy code paths and provides a safe migration for existing users.",
-      "observability_impact": "Migration script reports counts of imported records per source. Deprecation warnings logged for old API usage.",
-      "adversarial_partner": "Missing partner review.",
-      "adversarial_combatant": "Missing combatant review.",
-      "adversarial_architect": "Missing architect review.",
-      "planning_meeting": {
-        "trigger": "Need to migrate existing KNOWLEDGE.md entries and old agent.db memories into the unified sf.db store before deprecating legacy systems",
-        "pm": "Users may have valuable accumulated knowledge in KNOWLEDGE.md and extracted memories in agent.db. We need a safe migration path that preserves this data. The migration should be idempotent and leave old files as read-only backups.",
-        "researcher": "KNOWLEDGE.md format: plain markdown with sections (Rules, Patterns, Lessons Learned). Each entry is a table row. Agent.db format: sql.js database with tables: threads, stage1_outputs, jobs. stage1_outputs contains extraction_json (JSON array of memories).",
-        "partner": "Migration should: (1) parse KNOWLEDGE.md entries and insert as memories with source='migrated', (2) read agent.db stage1_outputs and insert as memories with source='extracted', (3) deduplicate against existing memories, (4) leave old files untouched as backups.",
-        "combatant": "Migration scripts are often buggy and can corrupt data. We should validate the migration with dry-run mode and backup the sf.db before running. Also, what if the old agent.db is corrupted or uses an older schema?",
-        "architect": "Agree on dry-run and backup. The migration script should: (1) create a backup of sf.db, (2) parse KNOWLEDGE.md with regex/table parsing, (3) parse agent.db extractions with JSON.parse, (4) check for duplicates before inserting, (5) report counts. For corrupted agent.db, skip with warning.",
-        "moderator": "Decision: Create a migration script with dry-run mode and automatic backup. Parse KNOWLEDGE.md entries as individual memories. Parse agent.db stage1_outputs as memory arrays. Deduplicate by content hash. Report migration statistics. Leave old files as read-only backups.",
-        "recommendedRoute": "planning",
-        "confidenceSummary": "Medium confidence. Migration scripts are inherently risky, but dry-run and backup mitigate this. The parsing logic is straightforward."
-      },
-      "sequence": 3,
-      "replan_triggered_at": null
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "id": "S04",
-      "title": "System Prompt Integration and Cleanup",
-      "status": "pending",
-      "risk": "medium",
-      "depends": [
-        "S03"
-      ],
-      "demo": "Starting a new session injects relevant ranked memories from sf.db into the system prompt; no references to old memory systems remain in the codebase",
-      "created_at": "2026-05-06T20:52:17.319Z",
-      "completed_at": null,
-      "full_summary_md": "",
-      "full_uat_md": "",
-      "goal": "Replace all system prompt injection paths to use the unified memory store, remove dead code, and verify end-to-end",
-      "success_criteria": "Not provided.",
-      "proof_level": "Not provided.",
-      "integration_closure": "Not provided.",
-      "observability_impact": "Not provided.",
-      "adversarial_partner": "Missing partner review.",
-      "adversarial_combatant": "Missing combatant review.",
-      "adversarial_architect": "Missing architect review.",
-      "planning_meeting": {
-        "trigger": "Need to integrate unified memory store into system prompt injection and remove all legacy memory code paths",
-        "pm": "The system prompt currently loads KNOWLEDGE.md and memory-store.js separately. We need to replace this with a single MemoryStore.getRelevantMemoriesRanked() call that injects ranked memories into the prompt. This simplifies the context assembly and ensures all memory flows through one path.",
-        "researcher": "system-context.js currently loads: knowledgeBlock (from KNOWLEDGE.md files) and memoryBlock (from memory-store.js getActiveMemoriesRanked). We need to replace both with a single call to MemoryStore.getRelevantMemoriesRanked(query, limit) where query is derived from the current task context.",
-        "partner": "The integration should: (1) replace knowledgeBlock + memoryBlock with a single memory injection, (2) use task context keywords for relevance scoring, (3) format memories with category headers, (4) respect token budget, (5) remove all imports of old memory systems.",
-        "combatant": "Removing KNOWLEDGE.md injection might lose important human-curated rules that aren't in the memory store yet. We should ensure migration is complete before removing the old path, or keep a fallback for unmigrated knowledge.",
-        "architect": "Agree on fallback. The integration should: (1) try MemoryStore first, (2) if empty, fall back to KNOWLEDGE.md (read-only, no new writes), (3) inject formatted memories into system prompt via before_agent_start hook, (4) track memory usage for feedback loop.",
-        "moderator": "Decision: Replace system prompt injection with MemoryStore.getRelevantMemoriesRanked(). Add a temporary fallback to KNOWLEDGE.md if MemoryStore returns empty (until migration is complete). Remove all old memory system imports. Add memory usage tracking. Run end-to-end test.",
-        "recommendedRoute": "planning",
-        "confidenceSummary": "High confidence. The integration is straightforward — replace two injection paths with one. The fallback ensures no regression during migration."
-      },
-      "sequence": 4,
-      "replan_triggered_at": null
-    }
-  ],
-  "tasks": [
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S01",
-      "id": "T01",
-      "title": "Audit and extend sf.db memories schema",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Read sf-db.js schema initialization to verify memories table exists with correct columns\n2. Add `source` column (TEXT, default 'manual') to distinguish: 'manual', 'extracted', 'migrated'\n3. Add `session_id` column (TEXT, nullable) to link extracted memories to source sessions\n4. Bump SCHEMA_VERSION in sf-db.js\n5. Write migration SQL that adds columns if they don't exist (for existing databases)\n6. Verify: run tests that create and query memories with new columns",
-      "estimate": "1h",
-      "files": [
-        "src/resources/extensions/sf/sf-db.js",
-        "src/resources/extensions/sf/memory-store.js"
-      ],
-      "verify": "npm test -- --grep \"memory schema\"",
-      "inputs": [
-        "src/resources/extensions/sf/sf-db.js"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/sf-db.js (modified)"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S01",
-      "id": "T02",
-      "title": "Create MemoryStore class wrapper",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Create src/resources/extensions/sf/MemoryStore.ts (or .js) that wraps existing memory-store.js functions\n2. Expose methods: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n3. Add source parameter to createMemory (default 'manual')\n4. Ensure all methods delegate to existing sf-db.js wrappers\n5. Export both the class and individual functions for backward compatibility\n6. Verify: unit tests pass for class and function APIs",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/MemoryStore.ts",
-        "src/resources/extensions/sf/memory-store.js"
-      ],
-      "verify": "npm test -- --grep \"MemoryStore\"",
-      "inputs": [
-        "src/resources/extensions/sf/memory-store.js",
-        "src/resources/extensions/sf/sf-db.js"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S01",
-      "id": "T03",
-      "title": "Write unit tests for MemoryStore CRUD and ranking",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Create src/resources/extensions/sf/tests/memory-store.test.ts\n2. Test CRUD: createMemory, getActiveMemories, updateMemory, reinforceMemory, supersedeMemory\n3. Test ranking: getActiveMemoriesRanked returns correct order by confidence * hit_count\n4. Test graceful degradation: returns empty arrays when DB unavailable\n5. Test source filtering: memories created with different sources are queryable\n6. Verify: all tests pass",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/tests/memory-store.test.ts"
-      ],
-      "verify": "npm test -- src/resources/extensions/sf/tests/memory-store.test.ts",
-      "inputs": [
-        "src/resources/extensions/sf/MemoryStore.ts",
-        "src/resources/extensions/sf/memory-store.js"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/tests/memory-store.test.ts"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S02",
-      "id": "T01",
-      "title": "Port session scanner to use MemoryStore",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Copy scanSessionFiles and filterSessionContent from memory extension\n2. Adapt to read from ~/.sf/agent/sessions/ directory structure\n3. Add session fingerprinting (hash of content) to avoid re-processing\n4. Check memory_processed_units table before processing\n5. Verify: scanner finds session files and filters correctly",
-      "estimate": "1h",
-      "files": [
-        "src/resources/extensions/sf/memory-pipeline.js"
-      ],
-      "verify": "npm test -- --grep \"session scanner\"",
-      "inputs": [
-        "packages/pi-coding-agent/src/resources/extensions/memory/pipeline.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/memory-pipeline.js"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S02",
-      "id": "T02",
-      "title": "Implement Phase 1: LLM extraction to MemoryStore",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Port runPhase1 from memory extension\n2. Use completeSimple from pi-ai for LLM calls\n3. Extract memories as JSON array with category/content/confidence\n4. Write each extracted memory to MemoryStore with source='extracted' and session_id\n5. Add rate limiting: max 10 memories per session, min confidence 0.6\n6. Mark session as processed in memory_processed_units\n7. Verify: extraction creates memories in sf.db",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/memory-pipeline.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "verify": "npm test -- --grep \"memory extraction\"",
-      "inputs": [
-        "src/resources/extensions/sf/memory-pipeline.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/memory-pipeline.js (phase 1)"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S02",
-      "id": "T03",
-      "title": "Implement Phase 2: Consolidation and /memory command",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Port runPhase2 consolidation logic\n2. Query MemoryStore for source='extracted' memories\n3. Group by category, deduplicate similar content (simple string similarity)\n4. Update consolidated memories with higher confidence\n5. Supersede duplicate memories\n6. Implement /memory command that reads from MemoryStore\n7. Verify: consolidation runs and /memory shows extracted memories",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/memory-pipeline.js",
-        "src/resources/extensions/sf/memory-commands.js"
-      ],
-      "verify": "npm test -- --grep \"memory consolidation\"",
-      "inputs": [
-        "src/resources/extensions/sf/memory-pipeline.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/memory-commands.js"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S03",
-      "id": "T01",
-      "title": "Create KNOWLEDGE.md parser and migrator",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Create src/resources/extensions/sf/migrate-knowledge.js\n2. Parse KNOWLEDGE.md sections (Rules, Patterns, Lessons Learned)\n3. Extract table rows as memory entries\n4. Map to memory categories: rule→convention, pattern→pattern, lesson→gotcha\n5. Insert into MemoryStore with source='migrated'\n6. Deduplicate by content hash against existing memories\n7. Add dry-run mode (report without inserting)\n8. Verify: dry-run reports correct counts",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/migrate-knowledge.js",
-        ".sf/KNOWLEDGE.md",
-        "KNOWLEDGE.md"
-      ],
-      "verify": "node src/resources/extensions/sf/migrate-knowledge.js --dry-run",
-      "inputs": [
-        ".sf/KNOWLEDGE.md",
-        "KNOWLEDGE.md",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/migrate-knowledge.js"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S03",
-      "id": "T02",
-      "title": "Create agent.db migrator",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Create src/resources/extensions/sf/migrate-agent-db.js\n2. Read old agent.db using sql.js\n3. Extract stage1_outputs.extraction_json fields\n4. Parse JSON arrays of memories\n5. Insert into MemoryStore with source='extracted'\n6. Deduplicate by content hash\n7. Handle corrupted databases gracefully (skip with warning)\n8. Add dry-run mode\n9. Verify: dry-run reports correct counts",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/migrate-agent-db.js",
-        "~/.sf/agent/agent.db"
-      ],
-      "verify": "node src/resources/extensions/sf/migrate-agent-db.js --dry-run",
-      "inputs": [
-        "packages/pi-coding-agent/src/resources/extensions/memory/storage.ts",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/migrate-agent-db.js"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S03",
-      "id": "T03",
-      "title": "Deprecate legacy memory systems",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Add deprecation warnings to memory-store.js exports (log warning on first use)\n2. Remove knowledge-compounding.js and knowledge-injector.js or make them no-op\n3. Update memory extension to use MemoryStore instead of sql.js\n4. Remove sql.js dependency from memory extension\n5. Update CHANGELOG.md with migration instructions\n6. Verify: no imports of old memory systems in active code paths",
-      "estimate": "1h",
-      "files": [
-        "src/resources/extensions/sf/memory-store.js",
-        "src/resources/extensions/sf/knowledge-compounding.js",
-        "src/resources/extensions/sf/knowledge-injector.js",
-        "packages/pi-coding-agent/src/resources/extensions/memory/index.ts",
-        "packages/pi-coding-agent/src/resources/extensions/memory/storage.ts",
-        "CHANGELOG.md"
-      ],
-      "verify": "grep -r \"knowledge-compounding\\|knowledge-injector\\|sql.js\" src/ packages/ --include=\"*.ts\" --include=\"*.js\" | grep -v \"test\\|spec\\|deprecated\"",
-      "inputs": [
-        "src/resources/extensions/sf/migrate-knowledge.js",
-        "src/resources/extensions/sf/migrate-agent-db.js"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/memory-store.js (deprecated)",
-        "CHANGELOG.md (updated)"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S04",
-      "id": "T01",
-      "title": "Integrate MemoryStore into system prompt injection",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Read system-context.js to understand current memory injection\n2. Replace knowledgeBlock + memoryBlock with MemoryStore.getRelevantMemoriesRanked()\n3. Derive query from task context (domain, keywords, technology)\n4. Format memories with category headers\n5. Respect token budget (truncate if needed)\n6. Add fallback to KNOWLEDGE.md if MemoryStore empty\n7. Verify: system prompt contains formatted memories",
-      "estimate": "1.5h",
-      "files": [
-        "src/resources/extensions/sf/bootstrap/system-context.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "verify": "npm test -- --grep \"system context memory\"",
-      "inputs": [
-        "src/resources/extensions/sf/bootstrap/system-context.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/bootstrap/system-context.js (modified)"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S04",
-      "id": "T02",
-      "title": "Remove legacy memory system imports",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Remove all imports of memory-store.js from active code\n2. Remove all imports of knowledge-injector.js\n3. Remove all imports of knowledge-compounding.js\n4. Update package.json to remove better-sqlite3 and sql.js from memory-related dependencies\n5. Verify: grep returns no active references to old systems",
-      "estimate": "1h",
-      "files": [
-        "src/resources/extensions/sf/**/*.js",
-        "src/resources/extensions/sf/**/*.ts",
-        "packages/pi-coding-agent/src/resources/extensions/memory/**/*.ts"
-      ],
-      "verify": "grep -r \"knowledge-injector\\|knowledge-compounding\\|memory-store\" src/ packages/ --include=\"*.ts\" --include=\"*.js\" | grep -v \"test\\|spec\\|deprecated\"",
-      "inputs": [
-        "src/resources/extensions/sf/bootstrap/system-context.js"
-      ],
-      "expected_output": [
-        "Clean codebase with no legacy memory imports"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    },
-    {
-      "milestone_id": "M001-6377a4",
-      "slice_id": "S04",
-      "id": "T03",
-      "title": "Write end-to-end integration tests",
-      "status": "pending",
-      "one_liner": "",
-      "narrative": "",
-      "verification_result": "",
-      "duration": "",
-      "completed_at": null,
-      "blocker_discovered": false,
-      "deviations": "",
-      "known_issues": "",
-      "key_files": [],
-      "key_decisions": [],
-      "full_summary_md": "",
-      "description": "1. Create end-to-end test: mock session with memory-worthy content\n2. Trigger memory extraction pipeline\n3. Start new session with matching context\n4. Verify system prompt contains the extracted memory\n5. Test graceful degradation: empty store returns empty string\n6. Test token budget: large memory set is truncated\n7. Verify: all tests pass",
-      "estimate": "2h",
-      "files": [
-        "src/resources/extensions/sf/tests/memory-e2e.test.ts"
-      ],
-      "verify": "npm test -- src/resources/extensions/sf/tests/memory-e2e.test.ts",
-      "inputs": [
-        "src/resources/extensions/sf/bootstrap/system-context.js",
-        "src/resources/extensions/sf/memory-pipeline.js",
-        "src/resources/extensions/sf/MemoryStore.ts"
-      ],
-      "expected_output": [
-        "src/resources/extensions/sf/tests/memory-e2e.test.ts"
-      ],
-      "observability_impact": "",
-      "full_plan_md": "",
-      "sequence": 0
-    }
-  ],
-  "decisions": [
-    {
-      "seq": 1,
-      "id": "D001",
-      "when_context": "",
-      "scope": "environment",
-      "decision": "How to handle missing libsecret dependency for keytar native addon on Linux.",
-      "choice": "Add libsecret-1-0 / libsecret-1-dev to all Linux environment definitions (Docker, Nix, CI).",
-      "rationale": "@github/keytar (used by @google/gemini-cli-core) requires libsecret on Linux. Without it, keychain initialization fails with a cryptic \"Failed to load keytar native addon\" error. Adding it to the core environment definitions ensures it is present in all official SF environments.",
-      "revisable": "Yes",
-      "made_by": "agent",
-      "superseded_by": null
-    }
-  ],
-  "verification_evidence": []
-}
\ No newline at end of file
+	"version": 1,
+	"origin": "db-projection",
+	"source_db_path": "/home/mhugo/code/singularity-forge/.sf/sf.db",
+	"exported_at": "2026-05-06T21:01:36.517Z",
+	"milestones": [
+		{
+			"id": "M001-6377a4",
+			"title": "Consolidate Memory Systems into Unified node:sqlite Store",
+			"status": "active",
+			"depends_on": [],
+			"created_at": "2026-05-06T20:51:08.963Z",
+			"completed_at": null,
+			"vision": "Replace the three fragmented memory systems (KNOWLEDGE.md, memory-store.js/sf.db, and memory extension/sql.js) with a single unified memory store backed by node:sqlite. All memory ingestion, querying, and prompt injection flows through one canonical database table with semantic search, confidence ranking, and automatic maintenance.",
+			"success_criteria": [
+				"A single `memories` table in `sf.db` stores all project memory with categories, confidence, hit_count, and embeddings",
+				"Session transcript extraction pipeline writes directly to `sf.db` instead of separate `agent.db`",
+				"KNOWLEDGE.md is retired as a live system; existing entries are optionally migrated",
+				"Memory query returns ranked results via `getRelevantMemoriesRanked(query, limit)` using node:sqlite",
+				"System prompt injection uses the unified memory store, not multiple competing sources",
+				"All better-sqlite3 and sql.js dependencies for memory are removed"
+			],
+			"key_risks": [
+				{
+					"risk": "Data loss during migration from existing memory stores",
+					"whyItMatters": "Users may have valuable accumulated memories in the old systems that would be lost if we don't migrate or preserve them"
+				},
+				{
+					"risk": "Performance regression from consolidating into a single database with WAL contention",
+					"whyItMatters": "The sf.db already has a single-writer invariant; adding memory extraction writes could create lock contention during session startup"
+				},
+				{
+					"risk": "Breaking the memory extension API contract for external consumers",
+					"whyItMatters": "The memory extension is a Pi extension with hooks and commands; changing its storage backend changes observable behavior"
+				}
+			],
+			"proof_strategy": [
+				{
+					"riskOrUnknown": "Single database performance under write load",
+					"retireIn": "S02",
+					"whatWillBeProven": "Memory extraction during session startup completes without blocking dispatch, proven by timing telemetry and zero timeout errors"
+				},
+				{
+					"riskOrUnknown": "Migration path preserves existing memories",
+					"retireIn": "S03",
+					"whatWillBeProven": "Existing KNOWLEDGE.md entries and memory extension extractions are importable into the new schema with no data loss"
+				}
+			],
+			"verification_contract": "Unit tests for memory CRUD, ranking, and embedding fallback. Integration test for full pipeline: session scan → extraction → storage → query → prompt injection.",
+			"verification_integration": "End-to-end test: run a mock session, trigger memory extraction, verify the memory appears in sf.db and is injectable into a prompt.",
+			"verification_operational": "Verify no memory-related background processes leak or consume excessive resources. Confirm WAL checkpoint behavior under memory write load.",
+			"verification_uat": "Manual verification: run `/memory view` command and confirm it reads from sf.db. Check that system prompt contains relevant memories from previous sessions.",
+			"definition_of_done": [
+				"All three legacy memory systems are removed or deprecated",
+				"Unified memory store passes all new tests",
+				"Migration script can import existing KNOWLEDGE.md and old agent.db data",
+				"Documentation updated to describe the single memory system",
+				"No references to better-sqlite3 or sql.js in memory-related code"
+			],
+			"requirement_coverage": "Covers: unified memory storage, semantic search, session-based learning. Partially covers: cross-session context persistence. Leaves for later: multi-project memory sharing, cloud-synced memory.",
+			"boundary_map_markdown": "### S01 → S02\n\nProduces:\n- Unified `memories` table schema in `sf.db`\n- `MemoryStore` class with CRUD, query, and ranking API\n- `formatMemoriesForPrompt()` for system prompt injection\n\nConsumes:\n- Existing `sf-db.js` adapter and schema initialization\n\n### S02 → S03\n\nProduces:\n- Session transcript scanner and LLM extraction pipeline\n- Direct write path to `sf.db` memories table\n- `/memory` command reading from unified store\n\nConsumes:\n- `MemoryStore` API from S01\n- `completeSimple` LLM call from pi-ai\n\n### S03 → S04\n\nProduces:\n- KNOWLEDGE.md importer\n- Legacy agent.db migration script\n- Deprecation warnings for old memory paths\n\nConsumes:\n- `MemoryStore` API from S01\n- Existing KNOWLEDGE.md and agent.db files",
+			"vision_meeting": null
+		}
+	],
+	"slices": [
+		{
+			"milestone_id": "M001-6377a4",
+			"id": "S01",
+			"title": "Unified Memory Schema and Core API",
+			"status": "pending",
+			"risk": "high",
+			"depends": [],
+			"demo": "A test creates a memory, queries it by relevance, and formats it for prompt injection — all via node:sqlite",
+			"created_at": "2026-05-06T20:52:17.318Z",
+			"completed_at": null,
+			"full_summary_md": "",
+			"full_uat_md": "",
+			"goal": "Create the canonical `memories` table in `sf.db` and a `MemoryStore` class that replaces memory-store.js with a node:sqlite implementation",
+			"success_criteria": "- `sf.db` has a `memories` table with: id, category, content, confidence, hit_count, source, session_id, source_unit_type, source_unit_id, created_at, updated_at, superseded_by\n- `MemoryStore` class exposes: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n- All methods gracefully degrade when DB is unavailable\n- Unit tests cover CRUD, ranking, source filtering, and embedding fallback",
+			"proof_level": "contract",
+			"integration_closure": "Produces the MemoryStore API that all downstream slices consume. No external wiring yet.",
+			"observability_impact": "No runtime signals added yet; tests verify table schema and query correctness.",
+			"adversarial_partner": "Missing partner review.",
+			"adversarial_combatant": "Missing combatant review.",
+			"adversarial_architect": "Missing architect review.",
+			"planning_meeting": {
+				"trigger": "Need to understand current memory-store.js implementation and sf-db.js schema before designing the unified table",
+				"pm": "The memory-store.js already has a working API with categories, confidence, hit_count, superseded_by, and embeddings. We should preserve this API surface but ensure it uses node:sqlite natively (not better-sqlite3 wrappers). The sf-db.js already uses node:sqlite DatabaseSync.",
+				"researcher": "Current memory-store.js uses sf-db.js which uses node:sqlite DatabaseSync. The schema already has memories table with: seq (autoincrement), id, category, content, confidence, source_unit_type, source_unit_id, created_at, updated_at, superseded_by, hit_count. Embeddings are in a separate memory_embeddings table. Relations in memory_relations. Processed units tracked in memory_processed_units.",
+				"partner": "The existing memory-store.js is already well-designed. S01 should: (1) verify the schema is correct in sf-db.js, (2) create a clean MemoryStore class that wraps the existing functions, (3) add unit tests, (4) ensure graceful degradation. We don't need to redesign the schema — it's already good.",
+				"combatant": "The current memory-store.js exports individual functions, not a class. Creating a class is an unnecessary abstraction if the functions already work. Also, the existing schema might be missing fields needed for the session extraction pipeline (like thread_id, session_id). We should check if the memory extension's schema has fields we need to add.",
+				"architect": "Agree with combatant on checking schema compatibility. The memory extension's pipeline stores: thread_id, file_path, file_size, file_mtime, cwd, status, extraction_json. These are pipeline metadata, not memory content. The actual memories are extracted JSON with category/content/confidence. So the existing memories table schema should work. However, we might want to add a `source` column to distinguish manual capture_thought vs auto-extracted vs KNOWLEDGE.md migrated entries.",
+				"moderator": "Decision: Keep the existing memories table schema (it's already correct). Add a `source` column to distinguish ingestion paths. Create a MemoryStore class that wraps the existing function-based API for cleaner consumption by the pipeline. Write unit tests. Do NOT redesign the schema from scratch.",
+				"recommendedRoute": "planning",
+				"confidenceSummary": "High confidence. The existing memory-store.js and sf-db.js are already well-designed. We just need to add a source column, create a class wrapper, and add tests."
+			},
+			"sequence": 1,
+			"replan_triggered_at": null
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"id": "S02",
+			"title": "Session Transcript Extraction Pipeline",
+			"status": "pending",
+			"risk": "medium",
+			"depends": ["S01"],
+			"demo": "After running a session, the pipeline extracts memories and stores them in sf.db; `/memory view` shows the extracted knowledge",
+			"created_at": "2026-05-06T20:52:17.319Z",
+			"completed_at": null,
+			"full_summary_md": "",
+			"full_uat_md": "",
+			"goal": "Port the two-phase extraction pipeline from the memory extension to write directly into the unified sf.db store instead of separate agent.db",
+			"success_criteria": "- Session .jsonl files are scanned and filtered for the current cwd\n- Phase 1: LLM extracts durable knowledge from transcripts, writes to sf.db memories table with source='extracted'\n- Phase 2: Consolidation deduplicates and ranks memories\n- /memory command reads from unified store\n- Pipeline runs fire-and-forget on session start without blocking dispatch\n- Rate limiting prevents store bloat (max 10 memories per session)",
+			"proof_level": "integration",
+			"integration_closure": "Wires the extraction pipeline into the session lifecycle and replaces the old memory extension storage path.",
+			"observability_impact": "Pipeline logs extraction stats (processed, errors) to the console. Memory store tracks hit_count for reinforcement.",
+			"adversarial_partner": "Missing partner review.",
+			"adversarial_combatant": "Missing combatant review.",
+			"adversarial_architect": "Missing architect review.",
+			"planning_meeting": {
+				"trigger": "Need to port the session transcript extraction pipeline from the memory extension (which uses sql.js/agent.db) to use the unified MemoryStore (sf.db with node:sqlite)",
+				"pm": "The memory extension's pipeline is well-designed but writes to a separate database. We should port it to use the unified MemoryStore while preserving the two-phase architecture (extract → consolidate). The /memory command should read from sf.db.",
+				"researcher": "The memory extension pipeline has: scanSessionFiles, filterSessionContent, runPhase1 (LLM extraction), runPhase2 (consolidation). It uses sql.js MemoryStorage class. We need to replace MemoryStorage with MemoryStore and write to sf.db instead of agent.db.",
+				"partner": "The pipeline should be preserved as-is but write to the unified store. Phase 1 extracts memories from session .jsonl files via LLM. Phase 2 consolidates and deduplicates. Both should write to sf.db memories table with source='extracted'.",
+				"combatant": "The pipeline might create too many memories and overwhelm the store. We need rate limiting and deduplication. Also, running LLM extraction on every session startup could be expensive and slow.",
+				"architect": "Agree on rate limiting. The pipeline should: (1) use the existing session scanner, (2) call LLM for extraction, (3) write to MemoryStore with source='extracted', (4) run consolidation as a separate phase. We should add a processed_units check to avoid re-processing the same sessions.",
+				"moderator": "Decision: Port the pipeline to use MemoryStore. Add session_id tracking to avoid re-processing. Keep the two-phase architecture. Add rate limiting (max memories per session). Run extraction asynchronously without blocking dispatch.",
+				"recommendedRoute": "planning",
+				"confidenceSummary": "Medium-high confidence. The pipeline code is well-understood. The main risk is performance and store bloat, which we mitigate with rate limiting and deduplication."
+			},
+			"sequence": 2,
+			"replan_triggered_at": null
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"id": "S03",
+			"title": "Migration and Legacy Deprecation",
+			"status": "pending",
+			"risk": "low",
+			"depends": ["S02"],
+			"demo": "Running a migration command imports existing KNOWLEDGE.md and agent.db memories into sf.db; old files are left as read-only backups",
+			"created_at": "2026-05-06T20:52:17.319Z",
+			"completed_at": null,
+			"full_summary_md": "",
+			"full_uat_md": "",
+			"goal": "Provide a one-time migration path from KNOWLEDGE.md and the old agent.db into the unified store, then deprecate the legacy systems",
+			"success_criteria": "- Migration script can import KNOWLEDGE.md entries as memories with source='migrated'\n- Migration script can import agent.db stage1_outputs as memories with source='extracted'\n- Dry-run mode reports counts without modifying database\n- sf.db is automatically backed up before migration\n- knowledge-compounding.js and knowledge-injector.js are removed or no-op'd\n- Memory extension uses MemoryStore instead of sql.js\n- CHANGELOG documents the consolidation and deprecation",
+			"proof_level": "contract",
+			"integration_closure": "Cleans up legacy code paths and provides a safe migration for existing users.",
+			"observability_impact": "Migration script reports counts of imported records per source. Deprecation warnings logged for old API usage.",
+			"adversarial_partner": "Missing partner review.",
+			"adversarial_combatant": "Missing combatant review.",
+			"adversarial_architect": "Missing architect review.",
+			"planning_meeting": {
+				"trigger": "Need to migrate existing KNOWLEDGE.md entries and old agent.db memories into the unified sf.db store before deprecating legacy systems",
+				"pm": "Users may have valuable accumulated knowledge in KNOWLEDGE.md and extracted memories in agent.db. We need a safe migration path that preserves this data. The migration should be idempotent and leave old files as read-only backups.",
+				"researcher": "KNOWLEDGE.md format: plain markdown with sections (Rules, Patterns, Lessons Learned). Each entry is a table row. Agent.db format: sql.js database with tables: threads, stage1_outputs, jobs. stage1_outputs contains extraction_json (JSON array of memories).",
+				"partner": "Migration should: (1) parse KNOWLEDGE.md entries and insert as memories with source='migrated', (2) read agent.db stage1_outputs and insert as memories with source='extracted', (3) deduplicate against existing memories, (4) leave old files untouched as backups.",
+				"combatant": "Migration scripts are often buggy and can corrupt data. We should validate the migration with dry-run mode and backup the sf.db before running. Also, what if the old agent.db is corrupted or uses an older schema?",
+				"architect": "Agree on dry-run and backup. The migration script should: (1) create a backup of sf.db, (2) parse KNOWLEDGE.md with regex/table parsing, (3) parse agent.db extractions with JSON.parse, (4) check for duplicates before inserting, (5) report counts. For corrupted agent.db, skip with warning.",
+				"moderator": "Decision: Create a migration script with dry-run mode and automatic backup. Parse KNOWLEDGE.md entries as individual memories. Parse agent.db stage1_outputs as memory arrays. Deduplicate by content hash. Report migration statistics. Leave old files as read-only backups.",
+				"recommendedRoute": "planning",
+				"confidenceSummary": "Medium confidence. Migration scripts are inherently risky, but dry-run and backup mitigate this. The parsing logic is straightforward."
+			},
+			"sequence": 3,
+			"replan_triggered_at": null
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"id": "S04",
+			"title": "System Prompt Integration and Cleanup",
+			"status": "pending",
+			"risk": "medium",
+			"depends": ["S03"],
+			"demo": "Starting a new session injects relevant ranked memories from sf.db into the system prompt; no references to old memory systems remain in the codebase",
+			"created_at": "2026-05-06T20:52:17.319Z",
+			"completed_at": null,
+			"full_summary_md": "",
+			"full_uat_md": "",
+			"goal": "Replace all system prompt injection paths to use the unified memory store, remove dead code, and verify end-to-end",
+			"success_criteria": "Not provided.",
+			"proof_level": "Not provided.",
+			"integration_closure": "Not provided.",
+			"observability_impact": "Not provided.",
+			"adversarial_partner": "Missing partner review.",
+			"adversarial_combatant": "Missing combatant review.",
+			"adversarial_architect": "Missing architect review.",
+			"planning_meeting": {
+				"trigger": "Need to integrate unified memory store into system prompt injection and remove all legacy memory code paths",
+				"pm": "The system prompt currently loads KNOWLEDGE.md and memory-store.js separately. We need to replace this with a single MemoryStore.getRelevantMemoriesRanked() call that injects ranked memories into the prompt. This simplifies the context assembly and ensures all memory flows through one path.",
+				"researcher": "system-context.js currently loads: knowledgeBlock (from KNOWLEDGE.md files) and memoryBlock (from memory-store.js getActiveMemoriesRanked). We need to replace both with a single call to MemoryStore.getRelevantMemoriesRanked(query, limit) where query is derived from the current task context.",
+				"partner": "The integration should: (1) replace knowledgeBlock + memoryBlock with a single memory injection, (2) use task context keywords for relevance scoring, (3) format memories with category headers, (4) respect token budget, (5) remove all imports of old memory systems.",
+				"combatant": "Removing KNOWLEDGE.md injection might lose important human-curated rules that aren't in the memory store yet. We should ensure migration is complete before removing the old path, or keep a fallback for unmigrated knowledge.",
+				"architect": "Agree on fallback. The integration should: (1) try MemoryStore first, (2) if empty, fall back to KNOWLEDGE.md (read-only, no new writes), (3) inject formatted memories into system prompt via before_agent_start hook, (4) track memory usage for feedback loop.",
+				"moderator": "Decision: Replace system prompt injection with MemoryStore.getRelevantMemoriesRanked(). Add a temporary fallback to KNOWLEDGE.md if MemoryStore returns empty (until migration is complete). Remove all old memory system imports. Add memory usage tracking. Run end-to-end test.",
+				"recommendedRoute": "planning",
+				"confidenceSummary": "High confidence. The integration is straightforward — replace two injection paths with one. The fallback ensures no regression during migration."
+			},
+			"sequence": 4,
+			"replan_triggered_at": null
+		}
+	],
+	"tasks": [
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S01",
+			"id": "T01",
+			"title": "Audit and extend sf.db memories schema",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Read sf-db.js schema initialization to verify memories table exists with correct columns\n2. Add `source` column (TEXT, default 'manual') to distinguish: 'manual', 'extracted', 'migrated'\n3. Add `session_id` column (TEXT, nullable) to link extracted memories to source sessions\n4. Bump SCHEMA_VERSION in sf-db.js\n5. Write migration SQL that adds columns if they don't exist (for existing databases)\n6. Verify: run tests that create and query memories with new columns",
+			"estimate": "1h",
+			"files": [
+				"src/resources/extensions/sf/sf-db.js",
+				"src/resources/extensions/sf/memory-store.js"
+			],
+			"verify": "npm test -- --grep \"memory schema\"",
+			"inputs": ["src/resources/extensions/sf/sf-db.js"],
+			"expected_output": ["src/resources/extensions/sf/sf-db.js (modified)"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S01",
+			"id": "T02",
+			"title": "Create MemoryStore class wrapper",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Create src/resources/extensions/sf/MemoryStore.ts (or .js) that wraps existing memory-store.js functions\n2. Expose methods: createMemory, updateMemory, reinforceMemory, supersedeMemory, getActiveMemoriesRanked, getRelevantMemoriesRanked, formatMemoriesForPrompt\n3. Add source parameter to createMemory (default 'manual')\n4. Ensure all methods delegate to existing sf-db.js wrappers\n5. Export both the class and individual functions for backward compatibility\n6. Verify: unit tests pass for class and function APIs",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/MemoryStore.ts",
+				"src/resources/extensions/sf/memory-store.js"
+			],
+			"verify": "npm test -- --grep \"MemoryStore\"",
+			"inputs": [
+				"src/resources/extensions/sf/memory-store.js",
+				"src/resources/extensions/sf/sf-db.js"
+			],
+			"expected_output": ["src/resources/extensions/sf/MemoryStore.ts"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S01",
+			"id": "T03",
+			"title": "Write unit tests for MemoryStore CRUD and ranking",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Create src/resources/extensions/sf/tests/memory-store.test.ts\n2. Test CRUD: createMemory, getActiveMemories, updateMemory, reinforceMemory, supersedeMemory\n3. Test ranking: getActiveMemoriesRanked returns correct order by confidence * hit_count\n4. Test graceful degradation: returns empty arrays when DB unavailable\n5. Test source filtering: memories created with different sources are queryable\n6. Verify: all tests pass",
+			"estimate": "1.5h",
+			"files": ["src/resources/extensions/sf/tests/memory-store.test.ts"],
+			"verify": "npm test -- src/resources/extensions/sf/tests/memory-store.test.ts",
+			"inputs": [
+				"src/resources/extensions/sf/MemoryStore.ts",
+				"src/resources/extensions/sf/memory-store.js"
+			],
+			"expected_output": [
+				"src/resources/extensions/sf/tests/memory-store.test.ts"
+			],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S02",
+			"id": "T01",
+			"title": "Port session scanner to use MemoryStore",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Copy scanSessionFiles and filterSessionContent from memory extension\n2. Adapt to read from ~/.sf/agent/sessions/ directory structure\n3. Add session fingerprinting (hash of content) to avoid re-processing\n4. Check memory_processed_units table before processing\n5. Verify: scanner finds session files and filters correctly",
+			"estimate": "1h",
+			"files": ["src/resources/extensions/sf/memory-pipeline.js"],
+			"verify": "npm test -- --grep \"session scanner\"",
+			"inputs": [
+				"packages/pi-coding-agent/src/resources/extensions/memory/pipeline.ts"
+			],
+			"expected_output": ["src/resources/extensions/sf/memory-pipeline.js"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S02",
+			"id": "T02",
+			"title": "Implement Phase 1: LLM extraction to MemoryStore",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Port runPhase1 from memory extension\n2. Use completeSimple from pi-ai for LLM calls\n3. Extract memories as JSON array with category/content/confidence\n4. Write each extracted memory to MemoryStore with source='extracted' and session_id\n5. Add rate limiting: max 10 memories per session, min confidence 0.6\n6. Mark session as processed in memory_processed_units\n7. Verify: extraction creates memories in sf.db",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/memory-pipeline.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"verify": "npm test -- --grep \"memory extraction\"",
+			"inputs": [
+				"src/resources/extensions/sf/memory-pipeline.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": [
+				"src/resources/extensions/sf/memory-pipeline.js (phase 1)"
+			],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S02",
+			"id": "T03",
+			"title": "Implement Phase 2: Consolidation and /memory command",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Port runPhase2 consolidation logic\n2. Query MemoryStore for source='extracted' memories\n3. Group by category, deduplicate similar content (simple string similarity)\n4. Update consolidated memories with higher confidence\n5. Supersede duplicate memories\n6. Implement /memory command that reads from MemoryStore\n7. Verify: consolidation runs and /memory shows extracted memories",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/memory-pipeline.js",
+				"src/resources/extensions/sf/memory-commands.js"
+			],
+			"verify": "npm test -- --grep \"memory consolidation\"",
+			"inputs": [
+				"src/resources/extensions/sf/memory-pipeline.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": ["src/resources/extensions/sf/memory-commands.js"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S03",
+			"id": "T01",
+			"title": "Create KNOWLEDGE.md parser and migrator",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Create src/resources/extensions/sf/migrate-knowledge.js\n2. Parse KNOWLEDGE.md sections (Rules, Patterns, Lessons Learned)\n3. Extract table rows as memory entries\n4. Map to memory categories: rule→convention, pattern→pattern, lesson→gotcha\n5. Insert into MemoryStore with source='migrated'\n6. Deduplicate by content hash against existing memories\n7. Add dry-run mode (report without inserting)\n8. Verify: dry-run reports correct counts",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/migrate-knowledge.js",
+				".sf/KNOWLEDGE.md",
+				"KNOWLEDGE.md"
+			],
+			"verify": "node src/resources/extensions/sf/migrate-knowledge.js --dry-run",
+			"inputs": [
+				".sf/KNOWLEDGE.md",
+				"KNOWLEDGE.md",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": ["src/resources/extensions/sf/migrate-knowledge.js"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S03",
+			"id": "T02",
+			"title": "Create agent.db migrator",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Create src/resources/extensions/sf/migrate-agent-db.js\n2. Read old agent.db using sql.js\n3. Extract stage1_outputs.extraction_json fields\n4. Parse JSON arrays of memories\n5. Insert into MemoryStore with source='extracted'\n6. Deduplicate by content hash\n7. Handle corrupted databases gracefully (skip with warning)\n8. Add dry-run mode\n9. Verify: dry-run reports correct counts",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/migrate-agent-db.js",
+				"~/.sf/agent/agent.db"
+			],
+			"verify": "node src/resources/extensions/sf/migrate-agent-db.js --dry-run",
+			"inputs": [
+				"packages/pi-coding-agent/src/resources/extensions/memory/storage.ts",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": ["src/resources/extensions/sf/migrate-agent-db.js"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S03",
+			"id": "T03",
+			"title": "Deprecate legacy memory systems",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Add deprecation warnings to memory-store.js exports (log warning on first use)\n2. Remove knowledge-compounding.js and knowledge-injector.js or make them no-op\n3. Update memory extension to use MemoryStore instead of sql.js\n4. Remove sql.js dependency from memory extension\n5. Update CHANGELOG.md with migration instructions\n6. Verify: no imports of old memory systems in active code paths",
+			"estimate": "1h",
+			"files": [
+				"src/resources/extensions/sf/memory-store.js",
+				"src/resources/extensions/sf/knowledge-compounding.js",
+				"src/resources/extensions/sf/knowledge-injector.js",
+				"packages/pi-coding-agent/src/resources/extensions/memory/index.ts",
+				"packages/pi-coding-agent/src/resources/extensions/memory/storage.ts",
+				"CHANGELOG.md"
+			],
+			"verify": "grep -r \"knowledge-compounding\\|knowledge-injector\\|sql.js\" src/ packages/ --include=\"*.ts\" --include=\"*.js\" | grep -v \"test\\|spec\\|deprecated\"",
+			"inputs": [
+				"src/resources/extensions/sf/migrate-knowledge.js",
+				"src/resources/extensions/sf/migrate-agent-db.js"
+			],
+			"expected_output": [
+				"src/resources/extensions/sf/memory-store.js (deprecated)",
+				"CHANGELOG.md (updated)"
+			],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S04",
+			"id": "T01",
+			"title": "Integrate MemoryStore into system prompt injection",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Read system-context.js to understand current memory injection\n2. Replace knowledgeBlock + memoryBlock with MemoryStore.getRelevantMemoriesRanked()\n3. Derive query from task context (domain, keywords, technology)\n4. Format memories with category headers\n5. Respect token budget (truncate if needed)\n6. Add fallback to KNOWLEDGE.md if MemoryStore empty\n7. Verify: system prompt contains formatted memories",
+			"estimate": "1.5h",
+			"files": [
+				"src/resources/extensions/sf/bootstrap/system-context.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"verify": "npm test -- --grep \"system context memory\"",
+			"inputs": [
+				"src/resources/extensions/sf/bootstrap/system-context.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": [
+				"src/resources/extensions/sf/bootstrap/system-context.js (modified)"
+			],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S04",
+			"id": "T02",
+			"title": "Remove legacy memory system imports",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Remove all imports of memory-store.js from active code\n2. Remove all imports of knowledge-injector.js\n3. Remove all imports of knowledge-compounding.js\n4. Update package.json to remove better-sqlite3 and sql.js from memory-related dependencies\n5. Verify: grep returns no active references to old systems",
+			"estimate": "1h",
+			"files": [
+				"src/resources/extensions/sf/**/*.js",
+				"src/resources/extensions/sf/**/*.ts",
+				"packages/pi-coding-agent/src/resources/extensions/memory/**/*.ts"
+			],
+			"verify": "grep -r \"knowledge-injector\\|knowledge-compounding\\|memory-store\" src/ packages/ --include=\"*.ts\" --include=\"*.js\" | grep -v \"test\\|spec\\|deprecated\"",
+			"inputs": ["src/resources/extensions/sf/bootstrap/system-context.js"],
+			"expected_output": ["Clean codebase with no legacy memory imports"],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		},
+		{
+			"milestone_id": "M001-6377a4",
+			"slice_id": "S04",
+			"id": "T03",
+			"title": "Write end-to-end integration tests",
+			"status": "pending",
+			"one_liner": "",
+			"narrative": "",
+			"verification_result": "",
+			"duration": "",
+			"completed_at": null,
+			"blocker_discovered": false,
+			"deviations": "",
+			"known_issues": "",
+			"key_files": [],
+			"key_decisions": [],
+			"full_summary_md": "",
+			"description": "1. Create end-to-end test: mock session with memory-worthy content\n2. Trigger memory extraction pipeline\n3. Start new session with matching context\n4. Verify system prompt contains the extracted memory\n5. Test graceful degradation: empty store returns empty string\n6. Test token budget: large memory set is truncated\n7. Verify: all tests pass",
+			"estimate": "2h",
+			"files": ["src/resources/extensions/sf/tests/memory-e2e.test.ts"],
+			"verify": "npm test -- src/resources/extensions/sf/tests/memory-e2e.test.ts",
+			"inputs": [
+				"src/resources/extensions/sf/bootstrap/system-context.js",
+				"src/resources/extensions/sf/memory-pipeline.js",
+				"src/resources/extensions/sf/MemoryStore.ts"
+			],
+			"expected_output": [
+				"src/resources/extensions/sf/tests/memory-e2e.test.ts"
+			],
+			"observability_impact": "",
+			"full_plan_md": "",
+			"sequence": 0
+		}
+	],
+	"decisions": [
+		{
+			"seq": 1,
+			"id": "D001",
+			"when_context": "",
+			"scope": "environment",
+			"decision": "How to handle missing libsecret dependency for keytar native addon on Linux.",
+			"choice": "Add libsecret-1-0 / libsecret-1-dev to all Linux environment definitions (Docker, Nix, CI).",
+			"rationale": "@github/keytar (used by @google/gemini-cli-core) requires libsecret on Linux. Without it, keychain initialization fails with a cryptic \"Failed to load keytar native addon\" error. Adding it to the core environment definitions ensures it is present in all official SF environments.",
+			"revisable": "Yes",
+			"made_by": "agent",
+			"superseded_by": null
+		}
+	],
+	"verification_evidence": []
+}