diff --git a/src/resources/extensions/sf/skills/systematic-debugging/SKILL.md b/src/resources/extensions/sf/skills/systematic-debugging/SKILL.md
index 1eeecf1bd..d18bfe6fe 100644
--- a/src/resources/extensions/sf/skills/systematic-debugging/SKILL.md
+++ b/src/resources/extensions/sf/skills/systematic-debugging/SKILL.md
@@ -154,6 +154,32 @@ Use only when evidence collection runs in parallel:
 
 Do not delegate the fix — the main thread owns it. Subagents return paragraphs; you apply the change.
 
+## Milestone Bug Registry Guidance
+
+Some milestones (notably systematic bug-hunt milestones such as M008) ship with a curated `bug-registry.json` and a `triage-protocol.md` under `.sf/milestones/<MID>/`. When you're fixing a bug and either artifact is present for the active milestone, follow these additional rules:
+
+1. **Read the registry first** — before proposing a fix, check `.sf/milestones/<MID>/bugs/bug-registry.json` to see if the issue is already documented. If it is, use the `suggestedFix` field as a starting hypothesis, not as gospel. (Canonical example: `.sf/milestones/M008/bugs/bug-registry.json`.)
+2. **Follow the triage protocol** — classify severity using the definitions in `.sf/milestones/<MID>/triage-protocol.md`. HIGH and MEDIUM severity bugs require regression tests; LOW severity bugs should be batched with other fixes in the same cluster.
+3. **Update the registry** — after fixing, update `bug-registry.json`: set `status` to `"FIXED"`, add `fixedByTaskId`, and remove `assignedTo` if present.
+4. **Cluster-aware fixes** — if a fix touches files in multiple clusters, escalate to human review per the triage protocol escalation rules.
+5. **Confidence gate for batch fixes** — when fixing multiple LOW-severity bugs in one PR, each item must have a passing regression test and the batch confidence must be ≥ 0.80.
+
+When the active milestone has no `bug-registry.json`, this section does not apply — proceed with the normal evidence/repro/fix flow above.
+
+### Registry Fields Reference
+
+| Field | Meaning |
+|---|---|
+| `id` | Unique finding identifier (`<cluster>-<file>-<description>-<line>`) |
+| `file` | Source file where issue was found |
+| `lines` | Line number(s) |
+| `category` | Bug category (e.g. `logic`, `race condition`, `resource leak`) |
+| `severity` | `HIGH` / `MEDIUM` / `LOW` / `FALSE_POSITIVE` |
+| `status` | `CONFIRMED` / `FALSE_POSITIVE` / `FIXED` / `WONTFIX` / `IN_PROGRESS` |
+| `cluster` | Which cluster the finding belongs to |
+| `suggestedFix` | Recommended fix or `"N/A (intentional design)"` |
+| `fixedByTaskId` | Task ID that fixed the finding (set after fix) |
+
 ## Rules
 
 - Fix root cause, not symptom.
@@ -161,6 +187,7 @@ Do not delegate the fix — the main thread owns it. Subagents return paragraphs
 - Keep before/after repro for non-trivial bugs.
 - Three failed fix attempts → stop. Question the architecture. Discuss before attempt 4.
 - Never claim "fixed" without re-running the original repro path.
+- **Registry-backed milestones (e.g., M008)**: Never claim "fixed" without updating `bug-registry.json`.
 
 ## Red Flags — STOP
 
@@ -170,3 +197,4 @@ Do not delegate the fix — the main thread owns it. Subagents return paragraphs
 - Diagnosis stated without `Observed:` evidence.
 - Third fix attempt without questioning architecture.
 - Test deleted to make a build green — that's encoding the bug, not fixing it.
+- Fixing a bug listed in a milestone bug-registry without reading the registry first.