- Set resourceVersionOnStart on paused-session resume so resource
staleness detection works for resumed sessions.
- Re-register setLevelChangeCallback on resume so health-level
transition notifications fire after process restart.
- Persist unitType/unitId in paused-session metadata and restore them
on resume so recovery synthesis framing text shows the actual unit
instead of "unknown"/"unknown".
- Check worktreePath existence before showing "(worktree)" in resume
notification to avoid misleading the user when the worktree was
already torn down.
- Fix showDiscuss skip_milestone to use step:false (matching
discuss_draft and discuss_fresh) instead of hardcoded step:true.
- Initialize autoStartTime, originalModelId, originalModelProvider, and
autoModeStartModel on file-based paused-session resume path so
dashboard elapsed time, model restore on stop, and error-recovery
model fallback all work correctly.
- Validate worktreePath existence before using it as assessment base;
fall back to project root when the worktree has been torn down.
- Add tests for paused-session-only (no lock) with resumable disk state
and for worktreePath fallback when the directory no longer exists.
- Re-derive state from project root after stale/recoverable cleanup in
guided flow; the assessment may have derived state from a worktree
path that was cleaned up, leading to stale state being used downstream.
- Add test for the "none" classification path (clean startup with no
lock and no paused session) to close coverage gap.
- Remove redundant `void verboseMode` suppression (parameter is used at line 364)
- Simplify hasStrongRecoverySignal: `hasResumableDiskState || recoveryToolCallCount > 0`
(the previous expression had a redundant third disjunct)
- Remove extra blank line in auto.ts between stepMode assignment and lock-clear block
Use paused-session worktree metadata when deriving interrupted-session state so
worktree isolation decisions follow the authoritative disk state without
changing /gsd discuss auto handoff semantics.
Remove a dead running-state comparison after the early return in startAuto so
extension typechecking passes alongside the interrupted-session regressions.
Discard stale paused-session metadata on direct auto starts and pass paused
step-mode through guided resume so the selected interrupted-session action
matches the actual resumed mode.
Pass paused step-mode metadata through guided resume actions and tighten source
coverage around stale paused-session handling so the interrupted-session flow
matches the selected resume mode.
Remove the duplicated auto bootstrap block, keep paused step-mode sessions on
/gsd next, and treat stale paused-session metadata as non-resumable when disk
state shows no unfinished work.
Treat interrupted sessions as resumable only when paused metadata, crash trace,
or unfinished derived state indicate real work remains. This clears dead locks
for completed/no-op repos and keeps guided flow plus /gsd auto consistent.
* fix: prune stale env-utils.js from extensions root, preventing startup load error
- Move env-utils.ts from extensions/ root into gsd/ subdirectory
- Update all import paths to reflect new location
- Add manifest-based tracking in resource-loader to record which root-level
extension files are installed, so future upgrades can detect and prune files
that get removed or relocated (preventing recurrence)
- Add known-stale fallback for pre-manifest upgrades (explicitly removes
env-utils.js which was moved into gsd/ in this release)
- Remove re-export block from auto.ts that referenced relocated symbols
- Clean up session_start handler in native-search.ts (remove provider diagnostics
that were duplicating info already shown by model_select)
- Update welcome-screen layout to two-panel bar design for visual consistency
* fix: resolve PR1655 extension load and compile regressions
* fix: remove duplicate _clearGsdRootCache export
* fix: restore native-search session_start diagnostics
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* feat: add Skill tool resolution for Pi agent
Expose a built-in Skill tool so dispatched prompts can resolve skill names without guessing file paths. This aligns runtime behavior with skill activation prompts and adds coverage for exact activation and unknown-skill handling.
Replaces the rounded box-corner two-panel layout (╭╮╰╯) with full-width
cyan ─ bars at top and bottom, matching the auto-mode progress widget's
ui.bar() style exactly. The inner │ divider and ├─ section separators are
kept (dimmed) so the two-panel logo/info layout is preserved.
Changes:
- Top/bottom borders: chalk.cyan('─'.repeat(termWidth)) — same as widget ui.bar()
- Outer vertical box borders removed; inner │ divider kept as dim separator
- Section dividers changed to dim ├──── style
- Trailing spaces removed from hint/version strings (no closing │ to pad against)
- Panel width formula updated: 1 + LEFT_INNER + 1 + RIGHT_INNER = termWidth
* feat: surface real doctor issue details in progress score widget
Previously the progress score traffic light (green/yellow/red) only
showed generic labels like "2 consecutive error units" or "Health
trend declining". The actual doctor issue descriptions were computed
in auto-post-unit but discarded before reaching the widget — only
aggregate counts were stored in HealthSnapshot.
Now the full data flows through:
- HealthSnapshot stores issue details (code, message, severity,
unitId) and fix descriptions alongside the counts
- recordHealthSnapshot() accepts optional issue/fix arrays
(backwards compatible — existing callers unchanged)
- getLatestHealthIssues() and getLatestHealthFixes() retrieve the
most recent details for display
- computeProgressScore() surfaces up to 5 real issue messages
(errors first) and up to 3 recent fixes as ProgressSignals
when the level is yellow or red
- Dashboard overlay renders signal details with ✓/✗/· icons
below the traffic light when degraded
This gives real-time visibility into what the auto-doctor is
detecting and fixing, without requiring manual /gsd doctor runs
or opening the full dashboard to investigate.
* feat: integrate doctor health data into visualizer and HTML reports
Phase 2b: close visibility gaps across visualizer and export surfaces.
Persistence (doctor.ts):
- Enrich DoctorHistoryEntry with issue details (severity, code,
message, unitId) and fix descriptions
- appendDoctorHistory now persists up to 10 issues per entry and
all fix descriptions to doctor-history.jsonl
- Export DoctorHistoryEntry type for consumers
Data layer (visualizer-data.ts):
- Add VisualizerDoctorEntry and VisualizerProgressScore types
- Extend HealthInfo with doctorHistory (last 20 persisted entries)
and progressScore (current in-memory traffic light)
- loadHealth reads doctor-history.jsonl synchronously and snapshots
current progress score when health data exists
TUI visualizer (visualizer-views.ts):
- Health tab now shows "Progress Score" section with traffic light
icon, summary, and all signal details (✓/✗/· prefixed)
- Health tab now shows "Doctor History" section with timestamped
entries, issue messages, and applied fixes
HTML export (export-html.ts):
- Health section includes progress score with colored indicator
and signal breakdown
- Health section includes "Doctor Run History" table with
timestamps, error/warning/fix counts, issue codes, expandable
issue messages, and fix descriptions
* feat: fill remaining health gaps — scope tagging, level notifications, human-readable logs
Gap fills:
Per-milestone/slice scope tagging:
- HealthSnapshot now stores scope (e.g. "M001/S02") from the
doctor run's unit context
- DoctorHistoryEntry persists scope to doctor-history.jsonl
- Visualizer and HTML reports display scope tags per entry
State transition notifications:
- setLevelChangeCallback() registers a handler for progress level
changes (green→yellow, yellow→red, red→green, etc.)
- auto-start.ts wires the callback to ctx.ui.notify on start
- auto.ts clears it on stop
- Notifications include the triggering issue message
Human-readable formatting throughout:
- formatHealthSummary() uses full words: "2 errors, 3 warnings ·
trend degrading · 1 fix applied · 1 of 5 consecutive errors
before escalation · latest: Missing PLAN.md for S03"
- DoctorHistoryEntry stores a human-readable summary field
built from error counts, fix counts, and top issue message
- Visualizer doctor history shows summary instead of "2E 1W 0F"
- HTML export doctor table uses summary column with scope tags
- Post-unit notification says what was fixed ("Doctor: rebuilt
STATE.md; cleared stale lock") instead of "applied 2 fix(es)"
Test updates:
- formatHealthSummary assertions updated for new readable format
* fix: default UAT type to artifact-driven to prevent unnecessary auto-mode pauses (#1651)
When a UAT file has no `## UAT Type` section, `extractUatType()` returns
`undefined`. The fallback was `"human-experience"`, causing `pauseAfterDispatch:
true` in the auto-dispatch rule. Since doctor-generated UAT placeholders never
include a UAT Type section and LLM-executed UATs are always artifact-driven,
the correct default is `"artifact-driven"`.
Closes#1649
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: remove duplicate doctorScope declaration (CI build fix)
* fix: resolve PR1644 regressions in health views and post-unit hook
---------
Co-authored-by: TÂCHES <afromanguy@me.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: add PhaseResult/IterationContext/LoopState types to auto-loop
No behavioral changes. Pure type additions for upcoming phase extraction.
* refactor: extract runFinalize() from autoLoop body
No behavioral changes. Pure structural extraction.
* refactor: extract runUnitPhase() from autoLoop body
No behavioral changes. Pure structural extraction.
* refactor: extract runGuards() from autoLoop body
No behavioral changes. Pure structural extraction.
* refactor: extract runDispatch() from autoLoop body
No behavioral changes. Pure structural extraction.
* refactor: extract runPreDispatch() from autoLoop body
No behavioral changes. Pure structural extraction.
Completes autoLoop pipeline phase decomposition:
runPreDispatch → runGuards → runDispatch → runUnitPhase → runFinalize
* refactor: hoist loopState before autoLoop loop, drop sync-back hacks
loopState was created inside the loop each iteration, requiring 3 manual
sync-backs for stuckRecoveryAttempts (number copy-by-value). Hoist it
before the loop so it's a true persistent mutable struct across iterations.
The forensics prompt listed only 5 source files and told the agent to
"analyze the report." This led to shallow analysis and hallucinated
paths because the agent had no knowledge of the source layout, runtime
paths, activity log format, or crash lock structure.
The rewritten prompt gives the forensics agent a complete source map
organized by domain, the full .gsd/ directory structure, data format
references for activity logs / crash locks / metrics, and a step-by-step
investigation protocol that requires tracing from symptom to specific
file:line in GSD source before filing an issue.
Closes#1656
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The forensics prompt references "activity logs" in natural language but
never provides the actual filesystem paths. This causes the LLM agent to
hallucinate paths like `activity-logs/` when it needs to inspect raw JSONL
logs beyond the pre-parsed forensic data.
Adds a "Key Runtime Paths" section with concrete `.gsd/` paths for
activity logs, debug logs, runtime state, crash lock, completed units,
and forensics reports.
Closes#1652
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The progress widget's render() synchronously accesses sessionManager
state via cmdCtx. When newSession() is in-flight, this can block the
TUI input loop, freezing the terminal. Guard render() to return the
last cached frame while a session switch is in progress.
Closes#1653
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Refactor GSD command/bootstrap modules
* fix: resolve TypeScript build errors in refactored db-tools and catalog
- db-tools.ts: add missing execute callback params (signal, onUpdate, ctx),
remove isError from return objects (not in AgentToolResult type), cast
details as any to avoid union type mismatch across error/success paths
- catalog.ts: use Object.entries() on TemplateRegistry.templates Record
instead of treating it as an array, use Record key as template id
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: update source-contract tests to reference refactored file locations
The god-file refactor moved code from index.ts and commands.ts into
bootstrap/agent-end-recovery.ts, bootstrap/register-hooks.ts, and
commands/handlers/core.ts. Update three test files to read from the
correct paths and adjust pattern assertions to match the new code
structure.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a UAT file has no `## UAT Type` section, `extractUatType()` returns
`undefined`. The fallback was `"human-experience"`, causing `pauseAfterDispatch:
true` in the auto-dispatch rule. Since doctor-generated UAT placeholders never
include a UAT Type section and LLM-executed UATs are always artifact-driven,
the correct default is `"artifact-driven"`.
Closes#1649
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a unit spawns background jobs via async_bash, job completion callbacks
fire follow-up messages after agent_end has resolved. The auto-loop has
moved on but the previous session's LLM processes these follow-ups, adding
12-45s of wasted time and ~14 unnecessary turns per unit.
Two complementary fixes:
1. Cancel all running background jobs on session_before_switch so
completion callbacks never fire for the old session.
2. Clear the follow-up queue after runUnit() completes as defense-in-depth,
discarding any already-queued notifications before the next session starts.
Closes#1642
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* feat(gsd): activate matching skills in dispatched prompts
Inject skill activations from installed skills, preferences, and task-plan handoff so GSD agents load the right skills automatically instead of relying on generic guidance. Align prompt templates and tests with the activation flow and current resource sync behavior.
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* fix(gsd extension): restore health widget build paths
* test(resource-loader): fix sibling cleanup assertion
When a milestone has a roadmap with unchecked slice checkboxes AND a
summary file, deriveState() incorrectly treated it as incomplete. The
summary check only ran inside the `if (isMilestoneComplete(roadmap))`
branch, so it was never reached when checkboxes weren't ticked.
This caused auto-mode to pick an already-completed milestone as active,
ignoring the actual current milestone entirely.
The fix adds summary-existence checks to all three resolution paths:
1. `getActiveMilestoneId()` — now checks for summary before returning
a milestone as incomplete
2. Phase 1 pre-scan in `deriveState()` — now adds milestones with
unchecked roadmaps + summaries to `completeMilestoneIds`
3. Phase 2 registry builder — now checks for summary before falling
through to the active/pending logic
This is consistent with the existing principle that the summary is the
terminal artifact (#864), which was already stated in a comment but not
enforced for the unchecked-roadmap case.
Adds two tests:
- Unchecked roadmap + summary → status is 'complete', next milestone
is active
- Unchecked roadmap + summary satisfies depends_on for downstream
milestones
Two targeted fixes that close the three remaining paths where .gsd/
tracked files can still be silently deleted after the v2.36.0 fix.
--- Path 1: hasGitTrackedGsdFiles fails open on git error (gitignore.ts)
nativeLsFiles() swallows git failures via allowFailure=true and returns
[], making hasGitTrackedGsdFiles() indistinguishable between "nothing
tracked" and "git failed". On any transient git failure (locked index,
binary not on PATH, corrupted .git/index), the function returned false
and .gsd was added to .gitignore, deleting all tracked state.
Fix: after nativeLsFiles returns [], verify git is reachable with a
cheap rev-parse call. If git is unavailable, return true (fail safe —
assume tracked). The outer catch also returns true instead of false.
--- Path 2: migration never cleans git index (migrate-external.ts)
migrateToExternalState() correctly creates the .gsd symlink/junction but
never ran `git rm -r --cached .gsd/`. All previously tracked .gsd/* files
remained in the git index pointing through the new symlink, which git
cannot follow — causing PROJECT.md, milestones/, REQUIREMENTS.md etc. to
appear as deleted in git status immediately after every migration.
Fix: after the symlink is verified, run:
git rm -r --cached --ignore-unmatch .gsd
--ignore-unmatch makes this a no-op on fresh/untracked projects.
--- Path 3: race between migration and ensureGitignore
Resolved by Path 2. If migration always cleans the index, the race
window (another process converting .gsd/ to a symlink between the
migrateToExternalState() and ensureGitignore() calls) is harmless —
the index is already clean and there is nothing to lose.
--- Tests added (gitignore-tracked-gsd.test.ts)
- hasGitTrackedGsdFiles returns true (fail-safe) when git is unavailable
(simulated via .git/index.lock to force git ls-files failure)
- migrateToExternalState cleans git index so tracked files don't show
as deleted after successful migration
Fixes residual vectors from #1364 (original fix: #1367, v2.36.0)
Template for projects to declare stack, build, test, and environment
details. Inlined into execute-task prompts when present.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds createDraftPR() to git-service.ts and hooks it into the milestone
transition block in auto-loop.ts. Best-effort, non-fatal on failure.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Native ESM import() ignores NODE_PATH and resolves packages by walking up
the directory tree. Extension files synced to ~/.gsd/agent/extensions/ have
no ancestor node_modules, so imports of @gsd/* packages fail with "Cannot
find package" errors during report generation and other dynamic-import paths.
Create a symlink ~/.gsd/agent/node_modules -> GSD's node_modules after
resource sync so Node's standard resolution finds @gsd/* packages. Also
migrate the most critical dynamic imports in auto-loop, exit-command, and
commands to use importExtensionModule (jiti-based) as a belt-and-suspenders
fix.
Closes#1594
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two bugs fixed:
1. recordHealthSnapshot counted ALL doctor issues including cross-milestone
stale errors, inflating consecutiveErrorUnits past the escalation threshold
from unfixable errors in other milestones. Now filters report.issues to
only the current milestone before summarizing for health tracking.
2. matchesScope used unitId.startsWith(scope) without a delimiter, so scope
"M004/S01" would false-match "M004/S010". Removed the redundant
delimiter-less startsWith branch — exact match and slash-delimited
startsWith are sufficient.
Closes#1579
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor(auto-loop): hoist MAX_RECOVERY_CHARS to module level
Constant was defined inside the while loop body on every iteration.
Moved to module level next to MAX_LOOP_ITERATIONS.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor(auto-loop): cache loadEffectiveGSDPreferences() once per iteration
Was called 9 times per loop iteration. Now called once at the top of the
try block and stored in `prefs`, used throughout the iteration.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
New UAT types skip human pause, enabling automated browser and script
verification by the engine.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
capPreamble() enforces MAX_PREAMBLE_CHARS via truncateAtSectionBoundary,
applied to all inlinedContext assembly points. Replaces deleted compression
subsystem with a simple deterministic cap.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the crude sameUnitCount counter with a sliding window (size 6)
that detects three stuck patterns:
1. Same error repeated twice in a row → stuck immediately
2. Same unit derived 3 consecutive times → stuck (was 5, now faster)
3. Oscillation pattern A→B→A→B → stuck (previously undetected)
Graduated recovery preserved: first detection triggers cache invalidation
+ retry, second detection triggers hard stop.
Exported detectStuck() function with 8 unit tests covering all rules
plus edge cases (truncation, priority, non-triggers).
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
dispatchWorkflow now resolves per-phase model preferences (e.g.,
models.planning, models.execution) via resolveModelWithFallbacksForUnit
and applies them with pi.setModel before dispatching the workflow message.
All 22 call sites pass the appropriate unit type context so planning,
research, execution, and completion phases each use the configured model.
Closes#1582
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
closeoutUnit() ran at the start of the next loop iteration, creating a
window where a crash between runUnit() returning and the next iteration
would lose all telemetry (metrics, activity log, memory extraction).
completed-units.json was also never flushed to disk, causing severe
staleness (3 entries for 322 completed units in production).
Closes#1590
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
smartStage() ran `git add -A` on the entire repo then unstaged exclusions,
causing indefinite hangs on repos with large untracked artifact trees (57GB+).
autoCommitDirtyState() bypassed smartStage() entirely via direct nativeAddAll().
Add nativeAddAllWithExclusions() using `git add -A -- ':!pattern'` syntax so
excluded paths are never hashed. Route autoCommitDirtyState() through it with
RUNTIME_EXCLUSION_PATHS.
Closes#1605
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a slice is marked [x] in ROADMAP but tasks are incomplete and no
summary exists, doctor detects slice_checked_missing_summary (declared
fixable) but had no shouldFix handler — creating an unrecoverable
deadlock. Add handler that unchecks the slice when tasks are incomplete,
and add markSliceUndoneInRoadmap to both doctor.ts and
roadmap-mutations.ts.
Closes#1591
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The MCP client passed raw "${VAR}" strings to child processes instead of
resolving them against process.env, breaking MCP servers that expect
resolved environment variable values.
Adds a resolveEnv() helper that interpolates ${VAR} patterns in env
config values before passing them to StdioClientTransport.
Closes#1599
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
After dispatchDoctorHeal fires pi.sendMessage({ triggerTurn: true }),
the function fell through to return "continue". The auto-loop treated
"continue" as "proceed to next unit", called newSession() while the
session manager was still processing the heal turn, and the 30s timeout
killed auto-mode.
Returning "dispatched" causes the auto-loop to break, letting the heal
turn complete and trigger its own handleAgentEnd to resume the loop.
Closes#1580
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
In manual sessions (no auto-mode), bootstrapAutoSession never runs, so the
GSD database is never opened. This causes gsd_save_decision,
gsd_update_requirement, and gsd_save_summary tools to always fail with
'GSD database is not available'.
Add ensureDbOpen() helper that checks isDbAvailable() first, then tries to
open the DB from the expected .gsd/gsd.db path if it exists. All three tool
handlers now use this helper instead of the check-only pattern.
The fix is backward-compatible: in auto-mode the DB is already open, so
ensureDbOpen() returns true immediately on the isDbAvailable() check.
Move `pendingResolve` and `sessionSwitchInFlight` from AutoSession to
module-level variables in auto-loop.ts (`_currentResolve`,
`_sessionSwitchInFlight`). Remove `pendingAgentEndQueue` entirely —
agent_end events arriving with no pending resolver are now dropped
(with a debug warning) instead of queued.
This eliminates the `_activeSession` singleton, the queue drain logic
in `runUnit`, and three properties from `AutoSession.reset()`.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extract validateProjectId() and validate at startup in
bootstrapAutoSession() so users get immediate feedback on invalid
values. repoIdentity() returns the custom ID directly when set.
Delete prompt-compressor, summary-distiller, and semantic-chunker modules
plus all associated tests. Replace all compression/distillation/chunking
call sites with section-boundary truncation via truncateAtSectionBoundary.
Remove compression_strategy preference, validation, and documentation.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Each cleanup group in stopAuto is wrapped in its own try/catch so a
failure in one step (e.g., worktree exit, DB close, model restore)
cannot abort remaining cleanup. Critical invariants (s.active=false,
s.paused=false, UI reset, pendingResolve=null) are moved into a
finally block that executes unconditionally.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(gsd): tighten prompt automation contracts
* fix(gsd): restore confirmation gates for reflection/requirements/roadmap, scope workflow autonomy by complexity
Amends PR #1556 to address two behavioral risks:
1. discuss.md: Remove "treat continuation as confirmation" fallthrough —
elaboration is not confirmation. Restore explicit confirmation gates
for requirements and roadmap preview.
2. workflow-start.md: Gate autonomy on {{complexity}} — low/medium
workflows keep moving by default, high complexity workflows confirm
at phase transitions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
