Merge pull request #420 from davesienkowski/fix/executor-completion-verification

fix(executor): add completion verification to prevent hallucinated success (#315)

agents/gsd-executor.md

@@ -678,6 +678,45 @@ During execution, these authentication requirements were handled:
 
 </summary_creation>
 
+<self_check>
+After writing SUMMARY.md, verify your own claims before proceeding.
+
+**1. Check created files exist:**
+
+Parse `key-files.created` from the SUMMARY frontmatter. For each file listed:
+```bash
+[ -f "path/to/file" ] && echo "FOUND: path/to/file" || echo "MISSING: path/to/file"
+```
+If `key-files.created` is empty (docs-only plans), skip this check.
+
+**2. Check commits exist:**
+
+Parse commit hashes from the "Task Commits" section. Verify each hash exists:
+```bash
+git log --oneline --all | grep -q "{hash}" && echo "FOUND: {hash}" || echo "MISSING: {hash}"
+```
+
+**3. Append self-check result to SUMMARY.md:**
+
+If ANY file or commit is missing, append to SUMMARY.md:
+```markdown
+## Self-Check: FAILED
+
+Missing files:
+- path/to/missing-file.ts
+
+Missing commits:
+- abc123f
+```
+
+If all checks pass, append:
+```markdown
+## Self-Check: PASSED
+```
+
+Do NOT skip this step. Do NOT proceed to state updates if self-check fails — the SUMMARY must reflect reality.
+</self_check>
+
 <state_updates>
 After creating SUMMARY.md, update STATE.md.
 

commands/gsd/execute-phase.md

@@ -76,7 +76,7 @@ Phase: $ARGUMENTS
    For each wave in order:
    - Spawn `gsd-executor` for each plan in wave (parallel Task calls)
    - Wait for completion (Task blocks)
-   - Verify SUMMARYs created
+   - Verify SUMMARYs created and spot-check claims
    - Proceed to next wave
 
 5. **Aggregate results**

get-shit-done/workflows/execute-phase.md

@@ -337,6 +337,20 @@ Execute each wave in sequence. Autonomous plans within a wave run in parallel **
    - Read SUMMARY.md to extract what was built
    - Note any issues or deviations
 
+   **Spot-check claims before trusting SUMMARY:**
+
+   For each completed plan's SUMMARY.md:
+   - Pick the first 2 files from `key-files.created` frontmatter — verify they exist on disk with `[ -f ]`
+   - Check `git log --oneline --all --grep="{phase}-{plan}"` returns at least 1 commit
+   - Check SUMMARY.md for `## Self-Check: FAILED` marker
+
+   If ANY spot-check fails:
+   - Do NOT proceed silently
+   - Report which plan failed verification and what was missing
+   - Route to failure handler (step 4): ask user "Retry plan?" or "Continue with remaining waves?"
+
+   If spot-checks pass: proceed normally.
+
    **Output:**
    ```
    ---

get-shit-done/workflows/execute-plan.md

@@ -460,6 +460,13 @@ For Pattern A (fully autonomous) and Pattern C (decision-dependent), skip this s
 
    D. Report completion
 
+   E. Self-check the aggregated SUMMARY:
+      - Parse `key-files.created` from the SUMMARY frontmatter
+      - Verify first 2 files exist on disk with `[ -f ]`
+      - Check `git log --oneline --all --grep="{phase}-{plan}"` returns at least 1 commit
+      - If any check fails: append `## Self-Check: FAILED` section to SUMMARY listing missing items
+      - If all pass: append `## Self-Check: PASSED`
+
 **Example execution trace:**
 
 ````