Session End Checklist

Before ending the session, complete these verification steps:

1. Work Verification

• All TodoWrite items marked as completed or documented as blocked
• All commits pushed to remote branch
• All code review suggestions addressed or documented as skipped (with reason)
• Tests pass: npm test
• Lint passes: npm run lint
• Pattern check passes: npm run patterns:check

2. CI Verification

If you modified any of these, verify they still work:

• .github/workflows/ci.yml - Main CI pipeline
• .github/workflows/docs-lint.yml - Documentation linting
• scripts/check-docs-light.js - Doc linter script
• scripts/check-pattern-compliance.js - Pattern checker
• eslint.config.mjs - ESLint configuration

3. Documentation Updates

• Update SESSION_CONTEXT.md with:
  • Work completed this session
  • Any new blockers discovered
  • Next steps for future sessions
• Update ROADMAP.md Active Sprint checkboxes if features completed
• Log any significant learnings in AI_REVIEW_LEARNINGS_LOG.md
• Archive completed/cancelled plans to docs/archive/completed-plans/
• Cross-document check: Review docs modified this session against DOCUMENT_DEPENDENCIES.md trigger matrix - update any dependent documents

3.1 Roadmap Sync Check (MANDATORY for Feature Work)

Session #69 Guardrail: Prevents roadmap staleness by ensuring feature work is reflected in ROADMAP.md.

If you implemented features, completed tasks, or made significant progress this session:

• Verify ROADMAP.md reflects current status
  • Check Active Sprint tasks - mark completed items with [x]
  • Check M1.6/M2 phase tasks - update status indicators
  • Update sprint progress percentage if significant work done
• No new features without roadmap entry
  • If you added a new feature not in ROADMAP.md, add it now
  • If you completed a planned feature, mark it done

Quick Check:

# See what code was changed this session
git diff --name-only HEAD~5

# If you see new components/features, verify they're in ROADMAP.md

Why This Matters:

• Roadmap drift causes planning confusion
• Future sessions won't know what's already implemented
• Session #69 found Sentry was 90% done but roadmap showed "Planned"

4. Learning Consolidation (AUTOMATIC - Session #69)

Consolidation now runs automatically during SessionStart when the threshold is reached (10+ reviews). No manual action required.

What happens automatically:

• When threshold is reached, npm run consolidation:run --auto runs
• Counter is reset in AI_REVIEW_LEARNINGS_LOG.md
• Patterns are analyzed and counted

Manual follow-up (optional, if you want to persist patterns):

npm run consolidation:check   # Check current status
npm run consolidation:run     # Preview what was consolidated
npm run patterns:suggest      # Find automatable patterns

If you want to manually add patterns to documentation:

• Add critical patterns to claude.md Section 4 (top 5 only)
• Add full patterns to docs/agent_docs/CODE_PATTERNS.md
• Add suggested patterns to check-pattern-compliance.js

Why this matters:

• claude.md is loaded at session START
• Patterns in claude.md will be in context for NEXT session
• This is how the AI "learns" from previous sessions

5. Code Review Completeness Audit

If you received code review feedback this session:

VERIFICATION CHECKLIST:
├─ Did you address ALL suggestions? (not just some)
├─ Did you test regex patterns for performance?
├─ Did you verify path-based filtering works correctly?
├─ Did you check for CI workflow impacts?
└─ Did you commit descriptive messages explaining WHY changes were made?

6. Automated Verification (RUN BEFORE MANUAL AUDIT)

Execute these automated checks to verify session compliance:

# 1. Verify expected skills were used based on activity
npm run skills:verify-usage

# 2. Check if any triggers are pending
npm run triggers:check

# 3. Review any overrides used this session
npm run override:list

# 4. View session activity summary
npm run session:summary

Interpreting Results:

• Skills verification failures indicate skills that should have been used but weren't. Either use the skill now, or document why it was skipped.
• Pending triggers should be resolved before push. If already resolved, document the resolution.
• Override history should be reviewed - ensure reasons were provided and are legitimate.
• Session summary shows what was done this session for documentation purposes.

7. Agent/Skill/MCP/Hook/Script Audit (MANDATORY)

Complete this audit for every session. If gaps found, document why or fix before ending.

7.1 Session Start Scripts

7.2 Agent Usage (based on work performed)

7.3 Skill Usage

7.4 MCP Servers

7.5 Hooks Executed

7.6 Audit Result

Overall Result: [ ] PASS [ ] FAIL

PASS criteria: All required items ran AND passed (or justified with explanation)

If FAIL, select disposition:

• Fixes applied this session (describe below)
• Documented for next session (note in SESSION_CONTEXT.md)

Remediation notes (if FAIL):

(describe what was invoked/run to address gaps, or next steps)

8. Key Learnings to Remember

Today's session reinforced these patterns:

DO:

• Read files before editing
• Use TodoWrite for multi-step tasks
• Check all code review items multiple times
• Add path-based filtering for context-specific patterns
• Use bounded regex ({0,N}?) instead of greedy (.*)
• Spread ESLint plugin configs in flat config format
• Exclude archive files (docs/archive/) from strict linting
• Add continue-on-error for pre-existing issues in CI

DON'T:

• Skip code review suggestions without documenting why
• Use greedy regex that can cause runaway matches
• Forget to test changes before committing
• Push without verifying CI impact
• Edit files without reading them first

9. Commit Summary

Provide a summary of all commits made this session:

git log --oneline -10

10. Update Session State (AUTOMATIC)

IMPORTANT: Run this command to update session state tracking:

npm run hooks:health -- --end

This updates the cross-session validation system so that:

• Next session-begin knows this session ended properly
• Session statistics are tracked
• No false warnings about incomplete sessions

Session complete. All work has been verified and documented.