version: "1.0.1" name: session-begin description: Complete verification steps before starting any work session

Session Begin Checklist

⚠️ IMPORTANT - Duplicate Detection:

Before proceeding with the full checklist, check if this session was already started:

1. Read the current conversation context - Have I already completed this

checklist in the current conversation?

1. Check SESSION_CONTEXT.md timestamp - Was "Last Updated" modified today?

• Note: Field stores date only (YYYY-MM-DD), not time. Sub-day duplicate

detection relies on conversation context check (#1) and session counter check (#3).

1. Check session counter - Did I already increment the session counter

earlier in this conversation?

If ANY of these are true:

• ✅ Session is already active
• ⚠️ DO NOT re-run the checklist
• ⚠️ DO NOT re-increment the session counter
• ⚠️ DO NOT re-run startup scripts
• 💬 Example response: "Session #35 already active (started earlier in this

conversation). Checklist completed earlier. What would you like to work on?"

If ALL are false:

• ✅ This is a new session
• ✅ Proceed with full checklist below

Before starting any work, complete these verification steps:

0. Secrets Decryption Check (REMOTE SESSIONS)

Check if MCP tokens need decrypting:

# Check secrets status
if [ -f ".env.local.encrypted" ] && [ ! -f ".env.local" ]; then
  echo "⚠️ Encrypted secrets found but not decrypted"
fi

If secrets need decrypting:

1. Ask the user for their passphrase - Example: "Your MCP tokens need

decrypting. What's your passphrase?"

1. Run the decrypt command using stdin (avoids shell history exposure):

``bash echo "<user_passphrase>" | node scripts/secrets/decrypt-secrets.js --stdin ``

1. Verify success - Check that .env.local now exists with tokens
2. Never store or log the passphrase - Only use it for the decrypt command

Security note: Using --stdin with echo pipe is safer than env vars, which can leak to shell history and process listings.

If secrets are already decrypted or no encrypted file exists:

• Skip this step and continue to Context Loading

0b. Cross-Session Validation (AUTOMATIC)

The SessionStart hook automatically checks if the previous session ended properly.

If you see a "Cross-Session Warning":

1. The previous session started but /session-end was not run
2. Consider running the missed session-end checklist items:

• Update SESSION_CONTEXT.md with progress from the incomplete session
• Check for uncommitted changes
• Run npm run hooks:health to see session statistics

Quick remediation:

# See session state
npm run hooks:health
 
# Check for uncommitted work from previous session
git status
git log --oneline -5

0c. Episodic Memory Search (RECOMMENDED)

Search past conversations for relevant context using episodic memory. This helps recover decisions, solutions, and patterns from previous sessions.

Use `mcp__plugin_episodic-memory_episodic-memory__search` with these queries:

// Search for context on current work
search({ query: ["current branch/feature name", "decisions"] });
// Search for past errors if debugging
search({ query: "error message or pattern" });
// Search for established patterns
search({ query: ["component/module name", "patterns"] });

When to search:

Tips:

• Single string = semantic search (fuzzy, meaning-based)
• Array of 2-5 terms = AND search (all terms must match)
• Use limit: 5 for focused results, limit: 20 for broader search
• Current conversation is NOT indexed yet (only previous sessions)

Summarize findings for the user if relevant context is found.

1. Context Loading (MANDATORY)

• [ ] Read SESSION_CONTEXT.md - Current status,

active blockers, next goals

• [ ] Increment session counter in

SESSION_CONTEXT.md

• [ ] Check ROADMAP.md for priority changes

1b. Session Gap Detection (AUTOMATIC - Session #138)

The npm run session:gaps script (run in Section 7) checks for undocumented sessions by comparing commit-log.jsonl against SESSION_CONTEXT.md.

If gaps are detected:

1. Run npm run session:gaps:fix to generate suggested session summaries
2. Review the suggestions and add them to SESSION_CONTEXT.md
3. Update MEMORY.md if any stale entries need correction

How the system works (3 layers):

• Layer A (commit-tracker.js): PostToolUse: Bash hook auto-logs every commit

to .claude/state/commit-log.jsonl with session number, files, timestamp

• Layer B (compaction-handoff.js): Enhanced handoff includes task states and

recent commits when context is getting large

• Layer D (check-session-gaps.js): Session-begin detects gaps between commit

log and documented sessions

This system prevents state drift caused by context compaction interrupting session-end updates.

1c. Stale Documentation Check (MANDATORY)

Documentation often drifts from reality. Before trusting any status in docs, verify against actual commits:

# Check recent commits to see actual work done
git log --oneline -30
 
# Check commits since last documented session date
git log --oneline --since="YYYY-MM-DD"

Compare commits against documented status:

1. Look for PR/feature commits (e.g., "feat:", "fix:", "refactor:")
2. Cross-reference with ROADMAP.md Active Sprint checkboxes
3. If commits show work done but docs show incomplete → UPDATE THE DOCS

Common discrepancies to check:

• Sprint track items: Check commits against Active Sprint checkboxes in

ROADMAP.md

• Session counter: Check AI_REVIEW_LEARNINGS_LOG.md version history for session

numbers

• Test counts: Run npm test to verify actual vs documented

If docs are stale:

1. Update the stale document with correct status
2. Note which sessions failed to update docs
3. Commit the corrections before proceeding

2. Consolidation Status Check

Check AI_REVIEW_LEARNINGS_LOG.md for the "Consolidation Trigger" section:

• If "Reviews since last consolidation" >= 10: ⚠️ CONSOLIDATION WAS MISSED
• This means patterns from previous reviews are NOT in claude.md context
• Previous session should have consolidated but didn't

If consolidation was missed:

1. Note this in your session summary
2. The patterns are still available in AI_REVIEW_LEARNINGS_LOG.md (read if

needed)

1. Consolidation will happen at THIS session's end

3. Documentation & Planning Awareness

• [ ] Check ROADMAP.md Active Sprint section for current

work

• [ ] Note: Archive files in docs/archive/ are excluded from linting
• [ ] Completed plans are archived to docs/archive/completed-plans/
• [ ] Reference: INTEGRATED_IMPROVEMENT_PLAN.md is ✅ COMPLETE (archived

2026-01-14)

4. Skill Selection (BEFORE starting work)

DECISION TREE:
├─ New project/domain? → Use '/find-skills' to discover capabilities
├─ Bug/Error? → Use 'systematic-debugging' skill FIRST
├─ Writing code? → Use 'code-reviewer' agent AFTER completion
├─ Security work? → Use 'security-auditor' agent
├─ UI/Frontend? → Use 'frontend-design' skill
├─ Complex task? → Check available skills with /skills
└─ Multi-step task? → Use TodoWrite to track progress

5. Code Review Handling Procedures

When receiving code review feedback (CodeRabbit, Qodo, etc.):

1. Analyze ALL suggestions - Read through every comment multiple times
2. Create TodoWrite checklist - Track each suggestion as a task
3. Address systematically - Don't skip items; mark as resolved or note why

skipped

1. Verify CI impact - Check if changes affect workflows (ci.yml,

docs-lint.yml)

1. Test after changes - Run npm test and npm run lint before committing

6. Anti-Pattern Awareness

Before writing code, scan claude.md Section 4 "Critical Anti-Patterns" and CODE_PATTERNS.md Quick Reference section (🔴 = critical patterns). Key patterns:

• Read before edit - Always read files before attempting to edit
• Regex performance - Avoid greedy .* in patterns; use bounded

[\s\S]{0,N}?

• ESLint flat config - Spread plugin configs, don't use directly
• Path-based filtering - Add pathFilter for directory-specific patterns
• Archive exclusions - Historical docs should be excluded from strict

linting

6b. Velocity & Task Dependencies (RECOMMENDED)

Show recent velocity and available tasks:

# Show velocity summary (if data exists)
node scripts/velocity/generate-report.js 2>/dev/null || true
 
# Show dependency-resolved available tasks
node scripts/tasks/resolve-dependencies.js 2>/dev/null || true

Share the output with the user at session start so they can see:

• Recent velocity trend and sprint burn-down
• Which tasks are unblocked and ready to work on

7. Session Start Scripts (AUTO-RUN)

Execute these scripts automatically when processing this command:

# Surface known anti-patterns (errors should be visible, not suppressed)
npm run patterns:check
 
# Check if multi-AI review thresholds reached
npm run review:check
 
# Surface past lessons relevant to current work
npm run lessons:surface
 
# Check for undocumented sessions (Layer D - compaction gap detector)
npm run session:gaps

Important: These scripts are required. If any script fails:

1. Note the error in session summary
2. Investigate if it's a real issue vs missing script
3. If script missing, note it as "N/A" in audit

Record results in session audit - these must be marked as "Ran" or "Failed (reason)" in /session-end audit.

8. Technical Debt Awareness (NEW - Session #98)

Check current technical debt status:

• [ ] Read Technical Debt INDEX for

prioritized tech debt

• [ ] Note any S0/S1 items that should be addressed this session
• [ ] Check if any items from previous session need updating

Key tracking documents:

After resolving tech debt items:

1. Mark item as resolved in docs/technical-debt/MASTER_DEBT.jsonl
2. Update ROADMAP.md if item was in a sprint track
3. Note in session summary

9. Cross-Document Dependency Check

Before starting work, verify cross-document consistency:

# Check cross-document dependencies
npm run crossdoc:check

Key dependencies to verify:

• ROADMAP.md ↔ SESSION_CONTEXT.md (priorities match)
• MASTER_DEBT.jsonl ↔ ROADMAP.md (tech debt section current)
• Audit findings ↔ MASTER_DEBT.jsonl (new findings consolidated)

See: DOCUMENT_DEPENDENCIES.md for full dependency matrix.

10. Incident Documentation Reminder

After encountering any significant errors or issues:

• Document the issue in

AI_REVIEW_LEARNINGS_LOG.md

• Use the standard "Review #XX" format
• Include: cause, fix, pattern identified, prevention steps
• This builds institutional knowledge for future sessions

Ready to begin session. What would you like to work on?