Process Session Notes Runbook

This runbook guides you through reviewing ephemeral session notes, extracting valuable knowledge, and cleaning up processed sessions.

Purpose

Session notes (.docent/sessions/*.md) capture detailed work logs from AI agent sessions - objectives, context, work performed, results, and lessons learned. This runbook helps:

1. Extract valuable insights into permanent documentation
2. Archive or delete processed ephemeral notes
3. Keep the sessions directory manageable
4. Prevent loss of important knowledge

Prerequisites

• .docent/sessions/ directory with session notes
• Write access to documentation directories
• Understanding of current project priorities

Procedure

Step 1: List Session Notes

Purpose: Identify all session notes to process

Commands:

# List all session notes sorted by modification time
ls -lt .docent/sessions/*.md

# Count total sessions
ls -1 .docent/sessions/*.md | wc -l

Review: Note which sessions are recent vs historical

---

Step 2: Review Each Session Note

Purpose: Read through sessions to identify valuable content

Action: For each session note, read and look for:

High-value knowledge:

• Architecture decisions → Should become ADRs
• Important discoveries → Should be in guides or architecture docs
• Problems solved → Should be in troubleshooting guides
• Lessons learned → Should be documented to prevent future issues
• Code reference updates → May need to be captured in changelogs

Medium-value knowledge:

• Implementation patterns discovered
• Configuration changes made
• Dependencies updated
• Tool usage insights

Low-value ephemeral content:

• Routine implementation steps
• Standard git operations
• Normal debugging process
• Obvious code changes

Decision Point: For each piece of knowledge, ask:

• Will this help future developers?
• Does this explain "why" something was done?
• Would I want to know this 6 months from now?
• Is this already documented elsewhere?

---

Step 3: Extract Architecture Decisions

Purpose: Capture significant technical decisions as ADRs

Action: For architecture decisions found in session "Lessons Learned" or "Context" sections

Use /docent:tell to create ADR:

/docent:tell create an ADR documenting [the decision]

Context from session:
- Decision: [what was decided]
- Rationale: [why it was decided]
- Date: [session date]
- Trade-offs: [from Lessons Learned section]

Location: .docent/adr/

Example Decisions to Extract:

• Technology choices (database, framework, library)
• API design decisions
• Data model changes
• Performance optimization approaches
• Security implementation choices

---

Step 4: Extract Implementation Insights

Purpose: Document useful patterns and discoveries

Action: For valuable implementation details from "Work Performed" or "Results" sections

Use /docent:tell:

/docent:tell add to [relevant guide]:

From [session-date]: [insight or pattern discovered]

Common locations:

• Implementation patterns → .docent/guides/
• API usage → .docent/guides/api-usage.md
• Tool configuration → .docent/guides/setup.md
• Troubleshooting solutions → .docent/guides/troubleshooting.md

Example Insights to Extract:

• "Config X requires Y to work properly"
• "The build fails if Z is not set"
• "Performance improves 50% with W approach"

---

Step 5: Update CHANGELOG and Documentation

Purpose: Ensure user-facing changes are documented

Action: Review "Results" section for changes that affect users or developers

Update locations:

• CHANGELOG.md - User-visible changes, new features, breaking changes
• README.md - New setup steps, changed requirements
• Architecture docs - Structural changes, new patterns

Commands:

# Check what's already in CHANGELOG
head -20 CHANGELOG.md

# Edit CHANGELOG if needed
# (Use Edit tool)

Example CHANGELOG entries:

## [Unreleased]

### Added
- New session processing runbook for knowledge management

### Changed
- Documentation structure consolidated from docs/ to .docent/

### Fixed
- Broken links in architecture documentation

---

Step 6: Capture Lessons Learned

Purpose: Prevent repeating mistakes and preserve insights

Action: Extract key points from "Lessons Learned" sections

Use /docent:tell:

/docent:tell document lesson learned:

[Copy lesson from session notes]

Common documentation targets:

• Project conventions → .docent/guides/conventions.md
• Common pitfalls → .docent/guides/troubleshooting.md
• Best practices → .docent/guides/best-practices.md
• Tool gotchas → .docent/guides/tooling.md

---

Step 7: Clean Up Session Notes

Purpose: Remove ephemeral content after extraction

Decision Point: What to do with processed sessions?

Option A: Archive by Year (Recommended)

Keep history organized but clearly mark as processed:

# Create archive structure
mkdir -p .docent/sessions/archive/$(date +%Y)

# Move processed sessions
mv .docent/sessions/2025-10-*.md .docent/sessions/archive/2025/

Benefits:

• Preserves complete history
• Easy to reference if needed
• Keeps active directory clean

Option B: Delete Ephemeral Sessions

Completely remove sessions after extraction:

# Review one more time
ls .docent/sessions/2025-10-*.md

# Delete after confirming extraction is complete
rm .docent/sessions/2025-10-*.md

Benefits:

• Cleaner repository
• Focuses on permanent docs only
• Matches gitignore patterns for ephemeral content

Recommended Approach:

• Archive sessions with significant decisions or complex work
• Delete sessions with only routine changes or already-documented work
• Keep recent sessions (last 7-14 days) for reference

---

Step 8: Standardize Session Naming

Purpose: Ensure consistent naming for easier processing

Action: Rename sessions to follow consistent pattern

Standard format: YYYY-MM-DD-session-NNN.md

Commands:

# Check current naming
ls .docent/sessions/

# Rename inconsistent files
mv .docent/sessions/session-2025-10-29.md \
   .docent/sessions/2025-10-29-session-001.md

Validation:

• All sessions follow date-first format
• Sessions on same day are numbered sequentially
• Easy to sort chronologically

---

Step 9: Update .gitignore if Needed

Purpose: Ensure ephemeral sessions aren't accidentally committed

Action: Check if sessions should be gitignored

Commands:

# Check current gitignore
grep -A2 -B2 session .gitignore

# Check what's currently tracked
git ls-files .docent/sessions/

Decision Point:

• Track sessions if they're valuable project history
• Ignore sessions if they're truly ephemeral agent state

If ignoring, add to .gitignore:

# Docent ephemeral session notes
.docent/sessions/
!.docent/sessions/archive/

Note: Current project tracks sessions, but you may prefer different approach

---

Step 10: Commit Documentation Updates

Purpose: Save extracted knowledge

Action: Create commit with changes

Commands:

# Check what changed
git status .docent/ CHANGELOG.md

# Stage changes
git add .docent/adr/ .docent/guides/ .docent/architecture/
git add CHANGELOG.md

# Create commit
git commit -m "docs: extract knowledge from session review

- Added ADRs for [decisions]
- Updated [guides] with implementation insights
- Documented [lessons learned]
- Cleaned up ephemeral session notes

Processed sessions: [date range]"

---

Validation & Verification

Success Criteria

How do you know processing succeeded?

• All valuable insights extracted to permanent docs
• Session notes archived or deleted appropriately
• Documentation is searchable via /docent:ask
• CHANGELOG updated if needed
• No broken references to deleted sessions
• Commit created with extracted knowledge

Health Checks

Post-processing validation:

# Verify sessions directory is clean
ls -la .docent/sessions/

# Test documentation search works
# (Use /docent:ask to find extracted content)

# Check git status
git status

---

Cadence

Recommended frequency:

• After each significant session - Extract while context is fresh
• Weekly - Process last week's sessions in batch
• Before releases - Ensure all valuable knowledge is documented
• Monthly - Deep review and aggressive cleanup

Time estimates:

• Quick session (routine work): 5 minutes
• Medium session (new features): 15 minutes
• Complex session (architecture changes): 30-45 minutes

---

Troubleshooting

Common Issues

Issue: Session contains valuable content but unclear where it belongs

Solution:

1. Use /docent:ask where should documentation about [topic] go?
2. Create new guide if no appropriate location exists
3. When in doubt, extract to .docent/notes/ and decide later

---

Issue: Session describes decision but lacks full context

Solution:

1. Search related code to understand context better
2. Check git history for related commits
3. Use /docent:tell to document what you can reconstruct
4. Note in ADR that details were from session reconstruction

---

Issue: Multiple sessions cover same topic

Solution:

1. Read all related sessions together
2. Extract consolidated knowledge (not duplicate content)
3. Create single comprehensive doc covering the topic
4. Archive or delete all related sessions together

---

Issue: Unsure if session content is still relevant

Solution:

1. Check if code/files referenced still exist
2. Verify approaches described are still in use
3. When in doubt, extract with date context: "As of [date], ..."
4. Mark potentially outdated content for future review

---

Tips

Efficient Processing:

• Process sessions immediately after completion (context is fresh)
• Use /docent:tell extensively to speed up extraction
• Focus on "why" and "lessons learned" - code shows "what"
• Create templates for common extraction patterns

Knowledge Triage:

• High value: Architecture decisions, non-obvious solutions, project context
• Medium value: Implementation patterns, tool discoveries, configuration gotchas
• Low value: Routine git operations, standard debugging, obvious code changes

Avoid Over-Documenting:

• Not every session needs extraction
• Routine implementation details don't need permanent docs
• Code reviews and simple refactors are often self-documenting

Session Writing Tips (for future):

• Write "Lessons Learned" with extraction in mind
• Make "Context" section rich for future reference
• Note "why" decisions were made, not just "what" was done
• Flag high-value insights as you work

---

Notes & Warnings

💡 Tip: Process sessions while memory is fresh - much faster than processing old sessions

⚠️ Warning: Don't delete sessions until after extraction is complete and committed

📝 Note: Sessions are intentionally more detailed than journals - expect significant distillation during extraction

---

References

• Process Journals Runbook - Similar workflow for journal entries
• Bootstrap Runbook - Initial docent setup
• Health Check Runbook - Project health validation

---

Change Log

2025-10-29: Initial version