Role

Librarian (on user request)

Trigger

USER MUST EXPLICITLY REQUEST - Not automatic

User triggers with phrases like:

• •"Archive this conversation"
• •"Document this session"
• •"Save this discussion about [topic]"
• •"/archive [optional-topic]"
• •"This is important, save it"

Purpose

Create permanent documentation from high-value conversations that might be missed by regular memory updates. Unlike automatic memory updates (which capture decisions), this archives complete conversations with full context.

When to Use

Use when:

• •User explicitly asks to save/archive a conversation
• •Complex decision-making process that should be preserved
• •Important design discussions
• •Knowledge transfer sessions
• •Architecture debates with valuable rationale
• •Debugging sessions with lessons learned

Do NOT use automatically - only on explicit user request.

Inputs

• •Current session conversation (all messages)
• •Optional topic name from user
• •Current memory files (for context)

Outputs

• •Transcript: docs/requirements/conversations/session-NNN-YYYY-MM-DD-topic.md
• •Summary: docs/requirements/summaries/session-NNN-summary.md
• •Verification report: docs/requirements/summaries/session-NNN-verification.md
• •Optional requirements: docs/requirements/official/REQ-[PREFIX]-NNN.md
• •Optional requirements verification: docs/requirements/official/REQ-[PREFIX]-verification.md
• •Updated index: docs/requirements/INDEX.md

Steps

Phase 1: Capture

1. •Confirm with user: "I'll archive this conversation about [topic]. Continue?"
2. •Read full session conversation (all messages since session start)
3. •Determine session number (increment highest existing)
4. •Create verbatim transcript in docs/requirements/conversations/
   • •Preserve exact wording
   • •Keep raw format (don't clean up tool markers)
   • •Include timestamps if available

Phase 2: Summarize

1. •

   Extract key elements:

   • •Decisions made
   • •Features discussed
   • •Insights/realizations
   • •Questions asked and answered
   • •Action items
   • •Open questions
2. •

   Create summary in docs/requirements/summaries/

   • •Concise but comprehensive
   • •Link to specific parts of transcript
   • •Highlight critical decisions

Phase 3: Verify Summary (Quality Loop)

1. •Cross-check summary vs transcript
2. •If gaps or conflicts found:
   • •Update summary
   • •Re-verify
3. •If ambiguity remains, ask user for clarification
4. •Create verification report documenting:
   • •What was checked
   • •Any corrections made
   • •Confidence level

Phase 4: Generate Requirements (Optional)

Only if conversation included formal requirements:

1. •Convert verified summary into formal requirements
2. •Use consistent prefix (e.g., REQ-CORE-001)
3. •Include traceability links
4. •Create requirements file in docs/requirements/official/

Phase 5: Verify Requirements (Optional)

Only if Phase 4 was executed:

1. •Cross-check requirements vs summary
2. •Ensure no omissions or scope creep
3. •Create requirements verification report

Phase 6: Finalize

1. •Update docs/requirements/INDEX.md
2. •Create docs/requirements/DOCUMENTATION_PROCESS.md if missing
3. •Report to user:
   • •What was archived
   • •Where files are located
   • •How to access them

Constraints

• •Scope: Current session only (unless user explicitly asks to backfill)
• •Verbatim: Must preserve exact user/agent wording
• •Single Prefix: All requirements in a module use one prefix
• •Traceability: Each requirement must link back to session summary
• •Idempotency: If session already exists, update in place (no duplicates)
• •User Control: User can skip phases (e.g., "just save the transcript")

Output Format

### 📄 Conversation Archived
**Topic**: [Topic Name]
**Session**: NNN
**Date**: YYYY-MM-DD

**Files Created**:
- Transcript: `docs/requirements/conversations/session-NNN-YYYY-MM-DD-topic.md`
- Summary: `docs/requirements/summaries/session-NNN-summary.md`
- Summary Verification: `docs/requirements/summaries/session-NNN-verification.md`
- Requirements: `docs/requirements/official/REQ-[PREFIX]-001-through-NNN.md` (if applicable)
- Requirements Verification: `docs/requirements/official/REQ-[PREFIX]-verification.md` (if applicable)
- Index Updated: `docs/requirements/INDEX.md`

**Key Decisions Captured**:
- [Decision 1]
- [Decision 2]

**Next Steps**:
- [Any follow-up actions]

Examples

Example 1: Simple Archive

User: "Archive this conversation about the auth system"

Agent: "I'll archive this conversation about the auth system. Creating transcript and summary..."

[Proceeds through all 6 phases]

Example 2: Just Transcript

User: "Just save the transcript of this debugging session"

Agent: "I'll save the transcript only. Skipping summary and requirements generation."

[Proceeds with Phase 1 only, confirms with user]

Example 3: Complete Documentation

User: "This architecture discussion is important, document everything"

Agent: "I'll create complete documentation including requirements. This will take a few minutes."

[Proceeds through all 6 phases, generates formal requirements from the discussion]

Integration with Regular Memory

This skill complements (not replaces) automatic memory updates:

• •Automatic memory updates: Capture decisions, constraints, patterns
• •Archive conversation: Capture full context, rationale, discussions

Both systems can reference each other via links.

File Structure

docs/requirements/
├── INDEX.md                          # Master index
├── DOCUMENTATION_PROCESS.md          # Process documentation
├── conversations/
│   └── session-001-2026-02-08-auth.md
├── summaries/
│   ├── session-001-summary.md
│   └── session-001-verification.md
└── official/
    ├── REQ-CORE-001.md
    └── REQ-CORE-verification.md