Archiving Completed Plans

Overview

Archive the active project in .writing/ into a structured archive directory, consolidate and polish project memory based on current repo facts, and reset .writing/ for the next task.

Core principle: The target is the active project (design, plan, findings, progress), not "the session."

Announce at start: "I'm using the archiving skill to archive this project and consolidate memory."

Prerequisite: .writing/progress.md and .writing/findings.md must exist with content beyond the empty template.

The Process

Step 1: Determine Archive Name

1. Read .writing/progress.md — extract task names from the Task Status Dashboard
2. If .writing/plan.md exists, derive the feature name from it
3. Otherwise, derive a short name from the main task descriptions
4. Use AskUserQuestion to let the user confirm or modify the archive name

Step 2: Generate Archive Summary

Read all active project files:

• .writing/design.md (if exists)
• .writing/plan.md (if exists)
• .writing/progress.md
• .writing/findings.md
• .writing/agents/*/findings.md (if any exist)

Generate a concise summary.md with this format:

# Archive: <name>
**Date:** YYYY-MM-DD

## Summary
<!-- 2-3 sentences: what was done, what was the outcome -->

## Key Decisions
<!-- From findings.md Technical Decisions table — only decisions with long-term value -->

## Lessons Learned
<!-- Patterns discovered, gotchas, things to remember for similar future work -->

## Key Files Changed
<!-- Important files that were created or significantly modified -->

Keep it concise. The summary is a quick-reference document. Aim for 30-60 lines. The full design, plan, findings, and progress are preserved alongside it.

Step 3: Save Archive

1. Generate a unique archive directory name:

mkdir -p .writing/archive
${CLAUDE_PLUGIN_ROOT}/scripts/unique-filename.sh .writing/archive "<name>" ""

1. Create the archive directory and save all active project files:

mkdir -p "<returned-path>"
${CLAUDE_PLUGIN_ROOT}/scripts/snapshot-save.sh "<returned-path>"

1. Write summary.md to the archive directory
2. Report: "Archive saved to .writing/archive/<name>/"

Step 4: Memory Consolidation & Polish

This step performs a fact-based memory maintenance pass — not just adding new findings, but optimizing existing memory against current repo state.

4a. Explore current facts

• Locate the current project's memory directory: find the auto-memory path that matches this project's working directory under ~/.claude/projects/. Read MEMORY.md and any topic files in that directory only — do NOT glob across all projects.
• Run git diff --stat to see recent changes
• Quick Glob/Read of key repo files to verify paths and patterns mentioned in memory are still accurate

4b. Extract new findings

• From .writing/findings.md, identify items worth long-term retention:

- Technical decisions with lasting rationale - Debugging lessons and root causes - Architectural patterns discovered - Dependency constraints or compatibility notes

• Filter OUT session-specific items:

- Task completion status - Temporary workarounds - Intermediate test results

4c. Generate unified optimization proposal

Present to user in this format:

Memory Optimization Suggestions:

New items (from this session's findings)
  - [new finding 1]
  - [new finding 2]

Compress (existing memory that can be condensed)
  - [memory item X] -> [compressed version]

Update (inconsistent with current repo facts)
  - [outdated memory Y] -> [corrected version]

Remove (no longer applicable)
  - [obsolete memory Z] — reason: [why]

If there are no suggestions in a category, omit that category entirely.

4d. User confirmation

• Use AskUserQuestion to present the suggestions
• User can approve all, or describe which to skip

4e. Execute writes

• Apply approved changes to memory files using Edit/Write tools
• Ensure MEMORY.md stays under 200 lines — move detailed content to topic files
• For topic files that don't exist yet, create them and link from MEMORY.md

Step 5: Reset .writing/

${CLAUDE_PLUGIN_ROOT}/scripts/writing-reset.sh

This removes design.md, plan.md, progress.md, findings.md, and agents/, then recreates clean templates from canonical sources. archive/ and stash/ are preserved automatically.

Step 6: Report Completion

Display a concise completion summary:

Archive complete:
- Archive: .writing/archive/<name>/
- Memory: <N> items added, <N> compressed, <N> updated, <N> removed
- .writing/ reset to clean state

Edge Cases

Empty .writing/: Run ${CLAUDEPLUGINROOT}/scripts/check-writing-state.sh — if it returns empty, warn the user and ask if they still want to archive. An empty archive has no value.

No memory changes needed: If there are no new findings worth persisting and existing memory is already accurate, skip Step 4c-4e entirely. Report "Memory already up to date."

Multiple sessions in one .writing/: If progress.md shows multiple session headers, include all of them in the archive summary.

Legacy archives: Existing single-file archives in .writing/archive/*.md remain valid. The archive listing and historical archive checks should handle both formats.

Key Principles

• Fact-based: All memory polishing grounded in current repo state, not assumptions
• Semi-automatic: LLM proposes, user confirms. Never write memory without approval
• Non-destructive: Archive is saved BEFORE any cleanup. Archive survives reset
• Complete: Archive preserves the entire project (design, plan, findings, progress) as a unit
• Concise: summary.md is a quick-reference doc, not a full log