speckit-memory

name	speckit-memory
description	Capture stable lessons from a speckit run into persistent project memory and create/update helper skills when repeated repo-specific workflows emerge.
compatibility	Requires spec-kit project structure with .specify/ directory
metadata	{"author":"automata-cli"}

User Input

$ARGUMENTS

Use any user-provided notes as additional context, but do not require them.

Goal

Persist reusable lessons from the current speckit-full execution so the next run starts with better repo-specific context and less friction.

Write durable memory to .specify/memory/speckit-memory.md. Create or update a helper skill under .agents/skills/ only when a repeated repo-specific workflow is stable enough to deserve its own instructions.

What Belongs In Memory

Record only information that is likely to help future runs:

1. Stable project defaults repeatedly chosen during autonomous execution
2. Repo-specific implementation patterns that beat generic defaults
3. Process friction and the reliable way around it
4. Reviewer or documentation expectations that should influence future runs
5. Reusable multi-step workflows that are better captured as helper skills

Do not record:

1. Feature-specific business requirements
2. Temporary branch names, PR URLs, or one-off timestamps
3. File-by-file implementation diaries
4. Incidents that are unlikely to recur
5. Anything contradicted by current code, AGENTS.md, or .specify/memory/constitution.md

Outline

1. Resolve the active feature with .specify/scripts/bash/check-prerequisites.sh --json --paths-only.

2. Read only the artifacts needed to extract durable lessons:

   • .specify/memory/constitution.md
   • AGENTS.md
   • Existing .specify/memory/speckit-memory.md if present
   • Active feature spec.md, plan.md, tasks.md
   • research.md, pr-report.md, and spec-decisions.md if present
   • The current diff or completed-task state when useful to confirm what actually worked

3. Extract candidate lessons and classify each one:

   • keep: stable and reusable
   • promote: stable enough for a helper skill
   • drop: too specific, stale, or noisy

4. If .specify/memory/speckit-memory.md does not exist, create it from .specify/templates/speckit-memory-template.md.

5. Update the memory file by section:

   • ## Autonomous Defaults
   • ## Implementation Patterns
   • ## Process Friction
   • ## Helper Skills

6. Use concise bullets in this format:

   - **<topic>**: <rule or preferred action>. Why: <short rationale>. Confirmed: <YYYY-MM-DD>.
   

7. De-duplicate semantically similar bullets, refresh the Confirmed date for surviving entries, and remove entries that are now contradicted.

8. Keep the file lean:

   • Prefer updating an existing bullet over adding a near-duplicate
   • Keep each section short and high-signal
   • If a section grows noisy, collapse repeated details into one stronger rule

Helper Skill Promotion Rule

Create or update a helper skill only if all of the following are true:

1. The workflow is likely to recur across multiple features
2. It contains at least three ordered steps, branching rules, or tool-specific gotchas
3. A short bullet in memory would be insufficient guidance
4. The workflow can be described without feature-specific product details

When promoting:

1. Create or update .agents/skills/<skill-name>/SKILL.md
2. Keep the helper skill narrow and procedural
3. Reference .specify/memory/speckit-memory.md only when needed; do not copy large memory sections into the skill
4. Add or refresh a matching bullet in the memory file's ## Helper Skills section

If a candidate workflow does not meet the bar, keep it as a memory bullet instead of creating a new skill.

Output

Report:

1. Memory file path
2. Bullets added, updated, or removed
3. Helper skills created or updated
4. Any candidate lessons intentionally discarded as too specific