ratchet-memory

Curates MEMORY.md by merging duplicates, pruning obsolete entries, and verifying issue and PR links.

2|Updated Jun 13, 2026
One-click install
npx skills add https://github.com/praveenvijayan/Ratchet --skill ratchet-memory-praveenvijayan
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ratchet-memory
Source: https://github.com/praveenvijayan/Ratchet/tree/main/plugin/skills/ratchet-memory
Command: npx skills add https://github.com/praveenvijayan/Ratchet --skill ratchet-memory-praveenvijayan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long-running agent-driven projects accumulate a memory file that grows stale and bloated over time. This Skill keeps memory/MEMORY.md a small, current cache by deduplicating entries, removing obsolete facts, and checking that every cited issue or PR still resolves. ## Core Features & Use Cases - Link Verification: Checks each entry citing an issue or PR with gh issue view / gh pr view and flags dead or wrong references without deleting them. - Deduplication and Pruning: Merges entries that say the same thing and removes superseded decisions, fixed gotchas, and retired environment facts, while marking uncertain entries with (verify?) instead of guessing. - Human-Reviewed Edits: Edits the file, reports entries kept, merged, pruned, and flagged, then stops without committing so a human reviews the diff. - Use Case: Run quarterly or when MEMORY.md grows past roughly 300 lines to compact a project's accumulated agent memory before it degrades context quality. ## Quick Start Ask the agent to compact memory/MEMORY.md by merging duplicates, pruning stale entries, and verifying all linked issues and PRs, then review the resulting diff before committing.

Frequently Asked Questions about ratchet-memory

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I compact a growing agent memory file?

Run the memory compaction workflow on memory/MEMORY.md: it merges duplicate entries, prunes obsolete facts, tightens each surviving entry to one or two lines with a source citation, and reports what changed. The edit is left uncommitted for human review.

How do I verify issue and PR links in a markdown memory file?

Each entry citing #N or PR #N is checked with gh issue view N or gh pr view N using the GitHub CLI. Dead or wrong references are flagged in the report rather than deleted, so no information is lost silently.

When should I run memory compaction on MEMORY.md?

Run it periodically, such as quarterly, or whenever MEMORY.md grows past roughly 300 lines. Raw history remains in closed issues, PRs, and git, so pruning the cache never loses underlying information.

Does memory compaction delete entries it is unsure about?

No. Entries whose truth is uncertain are kept and marked with a (verify?) flag instead of being removed. Only entries that are clearly superseded, fixed, or retired are pruned.

Can the skill edit or commit memory files automatically?

It edits memory/MEMORY.md but never touches the human-owned memory/USER.md, and it never commits or pushes. The diff is left for human review exactly like any other memory change.