Persists evidence-backed project lessons and facts into canonical memory locations.
npx skills add https://github.com/liuzhengdongfortest/easysdd --skill cs-keep
Please help me install this Agent Skill. Skill: cs-keep Source: https://github.com/liuzhengdongfortest/easysdd/tree/main/plugins/codestable/skills/cs-keep Command: npx skills add https://github.com/liuzhengdongfortest/easysdd --skill cs-keep
Valuable lessons learned during AI-assisted coding sessions are usually lost when the session ends, forcing teams to rediscover the same pitfalls repeatedly. This Skill captures evidence-backed findings and routes each one to its single canonical owner so knowledge survives across sessions.
Ask the agent to remember the lesson or pitfall you just discovered in this session, and it will verify the evidence and store it in the appropriate project memory location.
High-intent search queries and answers about installing and using this skill.
Tell the agent to remember the finding, and cs-keep verifies the evidence, checks for duplicates, and writes a dated lesson file under .codestable/lessons/ with scope, applicability boundaries, and up to three evidence pointers.
It routes by single canonical owner: mechanical guards like tests or linters first, then attention.md for high-frequency facts, ADRs for hard-to-reverse structural decisions, existing project docs for stable methods, and lesson files as staging for everything else.
No. Writing requires explicit authorization such as a user asking to remember something or approving a crystallization candidate. New rules, scope changes, promotions, deletions, and cross-project sharing all need separate explicit consent.
Search with grep -ri over .codestable/lessons/ and .codestable/compound/ using relevant keywords. Before applying a lesson, verify it against current code, tests, or canonical docs, and never apply retired entries.
Lessons start as observed and can only become validated when an independent later task applies them successfully; a lesson cannot validate itself. Retired lessons are never revived—new conclusions require a fresh observed lesson.
jasonwilmotCoordinate a structured debugging workflow to reproduce and resolve software bugs.
TeiNamRecord one-line lessons from repeated corrections into a single file.
AlexZio00Load handoff context and generate a ready signal with prioritized actions.
HemantSudarshanCapture and organize architectural decisions with ADR templates and lifecycle tracking.
