Append structured episodes to a JSONL memory store and classify lessons for promotion.
npx skills add https://github.com/yiwei79/root-azoth --skill remember-yiwei79
Please help me install this Agent Skill. Skill: remember Source: https://github.com/yiwei79/root-azoth/tree/main/.opencode/skills/remember Command: npx skills add https://github.com/yiwei79/root-azoth --skill remember-yiwei79
.azoth/memory/episodes.jsonl so knowledge persists across sessions. ## Core Features & Use Cases - Episode Capture: Append typed episodes (success, failure, decision, pattern, observation) with lessons, tags, and context to an append-only JSONL store. - Auto-Classification: Classify each episode by scope, reuse potential, and maturity to recommend a future home without triggering automatic promotion. - Promotion Proposals: Surface patterns reinforced across 2+ episodes as human-approved promotion proposals for skills, agents, or pattern files. - Use Case: At the end of a coding session where a TDD approach caught three edge cases, capture a success episode with specific lessons and tags so future sessions can surface and reinforce it. ## Quick Start Ask the agent to remember what was learned in this session by capturing an episode with the key lessons and tags.High-intent search queries and answers about installing and using this skill.
Append a structured episode to `.azoth/memory/episodes.jsonl` with a type (success, failure, decision, pattern, or observation), a summary, specific actionable lessons, and tags. Capture at least one episode per session during the closeout phase.
Each episode includes a UUID, timestamp, session ID, type, goal, summary, lessons array, tags, reinforcement count, and context object with phase, pipeline, and files changed. Episodes are append-only and never edited or deleted.
Create a new episode linked to the older one via `related_episode_id` or `supersedes_episode_id`, and tag it as stale, superseded, or contradicted. Never rewrite or delete the original episode; corrections happen by addition.
Propose promotion only when a pattern is reinforced across 2 or more episodes, is generic rather than repo-local, and is objective rather than personal preference. Promotion always requires explicit human approval.
Skip promotion for single-event lessons, highly repo-local or temporary behaviors, personal preferences, contradictory evidence, or unstable tags. Keep the episode in M3 memory and wait for reinforcement across sessions.
Soria-IncAnalyze pipeline session artifacts to propose ETHOS.md and skill changes.
changfengproCaptures skill experiences into multi-memory artifacts and proposes validated skill updates.
yiwei79Retrieves and ranks relevant memory episodes and patterns by goal tags before planning.
brucebanner010198-commitsAppend project-close lessons to the cross-project LESSONS.md ledger.