learn

Records verified project learnings as minimal rules in repository instruction files.

3|Updated Nov 8, 2014
One-click install
npx skills add https://github.com/mintuz/.dotfiles --skill learn-mintuz
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: learn
Source: https://github.com/mintuz/.dotfiles/tree/main/agents/.agents/skills/learn
Command: npx skills add https://github.com/mintuz/.dotfiles --skill learn-mintuz

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams repeatedly rediscover the same bugs, gotchas, and architectural decisions because hard-won knowledge never gets written down. This Skill captures durable project learnings into authoritative instruction files like AGENTS.md or CLAUDE.md so future work benefits from past experience. ## Core Features & Use Cases - Authority Detection: Locates the correct instruction file by following the repository's instruction hierarchy, and identifies overlapping existing rules to amend rather than duplicate. - Verification Gate: Keeps only verified, reusable learnings that prevent recurring bugs, record non-obvious invariants, or preserve architectural rationale; rejects hypotheses until confirming evidence is supplied. - Safety Filtering: Strips credentials, personal data, internal hostnames, and usable security-bypass instructions before anything is written. - Minimal Patching: Produces the smallest possible edit matching the target file's voice, with outcomes limited to no change, a proposed patch, or one applied patch. - Use Case: After debugging a recurring CI failure caused by a misconfigured cache key, ask the agent to record the verified fix as a one-line rule in AGENTS.md so the mistake never recurs. ## Quick Start Ask the agent to preserve the learning from the incident we just fixed as a rule in AGENTS.md.

Frequently Asked Questions about learn

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

FAQPage Schema
How do I save a project learning to AGENTS.md or CLAUDE.md?▼

Invoke the learn skill after an incident or decision, and it locates the authoritative instruction file, checks for overlapping rules, and writes the smallest safe rule. It can propose the patch first or apply it directly if you authorize the edit.

What kinds of learnings should be recorded in repository instructions?▼

Record verified learnings that prevent recurring bugs or security failures, capture non-obvious invariants or constraints, preserve architectural rationale, or save meaningful investigation time. Trivial, one-off, or unverified details are rejected with a named reason.

Does the skill write secrets or credentials into instruction files?▼

No, the skill never writes credentials, personal or customer data, or internal hostnames into instruction files. It names the owning secret or environment manager instead, and it documents security bypasses only by name and status, never with usable values.

What happens if a learning is based on an unverified hypothesis?▼

The skill records no rule that depends on an unverified cause or repair. It names the unverified fact, identifies the artefact that would confirm it, such as a log or metric, and asks you to supply that evidence first.

When should I not use this skill for documentation?▼

Do not use it for trivial or one-off details, incident chronologies, or content already covered by an existing rule. It is designed for durable, reusable rules in instruction files, not general documentation or retrospectives.