Skip to main content

experience-to-skill

Use when deciding whether a reusable lesson or repeated workflow should become a skill, memory, AGENTS rule, script, project doc, or stay unrecorded.

Zur Installation springen

Quellinformationen

Repository
hideaway007/codex-self-improvement-skills
Letzte Quellaktivität
29. April 2026 um 06:10
Erkannte Sprache von SKILL.md
Englisch
Sterne
0
Forks
0

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
2 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
experience-to-skill
description
Use when deciding whether a reusable lesson or repeated workflow should become a skill, memory, AGENTS rule, script, project doc, or stay unrecorded.
# Experience To Skill ## Overview Turn useful experience into the right durable artifact. Not every lesson should become a skill: choose the smallest place that will make future work better without bloating global context. ## When to Use Use this skill: - At the end of a complex or repeated task. - When the same workflow has appeared about three times. - When a non-obvious debugging path, verification gate, or tool sequence saved time. - When the user asks to "make this reusable", "turn this into a skill", "update memory", "sync up", or similar. - When a future agent would likely repeat the same mistake without written guidance. Do not use it for one-off facts, project secrets, raw transcripts, or normal code comments. ## Artifact Decision Choose one primary destination: | Destination | Use when | Avoid when | |---|---|---| | Skill | Reusable cross-project workflow or judgment pattern | It is only true for one repo | | Project `AGENTS.md` | Project-specific agent rule, route, command, or boundary | It should apply globally | | Global `AGENTS.md` | Stable cross-project user preference or machine rule | It is domain/project-specific | | Memory | Lightweight preference or historical fact useful across sessions | It needs executable steps | | Script | Deterministic repeated operation | Human judgment is central | | Docs/README | Humans or downstream systems need to use the knowledge | Only agents need it | ## Workflow 1. Extract the reusable lesson. - State the trigger: when should a future agent remember this? - State the action: what should it do differently? - State the evidence: why is this worth preserving? 2. Check for an existing artifact. - Search current skills and project docs first. - Prefer updating an existing skill or rule over adding a near-duplicate. 3. Pick the narrowest durable home. - Global skills stay small and general. - Project-specific behavior belongs in the project. - Repeated mechanical operations should become scripts, not prose. 4. Write or update the artifact. - For skills, use `skill-creator` or `writing-skills`. - Keep frontmatter trigger-focused. - Keep body concise and procedural. - Do not include secrets, private tokens, or raw sensitive logs. 5. Validate. - Confirm paths exist. - Confirm the trigger description would cause the skill to load at the right moment. - Confirm the new artifact does not conflict with existing rules. ## Skill Creation Criteria Create or update a skill only when most of these are true: - The pattern is reusable across projects. - The task involves judgment or a workflow, not just a fact. - The behavior is not already covered by an installed skill. - The trigger can be described clearly in frontmatter. - The skill will prevent a real future mistake or reduce repeated setup work. ## Common Mistakes - Turning every lesson into a global skill. - Appending new rules without deleting or merging stale ones. - Storing project-specific domain rules globally. - Writing a narrative of what happened instead of a reusable procedure. - Treating generated files or green tests as proof that knowledge is correct.
Auf GitHub ansehen