agentleFS
LuisUrrutiaLuisUrrutia/skills/skill-maintenance/SKILL.md

skill-maintenance

Skill2 starsChanged 15 days agoMITView on GitHub

Use when creating or maintaining a skill's origin.toml to record its goal, sources and design decisions, or when checking or updating a skill from its recorded sources or dependencies.

We split this skill into the rules it gives an agent

9 rules. 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Keep each skill's origin.toml true to the skill: what the skill is for, the decisions that shape it, and the outside material it builds on.

    line 8
  2. Distinctive:

    Read references/origin-format.md before writing a record.

    line 14
  3. Very specific:

    Read the skill, its record, and the change with the reason the user gave.

    line 33
  4. Distinctive:

    Write each decision once, as the entry a future editor needs:

    line 40
  5. Very specific:

    Update or remove the entries the change makes false, including where locations of renamed headings and moved files.

    line 52
  6. Very specific:

    When the change adds, alters or removes an entry of a skill with approach base, judge whether the base still fits.

    line 56
  7. Distinctive:

    When the change alters the skill's name, existence or authority, or what callers pass to it or receive from it, list its dependents and …

    line 60
  8. Distinctive:

    Read references/updates.md.

    line 78
  9. Distinctive:

    After any edit to a record, and after headings or files move in a skill that has one, run the checker described in Check the record.

    line 85

“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill maintenance

Keep each skill's origin.toml true to the skill: what the skill is for, the decisions that shape it, and the outside material it builds on. The record lets a later edit or update keep those decisions instead of guessing them. agent-instructions owns writing and evaluating the skill's text; this skill owns its record and its updates from recorded sources.

Read references/origin-format.md before writing a record. Then choose the task from the request:

  • Record: a skill is created, or a change adds, removes or alters a rule, an adopted idea, a deviation from its base, or a dependency. agent-instructions hands these changes over; a user can also ask directly.
  • Dependents: "which skills depend on this one" lists each record whose dependencies name it, as List dependents describes.
  • Check: "check this skill for updates" reports candidate changes from its sources and leaves the skill and its record unchanged.
  • Update: "update this skill from its sources" incorporates compatible changes and records the review.

Using a skill never fetches its sources, and none of these tasks creates a schedule.

Record decisions

  1. Read the skill, its record, and the change with the reason the user gave. A skill without a record gets one when the user asks or when another skill in its repository has an origin.toml; skills installed in the host outside that repository do not count. The new record states the approach, a goal that says what the skill produces, for whom and what success looks like, the invariants its text already enforces, and the dependencies it calls.
  2. Write each decision once, as the entry a future editor needs:
    • A rule the user wants kept, such as a scope boundary, an authority limit or a host constraint: an invariant.
    • A difference from the base, including an added rule, whether or not an invariant also states it: a deviation.
    • A concept taken from a source: an idea. A donor behavior left out that a later update could propose again: a decline. Material from a source the record does not list yet first gets its own source or snapshot entry.
    • A skill or tool it calls or hands work to at runtime: a dependency.
  3. Describe the decision, not the edit or how it was reached. A reader who never saw the conversation, the source or the diff must understand what the skill does and why; state the rule rather than copy the skill's text.
  4. Update or remove the entries the change makes false, including where locations of renamed headings and moved files. Remove an invariant or a decline only when the user's request or decision removes it; when only its reason seems gone, report that and ask.
  5. When the change adds, alters or removes an entry of a skill with approach base, judge whether the base still fits; when it does not, keep the approach, report the evidence and ask the user.
  6. When the change alters the skill's name, existence or authority, or what callers pass to it or receive from it, list its dependents and report each whose recorded role or decisions rely on the old contract. Report the full list, with those entries marked. Changing their text, other callers or their records needs a request that covers it; agent-instructions then checks and updates the text.

Take commits, hashes and dates from the repository or file, never from memory. Record only reasons the user stated or the evidence shows, credited and cited as Write the entries describes. An invariant without a known reason leaves reason out; a deviation or decline requires one, so ask the user when neither the request nor the evidence supplies it. The goal changes only on the user's statement, because every later decision is judged against it. Wording fixes and corrections that keep the meaning need no entry; Git history keeps them.

Check or update from sources

Read references/updates.md. It covers both modes, the review range of each source, installed dependencies, and when the review cursor may advance. A partial run reports what stayed unresolved; it never claims the skill is fully synchronized.

Validate the record

After any edit to a record, and after headings or files move in a skill that has one, run the checker described in Check the record. Fix every diagnostic. Then reread each new or changed entry as a newcomer to the source and the conversation would.

Finish when the record matches the skill and the checker passes, or report the diagnostics that remain. Commits and publication follow the active repository rules and user authorization.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.