agentleFS
puedingpueding/houserules/skills/fit/SKILL.md

fit

Skill1 starsChanged 3 months agoMITView on GitHub

Re-fit CLAUDE.md in both directions at once — to the model that reads it and to how you actually work — reconciling the two into a single numbered table. Use when the user runs /houserules:fit, or asks to run migrate and learn together, do a full CLAUDE.md pass, or check everything at once. Prefer this over running the two separately when both are wanted, because it reconciles their conflicts and writes in one safe pass.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. ontology-of-everything/handy-skills★ 2same name
    2. hailneed/painradar★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository3 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Follow the existing skills — do not reimplement their logic:

    line 23
  2. Very specific:

    Run them in either order; they only read.

    line 33
  3. Distinctive:

    Every finding from both sides lands in exactly one bucket.

    line 38
  4. Distinctive:

    The session record wins: Keep the rule.

    line 61
  5. Distinctive:

    Never resolve a conflict silently.

    line 74
  6. Distinctive:

    If both sides touch the same lines, produce a single replacement satisfying both.

    line 84
  7. Distinctive:

    Only the numbers named.

    line 129
  8. Very specific:

    Do not reimplement migrate or learn: Follow their phases 1-4.

    line 149

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

houserules:fit

Runs both analyses and merges them into one table.

This is not a wrapper that calls two commands in sequence. Two things make it a distinct skill, and both are correctness issues rather than conveniences:

  1. The two analyses disagree, and the disagreements are the most informative rows in the table. Presented as two separate reports the user has to reconcile them by hand, which is exactly the judgment this skill owes them.
  2. Two sequential write passes corrupt line numbers. migrate cites L89-113 against the file as it exists now. Let learn write first and every one of those citations is silently wrong. There is exactly one write pass.

Run the two analyses

Follow the existing skills — do not reimplement their logic:

  • skills/migrate/SKILL.md phases 1-4 — resolve target, identify the model, fetch the live guide, categorise (A-H).
  • skills/learn/SKILL.md phases 1-4 — extract typed prompts, cluster, evidence gate, check against the existing file.

Stop both before their Phase 5. Neither prints its own table. Collect their findings and continue here.

Run them in either order; they only read. If migrate's live fetch fails, say so and continue — a degraded half is still worth reporting, clearly marked.

Reconcile

Every finding from both sides lands in exactly one bucket.

Both sides sayBucketHandling
The same thingREINFORCEDOne row. Evidence from both. Strongest rows in the table.
Opposite things about the same ruleCONFLICTOne row. Resolve with the precedence rule below and show your work.
Different things about the same linesOVERLAPOne row. A single merged replacement — never two edits to one range.
Unrelated thingspass throughNormal rows.

REINFORCED — the guide recommends it and you keep asking for it

The strongest signal available. The guide says Opus 5 writes long documents and recommends a length rule; the session record shows four "too long" corrections. Same rule, two independent justifications.

Merge into one row. Cite both. Rank at the top — never split a reinforced finding across two rows, which reads as two weak findings instead of one strong one.

CONFLICT — observed behavior wins

migrate says DELETE rule X because the guide says the model handles it natively. learn says you corrected X anyway, three times, last week.

The session record wins. Keep the rule.

The guide describes what the model does by default. The correction history describes what this user actually needed. When they disagree, the evidence from this machine beats the general claim — the rule stays, and the row says why:

CONFLICT — kept against the guide
  guide:    "<verbatim quote recommending removal>"
  observed: 3 corrections across 2 sessions, most recent <date>
  -> KEEP. The guide describes the default; your sessions describe the exception.

Never resolve a conflict silently. The user should see that the two sources disagreed and that you chose, so they can overrule.

The one exception: if the corrections all predate the model change, the guide may now be right and the rule genuinely obsolete. Check the dates. If the corrections stopped when the model changed, say so and let the user decide — mark the row WEAK rather than picking for them.

OVERLAP — one edit per line range

If both sides touch the same lines, produce a single replacement satisfying both. Two rows editing one range means whichever applies second clobbers the first.

The table

Same two-tier contract as the other commands: numbered table, then complete untruncated before/after per row. Add one column so the user can see which analysis produced each row.

## Fit — /path/to/CLAUDE.md
188 lines / 8,682 B / ~2,346 tok per session
Model: <model> · guide fetched live   |   Sessions: <N>d, <S> sessions, <C> corrections

| #  | Lines   | What                  | Source     | Verdict | Why                              |
| -- | ------- | --------------------- | ---------- | ------- | -------------------------------- |
| 1  | (new)   | Deliverable length    | BOTH       | ADD     | guide recommends + 4 corrections |
| 2  | 89-113  | Sub-agent routing     | migrate    | DELETE  | all 7 agents missing from disk   |
| 3  | 12-14   | Verification step     | CONFLICT   | KEEP    | guide says cut; you asked 3x     |
| 4  | 63-66   | Scope Control para 1  | migrate    | KEEP    | duplicates harness, but 4 lines  |

Projected if all applied: 188 -> 174 lines

Order: REINFORCED first, then CONFLICT, then contradictions and unexecutable, then the rest. Source is one of BOTH / CONFLICT / migrate / learn.

Report both halves even when one is empty. "No new patterns in the last 14 days" is a real result and the user should see that it ran, not silently get a migrate-only table.

End with:

Reply with the numbers you want:
  apply 1,3     test 1,3     apply all     apply none

test N hands those rows to houserules:test, which A/B's each one against a real probe before anything is written. Offer it whenever a row is a REWRITE of existing text or its wording is a judgment call — those are the rows most likely to be quietly outcompeted by rules already in the file.

Apply — one pass

Only the numbers named.

  1. One backup, before anything. cp "<target>" "<target>.bak-$(date +%Y-%m-%d-%H%M%S)"
  2. Apply in descending line order. Editing L63 before L89 shifts every citation below it. Bottom-up keeps every unapplied row's line numbers valid.
  3. Pure additions (no line range) go last, into the section where they belong.
  4. Stamp once — not twice:
    <!-- houserules: reviewed <date> against <model>, sessions through <date>. -->
  5. Show the diff and the backup path. Name any rows the user skipped.

Anti-goals

  • Do not print two tables. One merged table is the entire point.
  • Do not write twice. One backup, one pass, descending line order.
  • Do not hide a conflict. Show both sources and your resolution.
  • Do not let the guide overrule the session record on a live conflict. A general claim about the model loses to evidence from this user's machine.
  • Do not reimplement migrate or learn. Follow their phases 1-4.
  • Do not decide for the user. Produce the table; they pick numbers.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.