agentleFS
AkirakiiAkirakii/harness-doctor/skills/fix/SKILL.md

fix

Skill0 starsChanged yesterdayMITView on GitHub

Fix what Harness Doctor marked as failing - go through each regression, failing rule and uncovered problem, worst first and one at a time, find why the rule, skill or note failed in those sessions, and propose the exact edit. Use when the user pastes "/harness-doctor:fix", or asks "why does X keep failing", "fix my failing rules", "what's wrong with skill X", "is my CLAUDE.md working", "diagnose X", "which of my rules are failing".

90 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use feature, frontend-code-review and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • bradtraversySkill

    feature

    Manage current feature workflow - start, review, explain or complete

    Also in 10 projects that use this

  • chendongqiSkill

    frontend-code-review

    Trigger when the user requests a review of frontend files (e.g., `.tsx`, `.ts`, `.js`). Support both pending-change reviews and focused file reviews while applying the checklist rules.

    Also in 4 projects that use this

  • GoogleSkill

    pr-creator

    Also in 4 projects that use this

  • PJuniszewskiSkill

    cook

    Feature development with guardrails. Plan → Review → Code → Ship.

    Also in 3 projects that use this

  • vbrevikSkill

    work

    Use when the user says 'plan', 'todo', 'copy plan', 'append plan', 'resume plan', 'priorities', or 'what's next'.

    Also in 4 projects that use this

  • Microsoft AzureSkill

    test

    Also in 12 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 83 edited versions we read keep these.

  • 5

    Distinctive

    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. Very specific:

    Get the report: Run this, adding --project <name> when the user named a project (the word after /harness-doctor:fix):

    line 14
  2. Very specific:

    If it says no sessions are judged, tell the user to open the page (/harness-doctor:open) and press the judge button.

    line 19
  3. Distinctive:

    Check the marks before believing them: Read the rule as written, then each bad moment.

    line 22
  4. Distinctive:

    When the one line and the quotes aren't enough, open the session log the report names and read around the quote.

    line 23
  5. Distinctive:

    Write it as the general rule the failure breaks, with no project or topic words from the sessions.

    line 31
  6. Distinctive:

    Wait for the user's yes, apply it, and go to the next item.

    line 32
  7. Distinctive:

    If the user has their own skill for saving rules or knowledge, use it to place the edit.

    line 33

“Common” is measured: kept by at least 6 in 10 of the 83 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

Fix what the doctor marked, one rule at a time

Harness Doctor judged the user's sessions against every rule and fact they wrote, and marked each rule's moments good or bad. The marks are a model's judgment, so they can be wrong: the user is the check, here in the conversation. This skill never writes to the doctor's own data, and it changes a file only after the user says yes.

Steps

  1. Get the report. Run this, adding --project <name> when the user named a project (the word after /harness-doctor:fix):
    python3 "${CLAUDE_PLUGIN_ROOT}/app/harness_doctor.py" --report --data "${CLAUDE_PLUGIN_DATA}"
    It lists, worst first: regressions (a rule that was clean before its last edit), failing rules, then problems nothing written covers. If it says no sessions are judged, tell the user to open the page (/harness-doctor:open) and press the judge button.
  2. Say what's there in one line, like "1 regression, 4 failing rules, 2 not covered", then start with the first one. Take them one at a time; never propose edits for several at once.
  3. Check the marks before believing them. Read the rule as written, then each bad moment. When the one line and the quotes aren't enough, open the session log the report names and read around the quote. Tell the user plainly when a mark looks wrong ("this one didn't break the rule, because …") and drop it; a rule isn't changed for a wrong mark.
  4. Find why it failed, from the moments that are right:
    • Didn't load (a skill's description, a note's "read when"): the user's words in those moments aren't in the loading line. The fix goes in that line, using the words the user actually said.
    • Bad use of a rule: it doesn't name this case, it's vague about when it applies, another rule pulls the other way (name that rule), or it's buried among others.
    • Regression: read the last edit the report shows. Say what the edit removed or changed that the bad moments needed, and whether putting it back fixes it.
    • A fact in a note: the fact is wrong or out of date. The fix is the right fact, checked, not a new instruction.
  5. Propose one edit, ready to apply: the exact text to replace and the new text, with one sentence on why. Write it as the general rule the failure breaks, with no project or topic words from the sessions; one of the moments can follow as a short example. Wait for the user's yes, apply it, and go to the next item. If the user has their own skill for saving rules or knowledge, use it to place the edit.
  6. For a problem nothing covers, decide with the user where the fix belongs: an instruction goes to CLAUDE.md, the output style or a skill; a fact goes to a note; something that must happen every time without thinking goes to a hook.

After

The edit shows on the page as a new version of that rule, with its marks starting empty. Later sessions, once judged, show whether it helped; if the rule turns red again, the page says "Regression".

Output

Short. For each item: what failed and why in two lines, then the edit. No recap at the end.

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.