agentsclimarketplace
Sign in

Treat

Skill flopeztancredi/slopmd/skills/treat

Apply the prescription from a SlopMD examination, fixing AI slop findings one intent per diff and verifying the result. Use when the user asks to fix, clean up, treat, or deslopify code after a diagnosis, or says to apply the prescription. Requires an examination report; runs the diagnose skill first if none exists.From its SKILL.md

Install
npx -y skills add flopeztancredi/slopmd --skill treat

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.6 KB, 498 tokens by cl100k_base, as published. Nobody here has run it

SlopMD: Treat

You are applying the prescription from a SlopMD examination. If there is no examination report in the conversation, run the diagnose skill first; never treat blind.

Triage order

  1. Security criticals (hardcoded secrets, injection, missing authz). Also tell the human that any exposed secret must be rotated; removing it from code is not enough.
  2. Silent failure modes (swallowed errors, masking defaults, tests that cannot fail).
  3. Structure (wrong-layer code, decorative wrappers, duplication of existing helpers).
  4. Local quality (naming, comments, magic values, type suppressions).
  5. Cosmetics (typographic punctuation, emoji, prose slop).

Rules of treatment

  • One intent per change-set. Each prescription item becomes its own coherent diff; if the project uses commits, commit each item separately with an imperative message stating intent and effect. Never mix a refactor with a behavior change.
  • Minimal incision. Fix the finding, touch nothing else: no reformatting untouched lines, no drive-by renames, no bonus improvements.
  • Verify after each item. Run the project's tests, build, or type check after every item. Broken by the fix means fix it or revert it before moving on; never leave the patient worse.
  • Judgment items need consent. Findings whose fix changes architecture, deletes a fallback branch, removes a public API, or touches behavior are proposed with a concrete plan and applied only after the user approves, unless the user already pre-approved aggressive treatment.
  • Respect the confidence gate. Do not apply anything the examination scored below 80; if you disagree with a gated finding during treatment, say so instead of silently acting.
  • Never delete or skip a failing test to get to green. If a test is genuinely wrong, state why and get agreement.

Discharge

After all items, re-run the scanner from the diagnose skill (node ../diagnose/scripts/labs.js <path> --json, diff mode if applicable) and report:

SLOPMD DISCHARGE SUMMARY
Slop Score: NN -> MM
Treated: <n items, one line each: what changed>
Deferred: <items awaiting approval or out of scope, with reasons>
Follow-up: <anything the human must do, e.g. rotate exposed credentials>

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,614. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.