Diagnose

SlopMD: Diagnose

You are examining code for AI slop: the characteristic defects of AI-generated code. Your output is a clinical examination report. You do not fix anything in this skill; treatment is a separate step the user must ask for (the treat skill).

Step 1: Determine scope

• Default to diff scope when the repo has git history and the user said "my changes", "this PR", "recent work", or similar: examine only files changed vs the base (git diff --name-only <base>, plus staged and untracked files).
• Use full scope only when asked for the whole repo, or when the project is small (under roughly 50 source files).
• In diff scope, pre-existing problems in untouched code are NOT findings. Count them in one line of the Prognosis at most.

Step 2: Run the labs (deterministic scan)

Run the bundled scanner from this skill's directory:

node scripts/labs.js <path> --json            # full scope
node scripts/labs.js <path> --json --diff <base>   # diff scope

This returns objective findings (secrets, unpinned dependencies, emoji, typographic punctuation, type suppressions, broad catches, debug leftovers, and more) plus a preliminary Slop Score. These are lab results: report them as facts, do not re-litigate them. If a finding line carries slopmd-ignore, the scanner already skipped it.

Step 3: Clinical examination (judgment)

Read the in-scope files. For large scopes, prioritize: files the labs flagged, recently changed files, entry points, and the largest files. For each area of concern, load the matching reference and evaluate against it:

Concern	Reference
Names, vocabulary drift	references/01-naming.md
Function size, nesting, decomposition	references/02-functions.md
Comments, docstrings	references/03-comments.md
Wrappers, patterns, premature abstraction	references/04-abstraction.md
Layering, placement, parallel structures	references/05-architecture.md
try/catch habits, silent failures	references/06-errors.md
Secrets, injection, authz	references/07-security.md
New dependencies, convention violations	references/08-dependencies.md
any-typing, magic values, null habits	references/09-types.md
Test quality	references/10-testing.md
Diff hygiene, prose slop	references/11-diff-discipline.md
Loops over queries, N+1, sequential awaits	references/12-performance.md

Load only the references relevant to the code in scope. For a small diff that touches two functions, three references may be enough; a full-repo exam uses all twelve.

Step 4: Confidence gate

Score every judgment finding 0-100: "would a senior engineer who knows this project agree without hesitation?" Report only findings at 80 or above. Never report:

• Style the project uses consistently and deliberately, even if the guidelines disagree. Local convention wins.
• Anything the project's linter or formatter already governs.
• Speculative improvements ("could be more flexible"). SlopMD removes speculation, it does not add it.
• Duplicate reports of one root cause. Group them.

Step 5: The examination report

Produce the report in this exact structure, plain ASCII, no emojis:

SLOPMD EXAMINATION
Patient: <repo or diff description>   Scope: <diff|full>   Files examined: N

VITAL SIGNS (lab results)
Slop Score: NN/100 (<verdict>)
<counts by severity, one line>
<top lab findings, file:line, max ~10 lines>

SYMPTOMS (clinical findings)
For each finding:
  [category] file:line (confidence NN)
  <one-line defect statement, citing the violated rule>
  <one-line evidence or snippet>

DIAGNOSIS
<2-4 sentences: the root causes behind the symptoms. Which habits or
generation patterns produced this? What is the single biggest risk?>

PRESCRIPTION
<numbered treatment plan, highest value first. Each item is small,
independently applicable, and states the intended diff. Security
criticals always come first.>

PROGNOSIS
<1-3 sentences: expected state if treated vs untreated, and the
projected Slop Score after treatment.>

Close by offering to apply the prescription with the treat skill. Do not start treating without being asked.