claude-code-plugins

Pre-computed context

check's config resolve ran at load time. The detector's --show-config output below names the
layer supplying each effective value; (detector unavailable) means it did not run:

!"${CLAUDE_PLUGIN_ROOT}/skills/audit/scripts/detect.sh" --show-config 2>/dev/null || echo "(detector unavailable)"

When it reads (detector unavailable) or [shell command execution disabled by policy], run the
detector's --show-config via Bash or read the three layers directly.

Purpose

The ai-slop detector ships neutral defaults. A consumer repo declares its own exemptions and
tuning in .claude/ai-slop.json, resolved per the config-cascade convention: user-global
(~/.claude/ai-slop.json), team (.claude/ai-slop.json, tracked), local overlay
(.claude/ai-slop.local.json, gitignored). Later layers refine earlier ones per key.

Keys

Key	Type	Meaning
excluded_paths	glob list	Files the audit never scans
em_dash_allowed_paths	glob list	Documents exempt from the zero-tolerance em-dash rule
vocab_add	word list	Additions to the AI-vocabulary list
vocab_remove	word list	Removals from the AI-vocabulary list
phrase_add	ERE fragment list	Additions to the model-era phrase roster (rule-model-era-phrases); whole fragments, spaces allowed, apostrophes spelled .
phrase_remove	ERE fragment list	Removals from the shipped phrase roster, matched verbatim against the shipped fragments
disabled_rules	rule slugs	Rules the audit skips entirely (reported as disabled)
thresholds	map	Per-rule density thresholds: ai_vocabulary, copulative_avoidance (matches per 1000 words; density rules also need at least 3 matches)
_comment	string	Free-text rationale for the choices in this file. Read by nobody; JSON has no comment syntax, and a config that disables a rule without recording why is the drift this skill exists to catch. Not drift, so do not flag it as an unknown key

check (default, read-only)

Report the current state and change nothing:

1. Read the pre-computed --show-config output (it names the layer supplying each value), or read
   the layers directly when it is unavailable; report which layer wins each key and which layers are
   absent.
2. Flag drift: unknown keys, an em-dash threshold key (the rule is zero-tolerance by design, and
   per-document exemption via em_dash_allowed_paths is the supported mechanism), globs that
   match nothing, a disabled_rules slug that names no shipped rule, a phrase_add fragment
   that is not a valid ERE (the detector skips it with a stderr note; surface it here as
   drift), or a phrase_remove fragment matching no shipped phrase.
3. End with what apply would change, if anything was flagged.

apply

Everything check does, then the confirmed write:

1. Interview the user for what they want changed. For em-dash exemptions, ask for the documents
   that genuinely require em dashes; the default posture is that most work does not.
2. Write ONLY the team layer (.claude/ai-slop.json), showing the diff and getting explicit
   confirmation before writing. Never write the user-global or overlay layers on the user's
   behalf; name them as options the user edits themselves.
3. Re-run --show-config via Bash (the pre-computed output predates the write) and report the
   new effective state. Then run the tracked-file pair on
   the written team file: git check-ignore -v .claude/ai-slop.json reports no match (a match is
   FAIL with the pattern) AND git ls-files --error-unmatch .claude/ai-slop.json exits 0
   (non-zero right after a fresh write means "written but untracked: commit it to share with the
   team", never success).