|
|
@@ -3,10 +3,11 @@
|
|
|
#
|
|
|
# Checks:
|
|
|
# 1. Component counts on disk vs claims in README.md header, AGENTS.md
|
|
|
-# overview bullets, and docs/PLAN.md inventory table
|
|
|
+# overview bullets, docs/PLAN.md inventory table, and selected README prose
|
|
|
# 2. Every skill directory has a row in a README skill table
|
|
|
# 3. Every repo-relative markdown link in README.md / AGENTS.md resolves
|
|
|
# to an existing file or directory (no ghost references)
|
|
|
+# 4. Skill frontmatter references only skills that exist on disk
|
|
|
#
|
|
|
# Exit 0 = clean, exit 1 = drift detected.
|
|
|
set -u
|
|
|
@@ -44,6 +45,27 @@ else
|
|
|
[ "$r_rules" = "$rules_disk" ] || err "README header: $r_rules rules claimed, $rules_disk on disk"
|
|
|
fi
|
|
|
|
|
|
+# --- Selected README prose count claims -------------------------------------
|
|
|
+# These five known count-bearing patterns are intentionally exhaustive; newly
|
|
|
+# introduced prose patterns must be added explicitly if they should be gated.
|
|
|
+check_readme_prose() { # $1=regex $2=disk-count $3=label
|
|
|
+ local line claim matched=0
|
|
|
+ while IFS= read -r line; do
|
|
|
+ [ -n "$line" ] || continue
|
|
|
+ matched=$((matched + 1))
|
|
|
+ claim="$(echo "$line" | grep -oE '[0-9]+' | head -1)"
|
|
|
+ [ "$claim" = "$2" ] || err "README.md: $claim $3 claimed, $2 on disk"
|
|
|
+ done < <(grep -E "$1" README.md || true)
|
|
|
+ # Zero matches = the count-bearing line was deleted or reworded, which is
|
|
|
+ # drift too — a guard that finds nothing to check must not stay silent.
|
|
|
+ [ "$matched" -ge 1 ] || err "README.md: prose pattern for '$3' matched no lines (deleted/reworded?)"
|
|
|
+}
|
|
|
+check_readme_prose '[0-9]+ specialized skills' "$skills_disk" "specialized skills"
|
|
|
+check_readme_prose '[0-9]+ on-demand skills' "$skills_disk" "on-demand skills"
|
|
|
+check_readme_prose 'Custom skills \([0-9]+\)' "$skills_disk" "custom skills"
|
|
|
+check_readme_prose 'Slash commands \([0-9]+\)' "$commands_disk" "slash commands"
|
|
|
+check_readme_prose 'Expert subagents \([0-9]+\)' "$agents_disk" "expert subagents"
|
|
|
+
|
|
|
# --- AGENTS.md overview bullets ---------------------------------------------
|
|
|
check_agents_md() { # $1=regex $2=disk-count $3=label
|
|
|
local claim
|
|
|
@@ -92,6 +114,34 @@ for doc in README.md AGENTS.md; do
|
|
|
| sed -E 's/^\]\(//; s/\)$//')
|
|
|
done
|
|
|
|
|
|
+# --- 4. Skill frontmatter ghost references ---------------------------------
|
|
|
+# These are Claude-Code-bundled skills, not skills shipped by this repo.
|
|
|
+# Keep this allowlist narrow: every other unknown reference is a failure.
|
|
|
+external_skills="frontend-design"
|
|
|
+
|
|
|
+for skill_file in skills/*/SKILL.md; do
|
|
|
+ while IFS= read -r refs; do
|
|
|
+ refs="${refs#*:}"
|
|
|
+ refs="${refs#"${refs%%[![:space:]]*}"}"
|
|
|
+ refs="${refs%"${refs##*[![:space:]]}"}"
|
|
|
+ refs="${refs#\"}"; refs="${refs%\"}"
|
|
|
+ IFS=',' read -ra names <<< "$refs"
|
|
|
+ for name in "${names[@]}"; do
|
|
|
+ name="${name#"${name%%[![:space:]]*}"}"
|
|
|
+ name="${name%"${name##*[![:space:]]}"}"
|
|
|
+ [ -z "$name" ] && continue
|
|
|
+ case " $external_skills " in *" $name "*) continue ;; esac
|
|
|
+ [ -d "skills/$name" ] || err "$skill_file: unknown skill reference '$name'"
|
|
|
+ done
|
|
|
+ done < <(awk '
|
|
|
+ NR == 1 && $0 == "---" { frontmatter=1; next }
|
|
|
+ frontmatter && $0 == "---" { exit }
|
|
|
+ frontmatter && /^metadata:/ { metadata=1; next }
|
|
|
+ metadata && /^[^ ]/ { metadata=0 }
|
|
|
+ metadata && /^ (related-skills|depends-on):/ { print }
|
|
|
+ ' "$skill_file")
|
|
|
+done
|
|
|
+
|
|
|
echo
|
|
|
if [ "$errors" -eq 0 ]; then
|
|
|
echo "doc-drift: clean"
|
0xDarkMatter