Bruno-Cunha-Souza/ValarMindSkills / skills/celebrimbor
celebrimbor
UnverifiedAnalyzes a target project (languages, frameworks, libs, purpose, docs, tests, CI, infra, existing AGENTS.md/skills) and proposes project-specific skills, agents, and tool scripts with file:line evidence; after the user approves, generates them in a portable layout (.agents/skills, .agents/agents, AGENTS.md routing block, optional per-runtime adapters) that works across AGENTS.md-aware agents. Read-only analysis; never overwrites without confirmation. Triggers: 'forjar skills para o projeto', 'analisar projeto e sugerir skills', 'criar agentes para este projeto', 'forge project skills', 'suggest skills for this project', 'generate project agents', '/celebrimbor'.
| File | Kind | Size |
|---|---|---|
| EXAMPLE.md | markdown | 28.3 KB |
| SKILL.md | skill md | 25.9 KB |
| references/ARTIFACT_FORMATS.md | markdown | 22.4 KB |
| references/CHECKLIST.md | markdown | 10.2 KB |
| references/DETECTION_CATALOG.md | markdown | 13.1 KB |
references/CHECKLIST.md
View on GitHub> Reference companion for the [celebrimbor](../SKILL.md) skill.
# Checklist
Phase 5 walks every item from the target root. A failing item blocks the report until it is fixed or recorded as skipped with a reason; every output line of the bash block below is a failure, except `rm -rf` / `shutil.rmtree(` hits scoped to `"$OUT_DIR"` and `skip:` lines (record those as skipped with their reason). Formats: [ARTIFACT_FORMATS](ARTIFACT_FORMATS.md).
## Profile
- [ ] Every non-null profile field carries `evidence: file:line`
- [ ] Gaps are recorded as `file (absent)`, not dropped
## Suggestions
- [ ] Every row has all schema fields: `id`, `type`, `name`, `family`, `priority`, `depth`, `rationale`, `evidence`, `triggers`, `depends_on`, `uses_existing`, `portability`, `generates`, `status`
- [ ] `evidence` is non-empty on every row
- [ ] Both `review` baseline rows are present (`code-review`, `code-security-review`, or `<project>-` prefixed when an installed skill has the name), `proposed` or dedupe `skipped`
- [ ] Dedupe applied: a `name` already in `existing_skills` / `existing_agents` is an `update:` row when its frontmatter has `source: celebrimbor` (`proposed` with each profile change or depth-floor gap cited as `file:line` / `path (absent)`, else `skipped` as up to date), otherwise an `extend` row, `skipped` by default; overlapping artifacts under other names appear in `uses_existing` with `path:line`, never as a silent drop
- [ ] The approval list the user gave (IDs) is recorded; only `approved` rows were generated
## Generated skills
- [ ] Frontmatter parses with `yaml.safe_load` (skipped when PyYAML is absent) and holds only `name`, `description`, `source: celebrimbor`
- [ ] `name` equals the directory name (kebab, ≤64, no `claude` / `anthropic`)
- [ ] `description` is third person and ends with `Triggers: '...', '/<name>'.` (PT and EN)
- [ ] Every `references/*.md` opens with the companion banner
- [ ] Depth floor met per tier ([ARTIFACT_FORMATS §3](ARTIFACT_FORMATS.md)): T3 has `scripts/run.sh` + ≥2 check scripts, ≥2 references with ≥8 check rows each plus `SEVERITY.md`, and `EXAMPLE.md`; T2 has `scripts/run.sh`, a reference with ≥5 check rows, and `EXAMPLE.md` (mechanical check below)
- [ ] Every `references/*.md` and `EXAMPLE.md` cites this project's paths as `file:line`; every check row has a detection, a project anchor and a fix in this project's idiom; adapted files name their source in the banner
- [ ] Both `review` skills have a `## Scope boundary` section: security findings need a security class and an attacker path, everything else is handed off with `file:line`
## Agent role files
- [ ] Frontmatter parses; `name` equals the file basename; `description` carries triggers
- [ ] No `tools:` key
- [ ] Every entry in `skills:` resolves to `.agents/skills/<name>/SKILL.md`; every `scripts:` path exists
## AGENTS.md block
- [ ] Exactly one `<!-- CELEBRIMBOR BEGIN` line and one `<!-- CELEBRIMBOR END -->` line, BEGIN before END (mechanical check below)
- [ ] Idempotent: content outside the markers is byte-identical before and after the run (re-running leaves it unchanged)
## Tool scripts
- [ ] `bash -n` passes on every `scripts/*.sh`; every `scripts/*.py` parses, is stdlib only, and has no `shell=True` / `os.system(`
- [ ] Forbidden-token grep is clean; any `rm -rf` hit targets `"$OUT_DIR"` only
- [ ] Every active script (sends traffic / reads `TARGET`) has the `I_HAVE_AUTHORIZATION=1` check and the exact-host gate from ARTIFACT_FORMATS §6 (scheme, path/query/fragment, userinfo, port stripped before the `case`) inline; every active `.py` re-checks `I_HAVE_AUTHORIZATION`
- [ ] No prefix glob in any gate (`localhost*`, `127.0.0.1*`, `*.test*`): a hit is a failure, since it accepts `localhost.evil.com`
- [ ] Matched secret values are redacted inside the jq filter (slice `.[0:4]+"…"+.[-4:]`, emitted via `@tsv`), never interpolated raw (`"\(.Secret)"`) nor read line-by-line; raw tool reports holding values are deleted
- [ ] `out/` is gitignored (`run.sh` or its `lib/common.sh` writes `$OUT_DIR/.gitignore`)
## Adapters
- [ ] Every adapter of a generated artifact under `.claude/` and `.cursor/` resolves; fallback copies match `.agents/` (`diff -rq` clean)
- [ ] `.agent/skills/<name>/` (Antigravity) of a generated skill is a copy, never a symlink, and in sync
## Overwrite
- [ ] No existing file was overwritten without a shown diff and an explicit yes; declined files are listed as skipped; `CLAUDE.md` was touched only after its own yes, never created
## Prompt audit
- [ ] Each generated SKILL.md and agent file has an audit line, or `prompt audit: skipped — not available`
## Mechanical checks
Run with `bash`, never zsh (zsh aborts a line on an unmatched glob). Export three values from the report file first: `PRE_HASH` (empty if AGENTS.md did not exist before Phase 3), `GENERATED` = the space-separated `generates` paths of every `approved` row (canonical `.agents/` paths), and `DEPTH` = `<name>:<tier>` per approved skill. Every check below runs only on those paths — pre-existing user skills, agents and adapters are never inspected or edited.
```bash
: "${GENERATED:?export GENERATED from the report first}"
skills=$(printf '%s\n' $GENERATED | grep -o '^\.agents/skills/[^/]*' | sort -u)
agents=$(printf '%s\n' $GENERATED | grep '^\.agents/agents/[^/]*\.md$' | sort -u)
scripts=$(for d in $skills; do find "$d/scripts" -name '*.sh' -not -path '*/out/*' 2>/dev/null; done)
py=$(for d in $skills; do find "$d/scripts" -name '*.py' -not -path '*/out/*' 2>/dev/null; done)
if python3 -c 'import yaml' 2>/dev/null; then python3 - $(printf '%s/SKILL.md\n' $skills) $agents <<'PY' # frontmatter, name == dir, triggers, tools:, paths
import os, sys, yaml
for p in (a for a in sys.argv[1:] if os.path.isfile(a)):
t, agent = open(p).read(), "/agents/" in p
want = os.path.basename(p)[:-3] if agent else os.path.basename(os.path.dirname(p))
try: fm = yaml.safe_load(t.split("\n---", 1)[0][4:]) if t.startswith("---\n") else None
except yaml.YAMLError: fm = None
if not isinstance(fm, dict): print("FRONTMATTER:", p); continue
d = str(fm.get("description", ""))
errs = [m for m, bad in [("name != " + want, fm.get("name") != want), ("tools: key", agent and "tools" in fm),
("no triggers", "Triggers:" not in d and not (agent and "Use when:" in d))] if bad]
errs += [f"missing {k} {v}" for k, f in (("skills", ".agents/skills/{}/SKILL.md"), ("scripts", "{}")) for v in (fm.get(k) or []) if not os.path.isfile(f.format(v))]
for e in errs: print(f"{p}: {e}")
PY
else echo "skip: PyYAML not installed — frontmatter check not run"; fi
for d in $skills; do for r in "$d"/references/*.md; do [ -f "$r" ] && ! head -1 "$r" | grep -q '^> Reference companion for the \[' && echo "BANNER: $r"; done; done
for s in $scripts; do [ -f "$s" ] && { bash -n "$s" || echo "SYNTAX: $s"; }; done
for p in $py; do python3 -c 'import ast,sys; ast.parse(open(sys.argv[1]).read())' "$p" 2>/dev/null || echo "PY SYNTAX: $p"; done # parse only: py_compile would write __pycache__
[ -z "$py" ] || grep -nE 'shell=True|os\.system\(|shutil\.rmtree\(|pip install' $py
[ -z "$py" ] || for f in $(grep -lE 'urllib\.request|http\.client|socket\.|requests\.' $py); do grep -q 'I_HAVE_AUTHORIZATION' "$f" || echo "NO GATE: $f"; done
rows() { grep -cE '^\| *[^ |-]' "$1" | awk '{print $1-1}'; } # table rows minus the header
for nt in ${DEPTH:-}; do n=${nt%%:*} t=${nt##*:}; d=.agents/skills/$n; [ -d "$d" ] || continue # depth floor (ARTIFACT_FORMATS §3)
case $t in T2|T3) for f in scripts/run.sh EXAMPLE.md; do [ -f "$d/$f" ] || echo "DEPTH: $n $t needs $f"; done ;; esac
best=0; nref=0; thin=""
for r in "$d"/references/*.md; do [ -f "$r" ] || continue; case $r in */SEVERITY.md) continue ;; esac
nref=$((nref+1)); k=$(rows "$r"); [ "$k" -gt "$best" ] && best=$k; [ "$k" -lt 8 ] && thin="$thin ${r##*/}($k)"; done
case $t in
T3) [ "$nref" -ge 2 ] || echo "DEPTH: $n T3 needs ≥2 check references (has $nref)"; [ -z "$thin" ] || echo "DEPTH: $n T3 references under 8 rows:$thin"
[ -f "$d/references/SEVERITY.md" ] || echo "DEPTH: $n T3 needs references/SEVERITY.md"
k=$(find "$d/scripts" \( -name '*.sh' -o -name '*.py' \) -not -name run.sh -not -path '*/lib/*' -not -path '*/out/*' 2>/dev/null | wc -l)
[ "$k" -ge 2 ] || echo "DEPTH: $n T3 needs ≥2 check scripts besides run.sh (has $((k+0)))"
grep -q '^## Scope boundary' "$d/SKILL.md" || echo "SCOPE: $n lacks ## Scope boundary" ;;
T2) [ "$best" -ge 5 ] || echo "DEPTH: $n T2 needs a reference with ≥5 check rows (best $best)" ;;
esac
for f in "$d"/references/*.md "$d/EXAMPLE.md"; do [ -f "$f" ] && ! grep -qE '[[:alnum:]_./-]+\.[[:alnum:]]+:[0-9]+' "$f" && echo "NO PROJECT ANCHOR: $f"; done
done
[ -z "$scripts" ] || grep -nE 'rm -rf|git (push|reset --hard|clean)|sudo |chmod 777|(curl|wget)[^|]*\| *(ba)?sh|DROP |TRUNCATE ' $scripts
[ -z "$scripts" ] || { grep -nE '(localhost|127\.0\.0\.1|\.test)\*' $scripts && echo "PREFIX GLOB in gate"; }
[ -z "$scripts" ] || grep -nE '\\\(\.(Secret|Match)\)' $scripts | sed 's/^/RAW SECRET INTERPOLATION: /'
active=$([ -z "$scripts" ] || grep -lE 'TARGET|curl |wget |zap-baseline' $scripts)
[ -z "$active" ] || for pat in 'I_HAVE_AUTHORIZATION:-}" = "1"' 'host="${host%%[/?#]*}"' "localhost|127.0.0.1|'[::1]'|*.localhost|*.test)"; do
for f in $(grep -LF "$pat" $active); do echo "NO GATE: $f lacks $pat"; done; done
for s in $scripts; do case $s in */scripts/run.sh) grep -qs '\.gitignore' "$s" "${s%/run.sh}/lib/common.sh" || echo "OUT NOT IGNORED: $s" ;; esac; done
for p in $skills $agents; do for a in ".claude/${p#.agents/}" ".cursor/${p#.agents/}" ".agent/${p#.agents/}"; do # adapters of generated artifacts only
[ -e "$a" ] || [ -L "$a" ] || continue
if [ -L "$a" ]; then [ -e "$a" ] || echo "BROKEN LINK: $a"; case $a in .agent/*) echo "ANTIGRAVITY SYMLINK: $a" ;; esac
else diff -rq "$p" "$a"; fi; done; done
[ ! -f AGENTS.md ] || awk '/^<!-- CELEBRIMBOR BEGIN/{b++;bl=NR}/^<!-- CELEBRIMBOR END -->/{e++;el=NR}END{if(b!=1||e!=1||bl>el)print "MARKERS: need exactly one BEGIN then one END (begin="b+0" end="e+0")"}' AGENTS.md
[ -z "${PRE_HASH:-}" ] || [ "$(sed '/^<!-- CELEBRIMBOR BEGIN/,/^<!-- CELEBRIMBOR END -->/d' AGENTS.md | awk 'NF{for(;n;n--)print "";print;next}{n++}' | shasum)" = "$PRE_HASH" ] || echo "OUTSIDE-MARKERS CHANGED"
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |