Files
Anthropic-Cybersecurity-Skills/tools/validate-agentskills.py
T
Mahipal 796d96c413 fix: replace three hand-rolled YAML parsers with a shared PyYAML loader
index.json shipped 604 of 817 descriptions truncated to their first line.
The cause was the inline regex parser in update-index.yml: it reconstructed
multi-line descriptions only for the YAML block-scalar styles ('>' and '|')
and silently dropped continuation lines for every other style.

A census of the corpus explains the blast radius:

  block scalar   (description: >-)    43
  single-quoted multiline            278
  plain unquoted multiline           496
  single-line                          0

So 774 of 817 skills (94.7%) used a style the parser mishandled. Commit
d56fc0a7 had fixed only the 43 block-scalar files, and CONTRIBUTING.md
recommends that one working style, which is why it stayed hidden.

- add tools/skill_frontmatter.py, the single PyYAML-backed loader
- add tools/generate-index.py so generation is testable outside CI, with
  a --check mode for use as a gate
- delete the hand-rolled parsers from validate-skill.py (98 lines) and
  validate-agentskills.py, routing both through the shared loader
- implement the reserved-word check that agentskills-skill.schema.json
  names validate-agentskills.py as the enforcement point for

Verified by a differential harness against yaml.safe_load ground truth:
index-vs-source mismatches 606 -> 0.
2026-08-23 17:15:12 +02:00

123 lines
4.7 KiB
Python

#!/usr/bin/env python3
"""Validate SKILL.md frontmatter against the strict agentskills.io standard.
Reports, per skill, any deviation from tools/agentskills-skill.schema.json plus
the two constraints JSON Schema can't express (name == parent dir; no angle
brackets in frontmatter). READ-ONLY; never edits files.
Usage:
python3 tools/validate-agentskills.py # summary + report
python3 tools/validate-agentskills.py --json # machine-readable JSON
python3 tools/validate-agentskills.py --strict # exit 1 if any non-compliant
"""
import os, re, sys, json, glob
from collections import Counter
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from skill_frontmatter import description_of, load_frontmatter, FrontmatterError
REPO = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
ALLOWED = {"name", "description", "license", "compatibility", "metadata", "allowed-tools"}
NAME_RE = re.compile(r"^[a-z0-9]+(-[a-z0-9]+)*$")
# The agentskills.io standard forbids reserved vendor words in a skill name.
# tools/agentskills-skill.schema.json names this script as the enforcement
# point, but the check was never actually implemented until now.
RESERVED_NAME_WORDS = ("anthropic", "claude")
def validate(path):
slug = os.path.basename(os.path.dirname(path))
problems = []
try:
fm = load_frontmatter(path)
except FrontmatterError as exc:
return slug, [str(exc)], []
keys = list(fm.keys())
if "name" not in keys:
problems.append("missing required key: name")
if "description" not in keys:
problems.append("missing required key: description")
name = str(fm.get("name", "") or "")
if name:
if not NAME_RE.match(name):
problems.append(f"name not lowercase-kebab-case: {name!r}")
if not (1 <= len(name) <= 64):
problems.append(f"name length {len(name)} out of 1..64")
if name != slug:
problems.append(f"name {name!r} != directory {slug!r}")
for reserved in RESERVED_NAME_WORDS:
if reserved in name.lower():
problems.append(f"name contains reserved word {reserved!r}: {name!r}")
desc = description_of(fm)
if desc:
if not (1 <= len(desc) <= 1024):
problems.append(f"description length {len(desc)} out of 1..1024")
elif "description" in keys:
problems.append("description empty")
# Angle brackets are an injection risk. Checking the PARSED values (rather
# than the raw text) means YAML block-scalar indicators like `>-` are never
# mistaken for content, so no indicator-stripping hack is needed.
for key, value in fm.items():
if isinstance(value, str) and ("<" in value or ">" in value):
problems.append(f"frontmatter value for {key!r} contains angle brackets "
"(injection risk / not allowed)")
break
# Additional top-level keys are PERMITTED by the standard (name+description
# are the only required fields). They are reported for information, not
# counted as compliance failures.
nonstd = [k for k in keys if k not in ALLOWED]
return slug, problems, nonstd
def main():
as_json = "--json" in sys.argv
strict = "--strict" in sys.argv
skills = sorted(glob.glob(os.path.join(REPO, "skills", "*", "SKILL.md")))
results = []
nonstd_hist = Counter()
compliant = 0
for p in skills:
slug, problems, nonstd = validate(p)
nonstd_hist.update(nonstd)
if not problems:
compliant += 1
results.append({"skill": slug, "compliant": not problems, "problems": problems})
noncompliant = [r for r in results if not r["compliant"]]
summary = {
"total": len(skills),
"compliant": compliant,
"noncompliant": len(noncompliant),
"nonstandard_key_frequency": dict(nonstd_hist.most_common()),
}
if as_json:
print(json.dumps({"summary": summary, "results": results}, indent=1))
else:
print(f"agentskills.io compliance: {compliant}/{len(skills)} compliant, "
f"{len(noncompliant)} non-compliant")
print("\nNon-standard top-level keys (count of skills carrying each):")
for k, n in nonstd_hist.most_common():
print(f" {k:20s} {n}")
# distinct problem types (excluding the per-key nonstd noise)
other = Counter()
for r in noncompliant:
for pr in r["problems"]:
if not pr.startswith("non-standard top-level key:"):
other[re.sub(r':.*$', '', pr)] += 1
if other:
print("\nOther (non-key) issues:")
for k, n in other.most_common():
print(f" {k}: {n}")
if strict and noncompliant:
sys.exit(1)
if __name__ == "__main__":
main()