A Claude Code skill bundle for bug hunting and external red-team work - 82 skills, 15 slash commands, 681 disclosed-report patterns curated across 24 core vulnerability classes, plus enterprise identity + infrastructure attack matrices.
Python ai-securityanthropicapplication-securitybug-bountybugbountybugcrowdclaudeclaude-codeclaude-skillsethical-hackinghackeroneoffensive-securitypentestingred-teamsecurity-toolsweb-security
#!/usr/bin/env python3
"""
verify_citations.py — content-verify every disclosed-report citation.
Credibility gate: a citation is only trustworthy if the report's ACTUAL subject
matches the skill/pattern it's cited under. Titles + login-gated bodies can drift
from what a pattern claims. This fetches each cited report's PUBLIC metadata
(title) + <meta> summary, re-runs the classifier over that content, and flags:
MISMATCH — content classifies to a different skill AND the citing skill's own
regex does not match the content (likely mis-bin / wrong cite).
WEAK — content classifies to nothing (thin public signal; not necessarily
wrong, but the cite can't be auto-corroborated — eyeball it).
OK — content corroborates the citing skill.
Reads docs/disclosed-reports/*.md for citations, reuses classify_reports rules,
reuses harvest_h1.meta_summary for the public snippet. Network-bound (one page
fetch per unique report) — polite 0.4s spacing. Cache in raw/ is reused.
Usage: python3 research/reports/verify_citations.py # verify all
python3 research/reports/verify_citations.py --skill hunt-saml
Stdlib only.
"""
import argparse, glob, json, os, re, sys, time
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import classify_reports as C
import harvest_h1 as H
ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
DR = os.path.join(ROOT, "docs", "disclosed-reports")
RAW = os.path.join(os.path.dirname(os.path.abspath(__file__)), "raw")
# citing-skill regex map (reuse the classifier's own rules, keyed by skill)
SKILL_RX = {s: rx for s, rx in C.COMPILED}
def citations():
"""report_id -> set(skills that cite it), from the corpus files."""
m = {}
for f in glob.glob(os.path.join(DR, "*.md")):
skill = os.path.splitext(os.path.basename(f))[0]
for rid in re.findall(r'reports/(\d+)', open(f).read()):
m.setdefault(rid, set()).add(skill)
return m
def cached_title(rid):
for f in glob.glob(os.path.join(RAW, "**", f"{rid}.json"), recursive=True):
try:
r = json.load(open(f))
return r.get("title") or "", r.get("summary") or "", r.get("cwe") or ""
except Exception:
pass
return None
def main(argv):
ap = argparse.ArgumentParser()
ap.add_argument("--skill", default=None, help="verify only citations under this skill")
ap.add_argument("--limit", type=int, default=0, help="cap reports checked (0=all)")
a = ap.parse_args(argv)
cites = citations()
ids = [r for r in cites if not a.skill or a.skill in cites[r]]
ids.sort()
if a.limit:
ids = ids[:a.limit]
ok = weak = mism = 0
flags = []
for i, rid in enumerate(ids):
c = cached_title(rid)
title, summary, cwe = c if c else ("", "", "")
url = f"https://hackerone.com/reports/{rid}"
if not summary: # fetch public snippet if not cached
summary = H.meta_summary(url)
time.sleep(0.4)
rec = {"title": title, "summary": summary, "cwe": cwe}
content_skill = C.classify_one(rec)
haystack = " ".join([title, summary, cwe])
for citing in sorted(cites[rid]):
if a.skill and citing != a.skill:
continue
rx = SKILL_RX.get(citing)
self_match = bool(rx and rx.search(haystack))
if not (title or summary):
verdict = "WEAK"; weak += 1
elif self_match or content_skill == citing:
verdict = "OK"; ok += 1
continue
elif content_skill and content_skill != citing:
verdict = "MISMATCH"; mism += 1
else:
verdict = "WEAK"; weak += 1
flags.append((verdict, citing, rid, content_skill, (title or summary)[:70]))
print(f"verify: {len(ids)} unique cited report(s) — {ok} OK, {weak} WEAK, {mism} MISMATCH")
for v, citing, rid, cs, t in sorted(flags):
print(f" {v:9} {citing:22} {rid:>8} content~{cs or '-':22} {t}")
return 0
if __name__ == "__main__":
sys.exit(main(sys.argv[1:]))