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
# shellcheck shell=bash
# =====================================================================
# hunt — bug-bounty engagement scaffolding
#
# Adds a `hunt` shell function that creates a per-target working folder
# under ~/Targets/ with CLAUDE.md, scope.md, submissions tracker,
# findings folder, evidence folder (gitignored), and notes scratchpad.
#
# Usage:
# hunt acme # creates ~/Targets/acme/ with full template
# hunt # shows usage
#
# Customize HUNT_BASE in your environment to override the parent dir:
# export HUNT_BASE="$HOME/security-research/Targets"
#
# Install: source this file from your ~/.zshrc or ~/.bashrc
# echo 'source ~/.claude/scripts/hunt.sh' >> ~/.zshrc
# source ~/.zshrc
# =====================================================================
hunt() {
local target="$1"
local base="${HUNT_BASE:-$HOME/Targets}"
local dir="$base/$target"
if [ -z "$target" ]; then
echo "Usage: hunt <target-name>"
echo "Creates a new engagement folder at \$HUNT_BASE/<target-name>"
echo "Default \$HUNT_BASE is $HOME/Targets"
return 1
fi
if [ -d "$dir" ]; then
echo "Target '$target' already exists at $dir"
echo "cd $dir to continue working on it."
return 0
fi
mkdir -p "$dir/findings" "$dir/evidence"
# ============== CLAUDE.md ==============
# Write heading lines that require $target interpolation explicitly,
# then append the static body via a quoted heredoc (no shell injection risk).
printf '# Engagement: %s\n\n' "$target" > "$dir/CLAUDE.md"
printf '**Target:** %s\n' "$target" >> "$dir/CLAUDE.md"
printf '**Started:** %s\n' "$(date -u +"%Y-%m-%d")" >> "$dir/CLAUDE.md"
cat >> "$dir/CLAUDE.md" <<'CLAUDEMD'
**Platform:** [TBD — Bugcrowd / HackerOne / Intigriti / Immunefi / private]
**Program URL:** [paste the program page URL here]
## Quick context for Claude
This folder is the working directory for a single bug-bounty engagement.
Files in this folder:
- `scope.md` — parsed scope, OOS list, focus areas, bounty bands
- `findings/` — one markdown file per finding draft (naming: `finding-<NN>-<short-name>.md`)
- `submissions.txt` — submission IDs tracker (used for chain cross-references)
- `evidence/` — screenshots, HARs, raw transcripts (gitignored — never share)
- `notes.md` — running notes, leads, dead ends, hypotheses
## Workflow
1. **Plan** — fill in `scope.md` from the program page. Use the `bb-methodology`
and `osint-methodology` skills. Note Focus Areas and Bounty bands.
2. **Recon** — `offensive-osint`, `web2-recon`, and `bb-local-toolkit` for
discovery. Pipe Burp through every browser session for proxy history.
3. **Hunt** — `web2-vuln-classes` (or per-class `hunt-*` skills if installed)
plus `security-arsenal` for payloads.
4. **Validate** — run `/triage` on every lead BEFORE drafting a report.
Apply the 7-Question Gate from `triage-validation`.
5. **Capture evidence** — `evidence-hygiene` BEFORE any screenshot.
Cookies / PII / HARs all redacted per protocol.
6. **Report** — `report-writing` for the body template. `bugcrowd-reporting`
if filing on Bugcrowd (VRT search, severity request, OOS rebuttals).
7. **Track** — append every submitted finding's UUID to `submissions.txt`
so chained reports can cross-reference each other.
## Engagement-specific rules
- All testing on accounts I own.
- Stop immediately on encountering other-user PII; document and report.
- No public disclosure until program explicitly approves.
- Test-account email: `<your-bugcrowdninja-alias>@bugcrowdninja.com`
- Burp proxy capturing through all browser sessions for this target.
## Useful commands during the engagement
- `/scope <asset>` — verify a specific asset is in scope
- `/triage` — quick 7-Question Gate on a finding
- `/validate` — full 4-gate finding validator
- `/report` — draft a submission-ready report
- `/remember` — log a finding to hunt memory
CLAUDEMD
# ============== scope.md ==============
printf '# Scope — %s\n' "$target" > "$dir/scope.md"
cat >> "$dir/scope.md" <<'SCOPEMD'
> Parse this from the program page (Bugcrowd / HackerOne / etc.) before
> doing any active testing. `/scope <asset>` can verify individual assets.
## In scope
- (paste in-scope asset list here)
## Out of scope
- (paste OOS list here, including any bug classes excluded by the program)
## Focus areas
- (paste Focus Areas / accepted impacts list — these are the highest-leverage targets)
## Bounty bands
| Severity | Band |
|---|---|
| P1 (Critical) | |
| P2 (High) | |
| P3 (Medium) | |
| P4 (Low) | |
| P5 (Info) | (often unrewarded) |
## Engagement rules / safe harbor
- (paste any researcher-conduct rules — testing-account requirements,
rate-limit caps, no-DoS clauses, KYC posture, contact email, etc.)
## Account / testing setup
- **Test account email:** (`[email protected]` if Bugcrowd)
- **Test account uid:**
- **Production vs QA:** (which environment is in scope, and any special access notes)
- **Mobile builds:** (Android APK / iOS IPA download URLs if provided)
- **Authentication notes:** (SSO, MFA enrollment status, etc.)
## OOS clauses worth pre-empting in submissions
> Note any OOS clauses that might be miscategorized against your findings.
> Use this to draft `In-scope justification` paragraphs proactively.
- (e.g., "Rate limiting on non-authentication endpoints" — applies to
non-auth endpoints only; verify_password / OTP / token-validate are auth)
- (e.g., "User Enumeration with low-risk info" — applies only when info
enumerated is low-risk; real-name PII is NOT low-risk)
SCOPEMD
# ============== submissions.txt ==============
printf '# Submissions tracker — %s\n' "$target" > "$dir/submissions.txt"
cat >> "$dir/submissions.txt" <<'SUBSEOF'
#
# Format (tab-separated):
# <UUID> <severity> <VRT-or-class> <one-line title>
#
# Use `/remember` after each submission to append the new ID and link
# back to chained primitives.
SUBSEOF
# ============== findings/README.md ==============
printf '# Findings — %s\n' "$target" > "$dir/findings/README.md"
cat >> "$dir/findings/README.md" <<'FINDREADME'
One markdown file per lead.
## Naming
`finding-<NN>-<short-name>.md`
Examples:
- `finding-01-graphql-apq-bypass.md`
- `finding-02-verify-password-no-rate-limit.md`
- `finding-03-update-password-no-stepup.md`
## Per-finding template
Each finding file should have:
1. Status (lead / validated / drafted / submitted / triaged / paid / closed)
2. The finding summary (use `triage-validation`'s 7-Question Gate format)
3. Reproduction steps with exact requests/responses
4. Evidence inventory (path to redacted screenshots in ../evidence/)
5. Severity reasoning + VRT mapping (if Bugcrowd)
6. Submission UUID (once filed)
7. Triager dialogue / status updates
FINDREADME
# ============== notes.md ==============
printf '# Notes — %s\n' "$target" > "$dir/notes.md"
cat >> "$dir/notes.md" <<'NOTESMD'
> Running scratchpad. Leads, hypotheses, dead ends.
## Leads to investigate
## Hypotheses being tested
## Dead ends (so I don't re-investigate)
## Tooling / setup notes for this target
NOTESMD
# ============== .gitignore ==============
cat > "$dir/.gitignore" <<'GITIGNORE'
# Never commit raw evidence — contains live cookies, PII, HARs
evidence/
# Common artifact extensions
*.har
*.png
*.jpg
*.jpeg
*.mp4
*.mov
# Local OS noise
.DS_Store
Thumbs.db
# Secrets / config
.env
*.pem
*.key
GITIGNORE
# ============== recon manifest ingestion (optional) ==============
# If `cbh recon <target>` produced a manifest, seed scope + notes from it.
# Resolve from: 2nd arg → $HUNT_MANIFEST → ./recon/<target>/manifest.json.
local manifest="${2:-${HUNT_MANIFEST:-}}"
if [ -z "$manifest" ]; then
for c in "./recon/$target/manifest.json" "recon/$target/manifest.json" "$HOME/recon/$target/manifest.json"; do
[ -f "$c" ] && manifest="$c" && break
done
fi
if [ -n "$manifest" ] && [ -f "$manifest" ] && command -v python3 >/dev/null 2>&1; then
python3 - "$manifest" "$dir" <<'PYIN'
import json, sys
try:
m = json.load(open(sys.argv[1], encoding="utf-8"))
except Exception as e:
print(f" (recon manifest present but unreadable: {e})"); sys.exit(0)
d = sys.argv[2]
hosts = [a["host"] for a in m.get("assets", []) if a.get("url")]
ranked = m.get("ranked_surface", [])
with open(d + "/scope.md", "a", encoding="utf-8") as f:
f.write("\n## Recon-seeded live hosts (from cbh manifest — VERIFY in-scope before testing)\n\n")
for h in hosts:
f.write(f"- {h}\n")
with open(d + "/notes.md", "a", encoding="utf-8") as f:
f.write("\n## Ranked surface (from recon manifest)\n\n")
for r in ranked:
cls = ", ".join(r.get("bug_classes", [])) or "no class match"
f.write(f"- [{r.get('priority','?')}] {r.get('url') or r.get('host')} - {cls}\n")
print(f" ✓ Ingested recon manifest: {len(hosts)} live hosts -> scope.md, {len(ranked)} ranked -> notes.md")
PYIN
fi
# ============== confirmation ==============
echo "Initialized $dir with CLAUDE.md and engagement template."
echo "cd $dir to start hacking."
echo ""
echo "Files created:"
echo " CLAUDE.md - Claude Code engagement context"
echo " scope.md - parsed scope template (fill from program page)"
echo " submissions.txt - submission UUID tracker"
echo " findings/README.md - findings folder convention"
echo " notes.md - running scratchpad"
echo " evidence/ - gitignored screenshot/HAR folder"
echo " .gitignore - excludes evidence + common secret patterns"
}