JN13012/ai-pentest-harness
Claude Code profile for authorized web pentests: specialized agents, with scope and evidence enforced by code. Based on a course by Lucas Romano.
ARCHETYPE
Fort Knox
Deny lists, pre-tool hooks, sandboxing. Nothing touches prod without a signature.
GUARDRAILS
4/5
Blocks destructive commands · Protects secrets · Pre-tool screening hook · No YOLO mode · details
Quality 56/100 Guardrails v2
Copy this rig
# review before running: this installs third-party code
$ npx degit JN13012/ai-pentest-harness/.claude ./rig-ai-pentest-harness # inspect, then merge into .claude/ MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit JN13012/ai-pentest-harness/.claude/skills/recon .claude/skills/recon $ curl -fsSL --create-dirs -o .claude/agents/artifact-reviewer.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/artifact-reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/auth-hunter.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/auth-hunter.md $ curl -fsSL --create-dirs -o .claude/agents/chain.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/chain.md $ curl -fsSL --create-dirs -o .claude/agents/class-hunter.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/class-hunter.md $ curl -fsSL --create-dirs -o .claude/agents/injection-hunter.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/injection-hunter.md $ curl -fsSL --create-dirs -o .claude/agents/recon.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/recon.md $ curl -fsSL --create-dirs -o .claude/agents/review.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/review.md $ curl -fsSL --create-dirs -o .claude/agents/ssrf-hunter.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/ssrf-hunter.md $ curl -fsSL --create-dirs -o .claude/agents/xss-hunter.md https://raw.githubusercontent.com/JN13012/ai-pentest-harness/main/.claude/agents/xss-hunter.md
Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"permissions": {
"deny": [
"Bash(rm:*)",
"Bash(git push:*)",
"Bash(python:*)",
"Bash(python3:*)",
"Bash(node:*)",
"Bash(ruby:*)",
"Bash(perl:*)",
"Bash(php:*)",
"Bash(sh:*)",
"Bash(bash:*)",
"Bash(zsh:*)",
"Bash(powershell:*)",
"Bash(pwsh:*)",
"Bash(cmd:*)",
"Bash(xargs:*)",
"Bash(env:*)",
"Bash(printenv:*)",
"Bash(wget:*)",
"Bash(nc:*)",
"Bash(ncat:*)",
"Bash(socat:*)",
"Bash(ssh:*)",
"WebFetch",
"Read(//proc/**)",
"Read(**/.env)",
"Read(**/.env.*)",
"Read(**/*.env)",
"Read(**/credentials*)",
"Read(**/*credentials*)",
"Read(**/.auth/**)",
"Read(**/.aws/**)",
"Read(~/.aws/**)",
"Read(**/.azure/**)",
"Read(~/.azure/**)",
"Read(~/.config/gcloud/**)",
"Read(**/.kube/config)",
"Read(~/.kube/config)",
"Read(**/.docker/config.json)",
"Read(~/.docker/config.json)",
"Read(**/.ssh/**)",
"Read(~/.ssh/**)",
"Read(**/.netrc)",
"Read(~/.netrc)",
"Read(**/.git-credentials)",
"Read(**/.npmrc)",
"Read(**/.pypirc)",
"Read(**/.claude/.credentials.json)",
"Read(~/.claude/.credentials.json)",
"Read(**/*token*.json)",
"Read(**/*.token)",
"Read(**/*.pem)",
"Read(**/*.p12)",
"Read(**/id_rsa*)",
"Read(**/id_ed25519*)",
"Edit(./scope.md)",
"Write(./scope.md)",
"Edit(./CURRENT_ENGAGEMENT.md)",
"Write(./CURRENT_ENGAGEMENT.md)",
"Edit(./engagements/**/evidence/**)",
"Write(./engagements/**/evidence/**)",
"Edit(./engagements/**/scope.md)",
"Write(./engagements/**/scope.md)",
"Edit(./engagements/**/scope.json)",
"Write(./engagements/**/scope.json)",
"Edit(./.claude/**)",
"Write(./.claude/**)",
"Edit(**/.env*)",
"Write(./config/**)",
"Edit(./config/**)",
"Write(./lib/**)",
"Edit(./lib/**)",
"Write(./scripts/**)",
"Edit(./scripts/**)",
"Write(./prompts/**)",
"Edit(./prompts/**)",
"Write(./third_party/**)",
"Edit(./third_party/**)",
"Write(./engagements/**/analysis/plan.json)",
"Edit(./engagements/**/analysis/plan.json)",
"Write(./engagements/**/artifacts/**)"
]
},
"hooks": {
"PreToolUse": [
{
"matcher": "^Bash$|Claude_Browser__(navigate|javascript_tool)$|^WebFetch$",
"hooks": [
{
"type": "command",
"command": "node .claude/hooks/scope-check.js"
}
]
}
]
}
} Skills (1)
Subagents (9)
| artifact-reviewer | Reviews one planned and hash-verified sanitized artifact for one domain; records source-backed candidates and proof gaps without making network calls. |
| auth-hunter | Tests authentication mechanism failures -- JWT alg confusion/none, expired or replayed token acceptance, password reset flaws, missing rate limiting. Use after recon has produced surface.json. OWASP T |
| chain | Joins confirmed findings across classes on capability_granted -> capability_required to build and validate exploit chains (e.g. an auth-bypass primitive feeding a code-execution primitive). Runs autom |
| class-hunter | Executes exactly one planned HTTP vulnerability class on one surface, loading only its selected method and checking its prerequisites. |
| injection-hunter | Tests injection -- SQL, NoSQL, OS command, template, LDAP -- using a differential against a stable baseline. Covers command-injection-to-RCE. Use after recon has produced surface.json. OWASP Top 10 A0 |
| recon | Maps the attack surface of the active engagement's target -- endpoints, parameters, object identifiers, and the auth model. Runs once at the start of an engagement, before any class-hunter subagent. U |
| review | Adversarially re-verifies every finding and chain against the evidence log, checks OWASP/API Top 10 coverage, and drafts the final report. Runs last, automatically, once recon/class-hunters/chain have |
| ssrf-hunter | Tests server side request forgery against every parameter taking a URL, hostname, path, webhook, or callback. Use after recon has produced surface.json. OWASP Top 10 A10, API Top 10 API7. |
| xss-hunter | Tests cross-site scripting -- reflected, stored, and DOM-based -- by finding rendering contexts and proving script execution, not just reflection. Use after recon has produced surface.json. OWASP Top |
Hooks (2)
| event | matcher | runs |
|---|---|---|
| PreToolUse | ^Bash$|Claude_Browser__(navigate|javascript_tool)$|^WebFetch$ | node .claude/hooks/scope-check.js |
| PostToolUse | ^Bash$|Claude_Browser__(navigate|javascript_tool|computer|read_page|get_page_text)$|^WebFetch$ | node .claude/hooks/evidence-log.js |
Permissions
deny (81)
Bash(rm:*)
Bash(git push:*)
Bash(python:*)
Bash(python3:*)
Bash(node:*)
Bash(ruby:*)
Bash(perl:*)
Bash(php:*)
Bash(sh:*)
Bash(bash:*)
Bash(zsh:*)
Bash(powershell:*)
Bash(pwsh:*)
Bash(cmd:*)
Bash(xargs:*)
Bash(env:*)
Bash(printenv:*)
Bash(wget:*)
Bash(nc:*)
Bash(ncat:*)
Bash(socat:*)
Bash(ssh:*)
WebFetch
Read(//proc/**)
Read(**/.env)
Read(**/.env.*)
Read(**/*.env)
Read(**/credentials*)
Read(**/*credentials*)
Read(**/.auth/**)
Read(**/.aws/**)
Read(~/.aws/**)
Read(**/.azure/**)
Read(~/.azure/**)
Read(~/.config/gcloud/**)
Read(**/.kube/config)
Read(~/.kube/config)
Read(**/.docker/config.json)
Read(~/.docker/config.json)
Read(**/.ssh/**)
Read(~/.ssh/**)
Read(**/.netrc)
Read(~/.netrc)
Read(**/.git-credentials)
Read(**/.npmrc)
Read(**/.pypirc)
Read(**/.claude/.credentials.json)
Read(~/.claude/.credentials.json)
Read(**/*token*.json)
Read(**/*.token)
Read(**/*.pem)
Read(**/*.p12)
Read(**/id_rsa*)
Read(**/id_ed25519*)
Edit(./scope.md)
Write(./scope.md)
Edit(./CURRENT_ENGAGEMENT.md)
Write(./CURRENT_ENGAGEMENT.md)
Edit(./engagements/**/evidence/**)
Write(./engagements/**/evidence/**)
ask (0)
—
allow (16)
Bash(curl:*)
Bash(jq:*)
Bash(git status)
Bash(git diff:*)
Bash(git log:*)
Read(./**)
Write(./engagements/**)
Edit(./engagements/**)
mcp__remote-devices__Claude_Browser__navigate
mcp__remote-devices__Claude_Browser__computer
mcp__remote-devices__Claude_Browser__read_page
mcp__remote-devices__Claude_Browser__get_page_text
mcp__remote-devices__Claude_Browser__javascript_tool
mcp__remote-devices__Claude_Browser__find
mcp__remote-devices__Claude_Browser__form_input
mcp__remote-devices__Claude_Browser__tabs_context
Similar rigs
Ian-Lo/application-doc-toolkit
Fact-governance for LLM-written documents: claims trace to a verified source entry, a scripted lint gates every draft, reviewers hold no write access. Applied to resumes and cover letters.
Pragmatist 8.4k tok · · Q18
gavraz/recon
tmux-native dashboard for managing Claude Code agents
Minimalist 797 tok · · Q0
AlmanacCode/codealmanac
A codebase wiki for AI coding agents. Captures what the code can't say: decisions, flows, invariants, gotchas.
Pragmatist 4.1k tok · · Q8
tony/claude-code-riper-5
Claude Code (Sub-agent, Custom Commands) for RIPER-5
Pragmatist 3.0k tok · · Q20