SploitAgent

A security-skills library for AI agents — offensive and defensive, for authorized pentest, bug-bounty, and blue-team work.

_{Documentation · Catalog · Coverage · Using · Contributing · Roadmap}

---

Overview

SploitAgent is a portable library of 121 security skills an AI agent loads on demand. It is not a scanner or a framework — it is the operating knowledge of the work itself (find the issue, prove the impact, escalate, pivot, report, and defend) captured as small, trigger-tagged SKILL.md files in the standard Agent-Skills format.

Any agent — Claude Code, Codex, Gemini, or a local model — reads a skill's trigger line, loads the one that matches the task, and follows its method. The skills are plain Markdown: no runtime, no build step, no lock-in. Each is mapped to OWASP, MITRE ATT&CK, and CWE, so coverage is measurable (COVERAGE.md).

Design principles

• Knowledge over tooling. Skills encode how to reason — the mechanism, the exact command, the failure modes — which outlasts any individual tool or model version.
• Offense and defense. Every offensive class has its defensive counterpart: detection, hardening, DFIR.
• Consistency by contract. A JSON schema (schemas/skill.schema.json) validates every skill in CI.
• Distilled from real work. Techniques are mined from write-ups, disclosed reports, and authorized engagements — generalized to the underlying class, never copied as walkthroughs (CONTRIBUTING.md).
• Authorized use, enforced first. Every engagement begins with scope confirmation.

Quick start

Claude Code

Open it in the repository — CLAUDE.md instructs the agent on how to use the library and to confirm scope first:

git clone https://github.com/NoorQureshi/SploitAgent && cd SploitAgent
claude

Or expose the skills to every project once:

git clone https://github.com/NoorQureshi/SploitAgent
ln -s "$PWD/SploitAgent/skills" ~/.claude/skills/sploitagent

Codex, Gemini, and other agents

Clone the repository and open your agent inside it — it reads AGENTS.md, the tool-agnostic operating guide. For a one-off, load a single SKILL.md into context.

git clone https://github.com/NoorQureshi/SploitAgent && cd SploitAgent

Local models and other frameworks

There is no runtime to install. Give the agent read access to skills/ and AGENTS.md, or load the specific SKILL.md for the task. Per-agent instructions, including an Ollama Modelfile that embeds the operating guide, are in docs/USING.md.

Operating model

An agent works a target through a consistent loop and keeps its work in a per-engagement workspace. The full method is in methodology.md; the agent guide is AGENTS.md.

1. Scope       confirm authorization; record scope.txt     → tradecraft-scope-roe
2. Recon       map the attack surface                       → recon-*
3. Attack      route by domain (web / api / cloud / ad / …) → web-*, api-*, cloud-*, …
4. Foothold    drive a weakness to proven impact            → exploit-chaining
5. Escalate    privilege escalation and lateral movement    → privesc-*, ad-*, network-*
6. Report      findings with severity and evidence          → reporting-*
7. Defend      convert findings into detections             → defense-*

engagements/<target>/
  scope.txt   roe.md   notes.md   findings/   loot/     # git-ignored — never committed

Skill coverage

121 skills across 19 domains. The complete index is in CATALOG.md.

Domain	Skills	Coverage
web	36	XSS, SQLi, SSRF, SSTI, IDOR, XXE, CSRF, CORS, LFI, deserialization, OAuth, SAML, request smuggling, prototype pollution, cache poisoning, host-header, clickjacking, WebSocket, race conditions, business logic, file upload, JWT, account takeover, dependency confusion, client-side signing reversal, authenticated session handling, Python sandbox escape, Cypher injection, JDBC/connection-string RCE
ai-ml	9	Prompt injection, jailbreaks, RAG poisoning, model extraction, agent/tool and MCP abuse, insecure output handling, supply chain, unbounded consumption
cloud	9	IMDS credential theft, object-storage exposure, Kubernetes, container escape, exposed Docker/daemon API abuse, IAM privilege escalation, registries, GCP, Azure / Entra ID
api	8	BOLA/BFLA, GraphQL, gRPC, mass assignment, authentication attacks, fuzzing, version drift, NoSQL injection
recon	8	Subdomain enumeration, DNS analysis, content and JS discovery, OSINT, cloud-asset discovery, service enumeration, tech-stack fingerprinting
defense	6	Detection engineering (Sigma/ATT&CK), hardening baselines, DFIR triage, threat modeling, log analysis, purple teaming
code-review	6	Methodology, dangerous-sink catalog, secrets detection, CI/CD security, Python, Node.js
mobile	5	Android and iOS assessment, certificate-pinning bypass, deep-link abuse, WebView abuse
ad	5	Kerberoasting / AS-REP, ADCS (ESC1–8), ACL/DACL abuse, Kerberos delegation abuse (RBCD / S4U / coercion→relay), pivoting arsenal
network	6	Service attacks, pivoting and tunneling, NTLM coercion and relay, password spraying and credential stuffing, perimeter appliance and VPN offensive, hash and credential cracking
wireless	2	WPA2-PSK handshake/PMKID capture and cracking, evil-twin / rogue-AP enterprise (PEAP-MSCHAPv2) credential harvesting
privesc	3	Linux arsenal, GTFOBins (sudo/SUID/capabilities), Windows token impersonation
exploit-dev	2	Exploit chaining and impact amplification, PoC development
reverse-engineering	3	Native binary triage, deobfuscation (packed/JS/WASM/JSVMP), firmware extraction and analysis
payloads	2	WAF/filter bypass, XSS polyglots
reporting	3	Finding triage and validation, bug-bounty write-up, penetration-test report
automation	2	Recon pipelines, custom nuclei templates
tradecraft	2	Scope and rules of engagement, complex multi-stage engagements
social-eng	4	Authorized human-factor testing: methodology, phishing, vishing/pretexting, physical assessment

Skill format

Each skill is a single SKILL.md: a frontmatter block whose description is the load trigger, and a body that explains the mechanism rather than pasting a payload.

---
name: web-ssrf
description: Discover and escalate Server-Side Request Forgery. Load when the app
  fetches a URL you influence: webhooks, "import from URL", PDF/image rendering, …
domain: web
type: technique
modes: [pentest, bugbounty]
owasp: [A10:2021-SSRF]
cwe: [CWE-918]
---
## When it applies · Why it works · Method (exact commands) · Gotchas · Verify success

Authorized use

SploitAgent is intended solely for authorized security work: penetration-testing engagements under a signed scope, bug-bounty programs whose scope covers the target, and defensive assessment of systems you own or operate. The first skill loaded in any engagement is tradecraft-scope-roe, which requires an explicit authorization envelope before any activity. Do not use it against systems you are not authorized to test.

Contributing

Contributions are single Markdown files — no code required:

cp skills/_templates/technique.md skills/<domain>/<slug>/SKILL.md
python3 tools/catalog.py            # validate frontmatter and regenerate the indexes

Wanted skills are listed in ROADMAP.md (entries marked as good first contributions). The full guide, house style, and PR checklist are in CONTRIBUTING.md. Every pull request is schema-validated by CI.

Repository layout

skills/<domain>/<slug>/SKILL.md   the library
AGENTS.md · CLAUDE.md             operating guide read by any in-repo agent
methodology.md                    engagement loop, scope rule, note-taking standard
schemas/skill.schema.json         the skill contract (validated in CI)
tools/catalog.py                  validation and index generation
CATALOG.md · COVERAGE.md          generated indexes
docs/                             usage guide and project site

License

MIT.