JPeetz/agent-skills / skills/security/agent-red-teaming
agent-red-teaming
FlaggedUse this skill when adversarially testing an AI agent system — autonomous agents, tool-calling assistants, MCP/A2A fleets, or RAG-backed copilots — against OWASP Agentic Top 10 (ASI01–ASI10) and MITRE ATLAS. Covers threat modeling, capability reconnaissance, prompt injection, tool misuse, memory poisoning, privilege escalation, output-validation bypass, and evidence-backed reporting. Produces reproducible test cases and severity-rated findings, not a jailbreak demo.
| File | Kind | Size |
|---|---|---|
| CHANGELOG.md | markdown | 2.1 KB |
| LICENSE | other | 1.1 KB |
| SKILL.md | skill md | 23.8 KB |
| evals/evals.json | config | 9.9 KB |
| references/attack-payloads.jsonl | other | 22.8 KB |
| references/owasp-asi-mapping.md | markdown | 15.7 KB |
| scripts/score_findings.py | script | 7.0 KB |
CHANGELOG.md
View on GitHub# Changelog — Agent Red Teaming
## [1.0.0] — 2026-11-14
- Initial release from Skill Foundry
- Nine-step engagement process: scope/safety rails → threat model → capability
recon → prompt injection → tool misuse → memory poisoning → privilege
escalation → output-validation bypass → score/report/retest
- OWASP Agentic Top 10 (ASI01–ASI10) coverage with per-category test strategies,
evidence rules, and primary controls
- MITRE ATLAS tactic mapping per attack class, in the attack catalog and in
`references/owasp-asi-mapping.md`
- Attack vectors: direct/indirect prompt injection across 7 carriers, tool
misuse (traversal, SSRF, parameter pollution, chaining, scale abuse, confused
deputy), memory and context poisoning including agent-writable config/skill
surfaces, privilege and identity abuse (incl. inter-agent spoofing and
unattended self-escalation), and output-validation bypass (markup/ANSI,
schema breaks, downstream sink injection, beacon exfiltration)
- Canary-first evidence model: no finding without a trace excerpt and a canary
- Severity rubric scored on impact × reachability × precondition, with explicit
anti-inflation rules
- Verdict taxonomy `PASS | FAIL | PARTIAL | BLOCKED-ENV` (no blank verdicts)
- 3 worked examples (RAG support copilot exfiltration; coding-agent credential
read via repo-planted instructions; research-fleet memory poisoning with
circular corroboration)
- 11 anti-patterns and a 17-item pre-completion verification checklist
- `references/owasp-asi-mapping.md` — ASI01–ASI10 table with failure description,
test strategy, required evidence, and primary control layer
- `references/attack-payloads.jsonl` — 24 machine-readable probe records plus a
schema record, usable directly as a test backlog
- `scripts/score_findings.py` — stdlib-only validator/aggregator for
`tests.jsonl`: schema checks, coverage per ASI category, severity rollup, and
a markdown summary ready for the report appendix
- `evals/evals.json` — 8 eval cases covering trigger accuracy, near-misses, and
expected diagnostic content
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |