security-research — Claude Code skill: what it does, how to install
security-research runs a parallel security audit with five agents: three hunt for vulnerabilities, two build PoCs. Useful for a PR, a diff range or a release candidate.
no license 508 installs ★ 71,048 updated 2026-10-25
~/.claude/skills/security-research/SKILL.mdWhat security-research does
The skill spins up a five-agent team through team_create. Three hunters split the work: surface-hunter maps entry points and trust boundaries, auth-data-hunter covers authorization, data isolation and injection, and runtime-supply-hunter looks at the filesystem, subprocesses, dependencies, hooks and MCP. Two PoC engineers, poc-engineer-a and poc-engineer-b, try to reproduce what the hunters report. The second one also tries to knock findings down.
A run goes through scoping, a hunter pass, a PoC pass, a cross-check among all five agents, and a report. The report opens with a verdict: PASS, PASS WITH FINDINGS or BLOCK. A finding gets no severity unless someone can name an attack path. Destructive exploits against live services are off the table, so proofs run on local fixtures or as static arguments.
It needs team_* tools. If they are missing, the skill halts and asks you to set team_mode.enabled: true in .omo/omo.jsonc and restart opencode. It runs only in the main session. With no target given, it audits the current repository and the branch diff.
What it helps with
- check a PR or branch diff for exploitable holes before merging
- separate a vulnerability with an attack path from generic hardening advice
- get a small reproducible PoC for a suspected bug
- find authorization bypasses and data leaks between tenants
- review hooks, MCP and archive extraction for path traversal and command injection
Example prompts for the agent
- Run security-research on the diff of my current branch against its base
- Use security-research on the release candidate, no network calls and no destructive tests
- Audit this PR with security-research and give me the verdict first
How to install security-research
Manually: put the files in place
~/.claude/skills/security-research/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/security-research/
Manually: put the files in place
.agents/skills/security-research/SKILL.mdFor all projects use ~/.agents/skills/
Manually: put the files in place
.cursor/skills/security-research/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install code-yeongyu/oh-my-openagent packages/skills-loader-core/src/features/builtin-skills/security-research2Manually: put the files in place
.github/skills/security-research/SKILL.mdPlatform CLI
gemini skills install https://github.com/code-yeongyu/oh-my-openagent --path packages/skills-loader-core/src/features/builtin-skills/security-researchManually: put the files in place
.devin/skills/security-research/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about security-research
How do I install security-research in Claude Code?
Run npx skills add code-yeongyu/oh-my-openagent --skill security-research -a claude-code. The skill file sits at ~/.claude/skills/security-research/SKILL.md.
What does security-research do?
It assembles three vulnerability hunters and two PoC engineers, runs the audit in phases, and returns a report with a verdict, a findings table, CWE entries and minimal fixes.
What does security-research need to run?
Team-mode tools (team_*). Set team_mode.enabled: true in .omo/omo.jsonc and restart opencode. The skill also has to start from the main session, not from a background subagent.