| name | agent-scan |
| description | Use Snyk Agent Scan to inspect repo-local skills and optional MCP surfaces for prompt injection, secret handling, and agent supply-chain risks. |
Language and Coding Standards
- Communication: Always talk in Thai when interacting with users.
- Code & Technical Assets: All code, comments, documentation, and technical definitions must be in English.
Agent Scan
Use this skill when the task is to scan agent skills, Codex/Claude/Cursor surfaces, or MCP configuration with Snyk Agent Scan.
When to use
- The user asks to scan agent skills or MCP configs for prompt injection or supply-chain risk.
- The task touches
.agents/skills, .agents/agents, .codex/config.toml, or user-level agent surfaces.
- You need a dedicated agent-surface security pass beyond generic code scanning.
Safety rules
- Start with repo-local skills only before scanning MCP configs.
- Treat MCP config scans as execution-capable operations because upstream may start configured stdio MCP servers.
- Do not use
--dangerously-run-mcp-servers unless the environment is trusted and the user explicitly wants non-interactive execution.
- Require
SNYK_TOKEN; do not print it. The repo wrapper can auto-load it from .env.
Repo workflow
- Confirm
SNYK_TOKEN is present.
- Start with:
make agent-scan
- Escalate scope only when justified:
AGENT_SCAN_MODE=repo make agent-scan
AGENT_SCAN_MODE=home make agent-scan
AGENT_SCAN_MODE=auto make agent-scan
- For machine-readable output:
make agent-scan AGENT_SCAN_ARGS="--json"
- Summarize findings by scope:
- repo skills
- repo MCP config
- user/home surfaces
Output expectations
- State what scope was scanned.
- Call out whether MCP execution consent was involved.
- Separate confirmed findings from skipped surfaces or auth blockers.
- If the scan was not run, report the exact blocker such as missing
SNYK_TOKEN.
