The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local package manager commands to gather environment and configuration data.
Evidence: The script scripts/check_js_supply_chain.py uses subprocess.run to call pnpm --version and pnpm config get with a predefined set of policy keys.
Evidence: SKILL.md instructs the agent to run npm view pnpm dist-tags --json to verify the latest stable version of pnpm against the npm registry.
[INDIRECT_PROMPT_INJECTION]: The skill audits external repository content, which serves as a potential attack surface for indirect prompt injection.
Ingestion points: The tool reads configuration and manifest files from the target repository, including package.json, .npmrc, bunfig.toml, and GitHub Action workflow files.
Boundary markers: The skill instructs the agent to interpret findings within a "traffic-light analysis" framework and wait for human approval before proceeding, though the script output does not use specialized markers to delimit untrusted data.
Capability inventory: The skill has the capability to write to the file system and execute package manager commands based on the analysis of these untrusted files.
Sanitization: Data is parsed and filtered for specific security-relevant fields, minimizing the inclusion of arbitrary strings in the final report presented to the agent.

package-security-check