The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run various commands, including package manager operations (pnpm, npm), Git operations, and GitHub CLI (gh) tasks. It also directs the agent to execute Node.js scripts located within the repository checkout, such as node scripts/run-vitest.mjs and node scripts/test-docker-all.mjs.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes external, potentially untrusted data alongside its command execution capabilities.
Ingestion points: The agent is instructed to read CI failure logs, summary.json, failures.json, and contributor-submitted source code during PR testing to diagnose failures.
Boundary markers: The skill contains explicit safety directives: "Never execute untrusted contributor/fork code, wrappers, config locally" and "Contributor-controlled code must use the repository's approved untrusted-source route, not a credentialed local run."
Capability inventory: The workflow utilizes powerful tools including pnpm, npm, node, and gh to perform validation and testing tasks.
Sanitization: No specific data sanitization or escaping mechanisms for the diagnostic logs are described; the security model relies on directing untrusted execution to isolated, secretless CI environments.

openclaw-testing