The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub, including issue descriptions, PR comments, and repository files (README.md, CONTRIBUTING.md, AGENTS.md). This content could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: GitHub issue timelines, PR lists, repository documentation files, and CI logs (SKILL.md).
Boundary markers: The instructions emphasize that all external write operations must wait for explicit user confirmation.
Capability inventory: Access to GitHub CLI (gh) and the ability to run local shell commands for testing and building projects.
Sanitization: Relies on the user to review the proposed changes and the results of verification commands before execution.
[COMMAND_EXECUTION]: The skill is designed to run project-specific verification commands (such as npm test, lint, or build scripts) based on the configuration found in external repositories. Executing commands defined in untrusted repositories carries an inherent risk, though it is a standard part of the software contribution workflow.
Evidence: Instructions in SKILL.md direct the agent to "运行可执行的测试、lint、格式、构建或链接检查" (run executable tests, linting, formatting, builds, or link checks) based on the repository's .github/workflows/ and project-level configurations.

contributor