The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions for triaging pull requests direct the agent to execute arbitrary tests and commands provided within the pull request diff to verify functionality. If a contributor submits a malicious pull request, the agent may execute harmful code during this verification step.
[REMOTE_CODE_EXECUTION]: The process of verifying external pull requests requires checking out and running code from untrusted sources. This creates a vector for remote code execution where an attacker can supply malicious logic in a pull request that is then executed by the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external sources, making it vulnerable to indirect prompt injection attacks.
Ingestion points: The skill gathers context from issue and PR bodies, comments, author information, and code diffs as described in the "Triage a specific issue or PR" section of SKILL.md.
Boundary markers: There are no instructions for the agent to use specific delimiters or to disregard potential instructions embedded within the user-provided issue descriptions or comments.
Capability inventory: The agent has the ability to execute shell commands for verification and codebase exploration, write to the filesystem (creating or updating files in .out-of-scope/), and perform network operations (querying and posting comments to a remote issue tracker).
Sanitization: The skill does not define any sanitization, filtering, or validation steps for the external content before it is processed by the model to formulate recommendations or perform actions.

triage