The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from remote issue trackers and pull requests, creating a vulnerability surface where external actors could attempt to influence the agent's behavior.
Ingestion points: Remote task tracker (GitHub/GitLab issues and PRs) and local task bodies, specifically reading comments, author notes, and diffs as outlined in SKILL.md.
Boundary markers: The skill does not specify the use of delimiters or specific safety instructions to prevent the agent from executing instructions found within the untrusted external data.
Capability inventory: The skill requires the ability to execute shell commands to "reproduce a bug" and "run relevant tests or commands" for PR verification (evidence in SKILL.md step 3).
Sanitization: No explicit sanitization or filtering of external task content is mentioned before it is processed or used to trigger command execution.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to execute code and tests that are directly influenced by or derived from untrusted external sources (bug reports and PR diffs).
Evidence: SKILL.md step 3 under 'Triage a task or PR' instructs the agent to "reproduce a bug from the reporter's steps" and "run relevant tests or commands" for a PR diff.

triage