The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including Markdown specifications, GitHub issues, and task descriptions which could contain malicious instructions designed to influence agent behavior during implementation or verification.
Ingestion points: The skill accepts external input via the argument-hint <specs, tasks, issues, PRs, or milestone> and reads repository instructions and requirements.
Boundary markers: There are no explicit instructions or delimiters defined to separate untrusted task content from the agent's internal operational logic, nor instructions to disregard instructions found within the data.
Capability inventory: The skill possesses significant capabilities including reading source code, creating branches and worktrees, writing files (implementing tasks), committing code, pushing to remote repositories, and executing test suites via the /test tool.
Sanitization: The skill lacks specified mechanisms for validating or sanitizing external input before it is used to direct implementation or verify behavior.
[COMMAND_EXECUTION]: The workflow involves several operations that execute commands within the user's environment, including branch management, git operations (commit, push, merge), and the invocation of external tools like /test and /review for code verification.

task-to-pr