The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a robust security model by explicitly instructing the agent to treat all GitHub-authored content—including issue titles, bodies, and comments—as untrusted data. It provides specific isolation markers and strictly forbids following any instructions or commands found within external data.
[COMMAND_EXECUTION]: The skill utilizes standard command-line tools git and gh (GitHub CLI) to perform repository operations and API interactions. These subprocess calls are used legitimately to inspect the local checkout and fetch issue metadata.
[EXTERNAL_DOWNLOADS]: Fetches issue data from GitHub.com, which is a well-known and trusted service. Network operations are performed using the official GitHub CLI, ensuring communications are scoped to the intended platform.
[DATA_EXFILTRATION]: There is no evidence of unauthorized data transmission. All external communication is directed towards GitHub's official API for the purpose of issue management.
[REMOTE_CODE_EXECUTION]: The skill includes a mechanism to execute a validation script from a related senior planning skill. This dynamic execution is a controlled cross-skill interaction used to verify plans against higher-level requirements and is performed locally using the current Python interpreter.
[SAFE]: The workflow incorporates multiple 'fail-closed' security gates, such as checking for clean git checkouts and requiring explicit user authorization before performing any state-changing operations like branch creation or commenting.

github-issue-planner