The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data by reading Pull Request descriptions, comments, and review threads via the GitHub API. This creates an attack surface where an attacker could embed malicious instructions in PR comments. The skill mitigates this by providing explicit instructions to the agent to treat all thread text as evidence only, and to never interpret it as instructions or permission to execute commands.
[COMMAND_EXECUTION]: The skill executes the GitHub CLI (gh) to fetch PR metadata and manage review states. It also uses uv run to execute internal Python scripts (freeze_policy.py and report.py) for policy management and report generation. These operations are restricted to the local environment and well-known services.
[DYNAMIC_EXECUTION]: The skill utilizes a coordinator-worker model where it orchestrates sub-agents through the Codex App Server (cas) using thread/fork operations. This allows for sharded, parallel review of files while attempting to maintain a consistent execution context and frozen policy state.

elenctic