The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to install and maintain shell scripts that are automatically executed by AI agents during tool calls. These hooks (e.g., block-push-to-main.sh, check-pr-review.sh) enforce project policies at the execution layer by intercepting and validating agent commands.
[INDIRECT_PROMPT_INJECTION]: The verify-completion.sh script retrieves unresolved pull request review comments via the GitHub API and displays truncated snippets of these comments to the agent. This represents an ingestion point for untrusted data (comment bodies).
Ingestion points: GitHub PR review comments are fetched via gh api graphql in hooks/verify-completion.sh.
Boundary markers: The retrieved comments are presented as a markdown list in a blocking message without explicit isolation or safety delimiters.
Capability inventory: The script performs read-only operations to GitHub and local state files; its output is used to block task completion via exit codes.
Sanitization: Comment bodies are truncated to the first 80 characters of the first line to reduce the risk of malicious instructions influencing the agent.
[PRIVILEGE_ESCALATION]: The installation scripts (via lib-installer.sh) perform permission modifications (chmod +x) on wrapper scripts in the dispatcher directory to ensure they are executable by the agent. This is an expected operation for an installation utility and is scoped to the skill's own operational scripts.

autonomous-common