The Agent Skills Directory

[COMMAND_EXECUTION]: The skill installs and manages several shell hooks that are designed to execute automatically during git operations (commit, push) and AI agent tool use. These hooks are used to enforce a specific development workflow, such as blocking direct pushes to the main branch and ensuring test plans are created.
[INDIRECT_PROMPT_INJECTION]: The skill includes utilities that ingest data from external sources like GitHub issue bodies (mark-issue-checkbox.sh) and pull request review threads (verify-completion.sh, resolve-threads.sh). This data is used to determine workflow state or modify task completion status.
Ingestion points: GitHub issue bodies via itp_read_task and PR review thread details via chp_review_threads.
Boundary markers: The skill relies on structured JSON communication for hook data but does not use explicit boundary markers for the unstructured text data it processes.
Capability inventory: Uses gh api to patch issue bodies and resolve review threads; performs various git operations (commit, push, rebase, worktree); writes to local state files in .agents/state/.
Sanitization: Employs sed for character escaping in mark-issue-checkbox.sh and relies on jq for safe JSON handling.
[DYNAMIC_EXECUTION]: The install-codex-hooks.sh script executes an embedded Python script at install time to safely parse and migrate TOML configurations. Additionally, the core library (hooks/lib.sh) uses complex awk programs to perform static analysis of shell commands at runtime to detect git operations.

autonomous-common