The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides example shell scripts that use claude -p to perform development tasks non-interactively. These scripts also utilize standard command-line tools like the GitHub CLI (gh) for pull request management and CI/CD interactions.
[EXTERNAL_DOWNLOADS]: The documentation references third-party automation tools such as continuous-claude and Ralphinho. It includes explicit security warnings advising users to review external code before installation and avoid piping scripts directly to a shell.
[PROMPT_INJECTION]: The autonomous loop patterns describe workflows that ingest external data, creating an attack surface for indirect prompt injection. 1. Ingestion points: Data enters the context via specification files (e.g., docs/auth-spec.md), RFC/PRD documents, and shared task logs (SHARED_TASK_NOTES.md). 2. Boundary markers: The provided shell examples do not use explicit delimiters or "ignore instructions" wrappers for the ingested content. 3. Capability inventory: The loops use claude -p which can invoke tools for bash execution and file modification. 4. Sanitization: No input validation or sanitization logic is present in the architecture patterns.

autonomous-loops