The Agent Skills Directory

[COMMAND_EXECUTION]: Provides multiple shell script templates and command definitions (e.g., daily-dev.sh) that use claude -p to automate multi-step development workflows, including code implementation, testing, and Git operations.
[PROMPT_INJECTION]: The skill describes architectures (Infinite Agentic Loop, Continuous Claude PR Loop, and Ralphinho) designed to ingest and act upon untrusted external data such as specifications, RFCs, and shared task notes.
Ingestion points: Processes content from docs/auth-spec.md, specs/component-spec.md, and SHARED_TASK_NOTES.md as primary logic drivers in SKILL.md.
Boundary markers: The provided script examples and command definitions do not include boundary delimiters or instructions to the agent to ignore embedded commands within the processed data.
Capability inventory: The loops rely on claude -p, which possesses full capabilities for file modification, command execution via the Bash tool, and network access (e.g., through gh CLI commands mentioned in templates).
Sanitization: No sanitization or validation logic is demonstrated for the external content before it is interpolated into agent prompts.
[EXTERNAL_DOWNLOADS]: References external third-party tools and repositories for loop orchestration, specifically continuous-claude and Ralphinho. The skill correctly includes a warning recommending that users review the external code before execution.

autonomous-loops