The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the current development environment to create reusable skill proposals. While it includes strong directives for manual and automated review, the ingestion of external data combined with file-writing capabilities creates a potential surface for indirect injection.
Ingestion points: Processes task context, relevant diffs, docs, command output, and raw logs (mentioned in redaction rules) from the development environment.
Boundary markers: The skill relies on natural language instructions for the agent to 'inspect' and 'classify' data but does not define strict technical delimiters (e.g., XML tags or specific block markers) for the ingested content in its prompt templates.
Capability inventory: Capable of drafting and writing SKILL.md, agents/openai.yaml, and executable scripts/ to the local filesystem or proposing upstream patches via GitHub Pull Requests.
Sanitization: Explicitly mandates a redaction pass to remove secrets, credentials, private customer data, environment-specific paths, and private URLs before any drafting or delegation to subagents occurs.

skill-capture