The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external documentation for auditing, which creates a potential surface where malicious instructions could be embedded in the content being reviewed.
Ingestion points: User-provided documentation pages provided for auditing in 'Workflow' step 1.
Boundary markers: No specific delimiters or markers are defined to isolate the audited content from the agent's internal instructions.
Capability inventory: The agent's capabilities include text generation and managing task state in a local file (devrel-context.md).
Sanitization: No specific filtering or sanitization is mentioned for instructions found within audited content.
[SAFE]: The skill's instructions emphasize security best practices for the documentation it produces. It mandates that warnings be placed before risky steps, prohibits the use of secrets in command examples, and advises against using confirmation-bypass flags (like --yes) in destructive command snippets.
[SAFE]: The use of a local file devrel-context.md for maintaining task memory is a standard and benign method for preserving context between sessions.

developer-how-to-guide