The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data, including repository code, schemas, test results, and command outputs, which inherently creates a surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: SKILL.md (Section: 'Establish the reader and evidence'
Step 1: 'Read current code, schemas, tests, command output, and the relevant diff.').
Boundary markers: The skill mentions reading a local 'AGENTS.md' for terminology and scope, but does not specify explicit delimiters or 'ignore' instructions for the data it processes.
Capability inventory: The skill focus is documentation, though the underlying agent likely possesses file-read and potentially shell-execution capabilities to gather the evidence mentioned.
Sanitization: No specific technical sanitization or filtering of external content is described; the skill relies on the agent's 'verification pass' and 'plain-language pass' to maintain accuracy.
[SAFE]: The skill documentation explicitly promotes security best practices.
Evidence: SKILL.md (Section: 'Make examples usable'
'Never include real credentials.').
Evidence: SKILL.md (Section: 'Establish the reader and evidence'
'Verify claims, examples, defaults, paths, and failure behavior.').

technical-writing