The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted documentation files (e.g., .md, .mdx) and employs sub-agents with tools such as Bash, Edit, and Write to apply findings. Although it uses a verify-agent for adversarial checks, there are no explicit boundary markers defined in the instructions for sub-agents to separate documentation content from instructions.
Ingestion points: The skill reads documentation files via the Glob and Read tools as defined in agents/inventory-agent.md and agents/docs-ux-audit-agent.md.
Boundary markers: No explicit delimiters or 'ignore instructions' warnings are present in the prompts used by auditing sub-agents.
Capability inventory: The remediation-agent.md and docs-ux-audit-agent.md sub-agents have access to Bash, Edit, Write, and Glob tools.
Sanitization: The skill does not explicitly describe methods for sanitizing or escaping external content before interpolation into sub-agent contexts.
[COMMAND_EXECUTION]: The skill executes a shipped Python script (scripts/ste-lint.py) and calls native repository validation scripts (e.g., those found in package.json) via the Bash tool to perform documentation audits and linting.

technical-documentation