The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from the local codebase to generate documentation, which is a potential surface for indirect injection if the codebase contains adversarial content.
Ingestion points: The skill reads source files, configurations, and data models using grep, find, and the Explore agent to fill template placeholders.
Boundary markers: No specific delimiters or safety instructions are defined to separate ingested codebase context from the agent's instructions during the template population phase.
Capability inventory: The agent uses Read, Write (scoped to docs/adr/), Grep, Glob, and Task tools.
Sanitization: While the filename is sanitized (kebab-case), the content gathered from the codebase is not escaped or validated before being written to the ADR file.

docs-adr