The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes repository source code, manifests, and existing markdown files to generate documentation, which presents a surface for indirect prompt injection. The skill mitigates this risk through explicit instructions for grounded writing and a mandatory 'de-slopify' pass to manually recast generated content.
Ingestion points: Repository source code, configuration files (e.g., package.json, go.mod), and existing documentation in SKILL.md and references/readme-craft.md.
Boundary markers: Instructions emphasize grounding output in actual repository intent and verified executable behavior.
Capability inventory: Capabilities include writing documentation files, creating evidence directories, and executing local shell scripts for repository auditing.
Sanitization: Employs a 'de-slopify' pass (references/de-slopify.md) that requires manual line-by-line recasting of prose to remove prefabricated AI patterns and ensure factual accuracy.
[COMMAND_EXECUTION]: The skill executes bundled shell scripts, including scripts/audit-oss-docs.sh and scripts/validate.sh, to perform documentation coverage audits and verify the structural integrity of the skill package and linked resources.
[DYNAMIC_EXECUTION]: The scripts/validate.sh utility executes an inline Python block to dynamically validate that all internal documentation links resolve correctly to local files within the repository context.

doc