The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's publish mode generates a GitHub Actions workflow that relies on a third-party repository (Andrew-Chen-Wang/github-wiki-action@v5). This introduces a dependency on code from an external source not identified as a trusted vendor or well-known service.
[REMOTE_CODE_EXECUTION]: The installation of the Andrew-Chen-Wang/github-wiki-action@v5 within a GitHub Actions workflow represents the execution of remote code managed outside the skill's environment.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run an allowlist of read-only probes (e.g., --help, git log, ls-remote) to verify claims against the codebase. These operations require user consent and are restricted to specific side-effect-free commands.
[DYNAMIC_EXECUTION]: The publish mode dynamically constructs a GitHub Actions workflow file (.github/workflows/publish-wiki.yml) that includes a Perl script for URL transformation and flattening.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository (manifests, source code, CI configurations) to generate documentation, creating a potential attack surface for indirect prompt injection.
Ingestion points: As described in SKILL.md and modes/init.md, the skill reads manifests, CLI surfaces, routes, and environment variables across the project.
Boundary markers: The skill relies on "Grounding" rules and a human-in-the-loop "Propose the map" consent gate defined in SKILL.md to review all generated content.
Capability inventory: The skill uses Bash, Write, Edit, Grep, Glob, and Read tools across all operational modes.
Sanitization: Implements "Writing standards" and a verification mechanism against the live repository to ensure all documented features are present in the code.
[PERSISTENCE]: The skill establishes a persistent automated process by installing a GitHub Actions workflow that synchronizes documentation to the GitHub Wiki on every push to the repository.

wikikit