The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell command to verify the integrity of the documentation site after updates.
Evidence: Phase 4 (Verify) in SKILL.md triggers the command uv run mkdocs build --strict to check for build errors and broken links.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the repository to influence its logic and file-writing actions.
Ingestion points: Reads source code from synapse/, plugins/, and pyproject.toml, as well as documentation from guides/, docs/, and README.md to identify mapping relationships defined in references/source-site-mapping.md.
Boundary markers: The skill does not define explicit delimiters or use system-level instructions to ignore potential malicious prompts embedded within the source files it reads.
Capability inventory: The skill has the ability to read and write files in the site-docs/ directory, update mkdocs.yml and changelog.md, and execute shell commands via uv.
Sanitization: The risk is mitigated by a mandatory human-in-the-loop approval step in Phase 3, where proposed changes must be presented to and approved by the user before being applied to the filesystem.

github-pages-sync