The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the agent-spec CLI tool to perform operations like init, seed, query, and inventory on the repository code and wiki directory. It also executes a local shell script scripts/docs-lint.sh for documentation linting.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for ingesting data from Markdown files located in .agent-spec/wiki/. These files contain YAML frontmatter with fields such as external_sources and source_files.
Ingestion points: Articles and flow descriptions located in .agent-spec/wiki/projects/ and .agent-spec/wiki/flows/ are read by the agent to build project maps and perform checks.
Boundary markers: The skill requires strict YAML frontmatter validation and explicitly rejects symlinks during source and article traversal to prevent path traversal or unintended data access.
Capability inventory: The skill can execute local commands (agent-spec, bash) and write generated architecture files (JSON and Mermaid formats) to the local .agent-spec/wiki/architecture/ directory.
Sanitization: The skill performs field validation (e.g., rejecting empty or duplicate keys) and explicitly states that it does not perform external repository scans for external_sources URLs by default.

agent-spec-wiki