The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from potentially untrusted external sources, such as local repositories, trackers, and documentation.
Ingestion points: The instructions guide the agent to "open the repository", "probe active branches", and analyze the "issue tracker or the docs" to find facts and decisions.
Boundary markers: The skill includes strong constraints to minimize hallucination or blind obedience, such as "Never invent a fact", "No technology is proposed before the verdict", and "A decision without a concrete value is not decided".
Capability inventory: The agent has the capability to read local files, execute git commands, and write design artifacts in markdown format.
Sanitization: While there is no explicit data sanitization, the use of a highly rigid and structured output template (references/document-format.md) significantly limits the potential for external data to hijack the overall session flow.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to execute shell commands to analyze the repository context.
Evidence: The skill body directs the agent to calculate "churn" and "author counts" using git: "Churn does not say which option is right... read it in one command and move on". This implies the use of the agent's command execution capabilities on the local file system.

wtk-discover