The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git and cargo commands to inspect repository status, check for sync discrepancies, and build the project artifacts. These operations are limited to the local environment and require explicit invocation.- [REMOTE_CODE_EXECUTION]: The workflow involves running the repository's verification suite via cargo test and executing a locally built binary (codewhale). While this involves code execution, it is the primary stated purpose of the skill for environment verification and is explicitly staged with user confirmation requirements.- [SAFE]: The skill incorporates multiple safety constraints, including a prohibition on autonomous network operations (e.g., git fetch), a requirement for verbatim reporting to avoid model-generated hallucinations or prose, and the explicit-only invocation setting which prevents automatic execution without user request.- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the repository's version history and documentation.
Ingestion points: Processes content from CHANGELOG.md and git log output in SKILL.md.
Boundary markers: Defines strict caps on the amount of data processed (20 commits and 40 lines) and explicitly instructs the agent to quote findings verbatim.
Capability inventory: Has the ability to run build commands and project binaries as defined in SKILL.md.
Sanitization: Instructs the agent not to summarize, rank, or editorialize the ingested data, which serves as a control against following instructions embedded in the data.

contributor-onboarding