The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user's repository to inform its implementation and review cycles, creating a potential surface for instructions embedded in source code to influence agent behavior.
Ingestion points: Source code files and evaluation cases referenced in SKILL.md and variants/evaluation.md, as well as logs and measurement data ingested in variants/performance.md.
Boundary markers: The provided instructions do not specify the use of delimiters or 'ignore' instructions for the ingested code content.
Capability inventory: The skill possesses authority to perform scoped code edits, execute local repository scripts, and invoke external review skills (/council and /review-loop).
Sanitization: No specific sanitization or filtering of external content is described before interpolation into the agent's context.
[COMMAND_EXECUTION]: The skill logic involves executing existing scripts located within the user's repository to establish performance or behavior baselines.
Evidence: SKILL.md states that the process 'settles one capture command, preferring a capture script the repository already has'.
[EXTERNAL_DOWNLOADS]: The skill references an external documentation source for manual evaluation procedures.
Evidence: variants/evaluation.md contains a link to 'https://github.com/Gabriel-Lafrance/Skills/blob/6c2735ff407385ed805e6d5c698eecc204552b0a/evals/tasks.md#how-to-run'. This resource belongs to the skill's author.

hillclimb