The Agent Skills Directory

[SAFE]: The skill is designed as a read-only analysis tool with explicit boundaries preventing destructive actions such as file modification, commits, or browser-based reproduction.
[INDIRECT_PROMPT_INJECTION]: The skill processes external repository data (code, logs, ADRs) which represents an ingestion surface for untrusted instructions. However, it explicitly instructs the model to treat retrieved content as evidence/data rather than authority, using a 'Retrieved Evidence Boundary' to prevent data from overriding the skill's operational protocol. The severity is low as it defines clear boundaries and lacks capabilities for automated writing or network exfiltration.
[COMMAND_EXECUTION]: The skill operates within the agent's environment to read files and inspect repository state. While it interprets code, it contains instructions to stop with Status: Blocked or Status: Unverifiable if evidence is ambiguous, preventing hallucinated claims or unsafe inferences.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, package installations, or external script downloads were detected. The skill uses local synthetic fixtures for testing its investigation logic.
[DATA_EXFILTRATION]: There are no network operations or non-whitelisted domain accesses present. The skill's purpose is to summarize internal repository findings for the user.

tk-ask-repo