The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the exec tool to run development commands including bun test, bun run lint, tsc --noEmit, and bun run build within the specified working directory.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it ingests and processes the output of external commands which may contain data from untrusted sources (e.g., error messages from failed tests on external code).
Ingestion points: Command output logs generated by test and build tools mentioned in SKILL.md.
Boundary markers: The skill uses a "bounded tool call" intended to return only a sifted report (verdict, tail, and specific failure lines) rather than the raw log, which provides a natural delimiter.
Capability inventory: The skill is granted access to the exec tool to run shell commands.
Sanitization: The skill applies filtering logic to extract only failures and a tail of the output, which naturally reduces the surface area for potentially malicious instructions hidden in large log files.

algal-verify