The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts such as up, doctor, and down, as well as arbitrary commands found in the quick-start documentation of targeted skills. These executions are scoped to the .claude/skills/verify-*/ directory and are part of the intended audit process.- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface where untrusted data could influence agent behavior through the following components:
Ingestion points: The skill reads git log output and content from .claude/skills/verify-*/SKILL.md files.
Boundary markers: There are no delimiters or explicit instructions for the agent to ignore embedded commands when processing the data ingested from git logs or skill files.
Capability inventory: The skill can execute shell commands, run project-local scripts, and perform file system operations to fix drift.
Sanitization: No sanitization or validation is performed on the ingested data before the agent processes it.

maintain-verification-skill