The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from the repository, including tasks.md, design.md, and code diffs, to drive its verification logic.
Ingestion points: File content from tasks.md, design.md, and git diff outputs are read into the agent context.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within these files are specified.
Capability inventory: The skill is capable of executing arbitrary shell commands through the comet check run interface and performing file system writes.
Sanitization: No evidence of sanitization or validation of the input repository data is provided.
[COMMAND_EXECUTION]: The skill enables the execution of user-defined verification programs and arguments via the comet check run command. This capability allows for the execution of project-specific scripts which could be leveraged maliciously if the local environment is compromised.
[DYNAMIC_EXECUTION]: The instructions require loading an external skill, openspec-verify-change, at runtime when full verification is active. This dynamic loading of executable logic introduces a dependency on external components not defined within the skill itself.

comet-verify