The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the orca command-line utility and a local script scripts/mo-models.mjs to manage review tasks and model selection.
[COMMAND_EXECUTION]: System utilities including realpath, umask, and mktemp are executed to ensure the integrity and isolation of the work environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes reports generated by independent review workers, introducing a vulnerability surface for indirect instructions.
Ingestion points: Reads and parses worker_done reports from worker processes as described in SKILL.md.
Boundary markers: Implements strict anchoring using markers such as Review-Execution and End-Review to delimit agent output.
Capability inventory: The skill has capabilities to execute CLI commands via the Orca tool and perform file system operations like atomic linking.
Sanitization: Uses an AST-based parser to validate report structure and verifies candidate SHAs and dispatch IDs against internal expected values to ensure report authenticity.

mo-review-orca