The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a chain of Python scripts (e.g., preflight.py, init_run.py, ledger.py, partition.py, report.py) located in a relative path (../review-deep/scripts). It also performs file system operations such as creating directories and deleting temporary files using find.
[INDIRECT_PROMPT_INJECTION]: The skill ingests the entire repository content via repomix and provides this data to nine specialized subagents for review.
Ingestion points: The repomix.xml file, which snapshots the target directory/repository (SKILL.md, Step 4).
Boundary markers: None explicitly defined in the prompts to distinguish between code and instructions embedded in the code.
Capability inventory: The skill can execute shell commands, read/write files in the .llmtmp and cache directories, and dispatch subagents (SKILL.md, Steps 1, 3, 5, 7).
Sanitization: No evidence of input sanitization or escaping of the codebase content before it is processed by the reviewer agents.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx repomix and uv run to fetch and execute external tools and libraries, specifically requiring scikit-learn, scipy, and numpy at runtime (SKILL.md, Steps 4 and 7).
[PROMPT_INJECTION]: The instructions contain an explicit directive to "Never pause before dispatch to confirm scope, cost, or model spend," which is designed to bypass standard interaction patterns or safety/cost confirmations during autonomous operation.

review-full