The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to facilitate its iterative review and fix loop. According to SKILL.md, it runs mkdir to create workspace directories, invokes the cubic CLI tool for code analysis, and executes 'relevant validation (tests/lint/typecheck)' which typically involves project-specific scripts like npm test or pytest.\n- [PROMPT_INJECTION]: The skill accepts direct user input through a --prompt flag that is passed to the cubic CLI, allowing the user to influence the focus of the review. Additionally, it contains an indirect prompt injection surface related to processing CLI findings.\n
Ingestion points: The agent ingests data from JSON reports stored in .cubicloop/iteration-N.json which contain findings from the local code review.\n
Boundary markers: The instructions in SKILL.md do not define boundary markers or 'ignore' instructions when parsing the issues array from the JSON output.\n
Capability inventory: The agent possesses the capability to execute subprocesses (via cubic and validation scripts) and modify local source files to apply 'fixes' based on the findings.\n
Sanitization: There is no evidence of sanitization or safety checks performed on the titles or descriptions within the JSON findings before the agent acts on them to modify the source code.

cubicloop