The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes pnpm scripts defined in the local repository's package.json through a bash wrapper script. The wrapper performs validation to ensure requested gate names exist in the configuration before execution, preventing arbitrary command injection through the tool parameters.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and interpret diagnostic messages from compilers, linters, and test runners to automatically fix code errors, creating a vulnerability surface.
Ingestion points: The agent reads logs and parsed error files (e.g., raw.log, errors.txt) located in the .SS_WF/Agent/QUALITY_RUNS/ directory, which contain output from external execution environments.
Boundary markers: The instructions lack specific delimiters or "ignore instructions" warnings for the data ingested from tool outputs, which might contain malicious payloads embedded in source code comments or test failure messages.
Capability inventory: The agent is granted extensive capabilities to modify source files throughout the repository to satisfy the quality gates.
Sanitization: While the skill parses tool output for structure, the raw diagnostic messages are not sanitized or escaped before being processed by the agent, potentially allowing crafted code content to influence the agent's code-modification decisions.

master-static-code-quality