The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation describes the use of shell scripts (multi_linter.sh, protect_linter_configs.sh, stop_config_guardian.sh) as PreToolUse, PostToolUse, and Stop hooks. These scripts actively monitor and intercept file edits and shell commands to enforce quality standards and block legacy tools.
[DYNAMIC_EXECUTION]: The skill utilizes a delegation phase that spawns claude -p subprocesses. These sub-agents are dynamically invoked to fix code violations based on JSON output from linters, using a tiered routing system that selects model versions based on issue complexity.
[INDIRECT_PROMPT_INJECTION]: The automated workflow presents an indirect prompt injection surface by feeding file contents and linter violations directly into a subprocess agent.
Ingestion points: Every file edit or write operation performed by the main agent triggers the collection of data for the multi_linter.sh hook.
Boundary markers: The instructions do not specify any delimiters or safety prompts to prevent the sub-agent from following instructions embedded within the code files it is attempting to fix.
Capability inventory: The skill facilitates subprocess spawning (claude -p), broad filesystem monitoring via hooks, and the ability to block or redirect package manager commands.
Sanitization: No evidence is provided regarding the sanitization or escaping of linter results or source code before it is passed to the delegation subprocess.

plankton-code-quality