The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires downloading the Plankton toolkit from an external GitHub repository (https://github.com/alexfazio/plankton.git) and installing several third-party tools via package managers like brew and uv.- [COMMAND_EXECUTION]: The skill executes numerous shell scripts and system binaries (e.g., multi_linter.sh, protect_linter_configs.sh, ruff, biome, hadolint) through platform-specific hooks. It also spawns autonomous AI subprocesses using the claude -p command to automate code fixes.- [PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection. Linter violations, which are derived directly from the content of project files, are passed as input to a secondary AI subprocess without explicit sanitization or boundary protection.
Ingestion points: Linter violation messages generated from file content during the execution of multi_linter.sh.
Boundary markers: Absent; the documentation does not indicate the use of delimiters or instructions to ignore instructions embedded in the linter output.
Capability inventory: The skill can spawn claude -p subprocesses, execute shell scripts, and run multiple CLI linting and formatting tools.
Sanitization: Absent; while the linter output is structured as JSON, it is not filtered to remove potentially malicious natural language instructions.

plankton-code-quality