The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes PreToolUse and PostToolUse hooks to automatically execute local shell scripts, including multi_linter.sh, protect_linter_configs.sh, and stop_config_guardian.sh, on every file modification.
[REMOTE_CODE_EXECUTION]: Implements a three-phase architecture that spawns claude -p subprocesses to interpret violation data and automatically generate and apply code fixes to the project files.
[EXTERNAL_DOWNLOADS]: Recommends the installation of multiple developer tools and linters (such as jaq, ruff, uv, biome, and shellcheck) from standard package registries and system package managers.
[COMMAND_EXECUTION]: Enforces specific tooling preferences by intercepting and blocking standard shell commands for package managers like pip, npm, and poetry, redirecting usage to alternatives such as uv and bun.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface:
Ingestion points: multi_linter.sh collects violation messages and error codes from various external linters and tools (SKILL.md).
Boundary markers: None specified in the documentation regarding the isolation of linter output when passed to subprocess prompts.
Capability inventory: The system can execute shell commands, write to the filesystem, and spawn new AI agent instances (SKILL.md).
Sanitization: No sanitization, escaping, or validation mechanisms are described for the untrusted data retrieved from linter outputs before processing.

plankton-code-quality