The Agent Skills Directory

[COMMAND_EXECUTION]: The hook scripts (hooks/ste-gate.py, hooks/ste-pregate.py, and hooks/ste-refresh.py) utilize the subprocess.run function to execute a local linter script (scripts/ste-lint.py). The scripts resolve the path to the linter dynamically, looking within the skill's relative directory or at the standard installation path in the user's home directory.
[PERSISTENCE]: The install.py script automates the registration of hooks and output styles by modifying the agent's local configuration file (~/.claude/settings.json). It also creates symbolic links within the user's home directory to manage the skill's persistence and integration across different agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes data that may contain untrusted content.
Ingestion points: The PostToolUse hook (ste-refresh.py) reads markdown files written by the agent; the PreToolUse hook (ste-pregate.py) reads command-line inputs (e.g., git commit messages); and the Stop hook (ste-gate.py) reads the agent's interaction transcript.
Boundary markers: The linter results and feedback are injected into the agent's context using specific XML-like tags such as <asd-ste100-feedback> and <asd-ste100-standing-rule>.
Capability inventory: The skill has the ability to read local files, manage state in ~/.claude/ste-gate, and execute its own Python-based linter.
Sanitization: The scripts employ a strip_noise function to remove URLs and markdown table rows from the text before processing, and the ste-lint.py script is designed to ignore code blocks to avoid false positives and noise.
[DYNAMIC_EXECUTION]: The skill performs dynamic execution by locating the current Python interpreter at runtime (sys.executable) to run its internal linter script against the text being processed.

asd-ste100