The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to verify its own output by executing a local Python linter script (python3 scripts/ste-lint.py) before final delivery.
[DYNAMIC_EXECUTION]: Hook scripts such as ste-gate.py and ste-pregate.py use the subprocess.run method to execute the local linter script at runtime, providing automated feedback on text quality.
[PERSISTENCE]: An install.py script is provided to register the skill's hooks in the agent's persistent configuration file (~/.claude/settings.json) and establish symlinks for the skill's directory and style guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill monitors tool inputs (like commit messages) and written markdown files to provide automated linting; this ingestion of external data represents a standard attack surface for indirect prompt injection, though the linter itself performs only static regex-based analysis.
[DATA_EXPOSURE]: The skill processes transcript history and local file contents to evaluate writing style. This data remains local to the system and is used exclusively for style enforcement.

ste-writing