The Agent Skills Directory

[COMMAND_EXECUTION]: The test suite located in tests/test_e2e_pipeline.py uses the Python subprocess.run function to drive the claude command-line tool. This is a legitimate use of shell execution within a development context to perform end-to-end testing of the skill's output against defined stage gates.
[SAFE]: The skill explicitly defines an approval process in references/stages/approve.md that writes finalized documents to the user's filesystem (defaulting to ~/Desktop). This file-writing behavior is a core feature for artifact persistence and is clearly documented for the user.
[SAFE]: The references/doc/general/agent-skill-architecture.md file describes the use of platform-specific dynamic context injection (shell commands like pwd or find). These patterns appear within a guide for building AI skills and are not executed by the Kernel Write skill itself during normal operation.
[SAFE]: The skill processes untrusted user-provided content such as raw notes and transcripts through its 'Intake' stage. While this constitutes an attack surface for indirect prompt injection, the skill's implementation of strict, rule-based critique and editing stages provides a robust framework for normalizing and validating output before final approval.

kernel-write