The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses uv run and subprocess.run to execute scripts. Specifically, SKILL.md directs the agent to run scripts/invoke.py, which in turn executes the kntnt engine.
[DYNAMIC_EXECUTION]: The script scripts/invoke.py dynamically locates its execution target by scanning a large hardcoded list of home directory paths (including directories for Claude, Cursor, Windsurf, and many others). It then executes the first script found at these locations via subprocess.run. This 'search-and-execute' logic from computed paths is a high-risk dynamic execution pattern, as it could potentially execute code from untrusted or hijacked local directories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from paths, URLs, or direct input (ingestion point: SKILL.md). This content is passed to a subagent for correction. While the subagent is constrained by the correction.md brief to only modify specific slop patterns and preserve other content (sanitization logic), the ingestion of untrusted data into a subagent pipeline represents a vulnerability surface for indirect prompt injection. The skill includes a capability to create subagents but lacks explicit boundary delimiters for the input data.
[SAFE]: The skill employs security best practices by instructing the agent to use fresh, private temporary directories for every command and ensuring these directories are cleaned up immediately after execution, even in the event of failure.

unslop