The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/invoke.py script dynamically locates and executes an external engine (kntnt.py) by searching through a large predefined list of potential installation paths in the user's home directory (e.g., ~/.claude/skills/kntnt/, ~/.cursor/skills/kntnt/). This results in the execution of code from paths determined at runtime based on the local environment.
[COMMAND_EXECUTION]: The skill uses uv run and subprocess.run to execute various local Python scripts, including the discovered engine and helper scripts like languages.py and article_anatomy.py.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary external content from URLs and local files provided in a 'brief'. While it implements a sophisticated defense mechanism by using a dedicated subagent to verify source fidelity and check for discrepancies, the ingestion of untrusted external data represents a significant attack surface.
Ingestion points: The <brief> argument in SKILL.md can contain text, local file paths, and URLs which are gathered in Step 2 of the workflow.
Boundary markers: The references/source-check.md file defines a specific prompt for the checker subagent that uses block quotes and clear instructions to treat the provided material as data for comparison rather than instructions.
Capability inventory: The skill can execute shell commands via uv run, perform network requests to fetch URLs, and spawn subagents.
Sanitization: The skill relies on a 'Source and quotation comparison' process performed by a fresh subagent to validate the draft against the original material before delivery.
[EXTERNAL_DOWNLOADS]: The skill is instructed to gather material from URLs provided in the user's brief, leading to the retrieval of remote content during the drafting process.

write