The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/invoke.py script executes an external engine (kntnt.py) using subprocess.run. Additionally, the skill's instructions and subagent prompts (references/correction.md) include commands to run helper scripts such as languages.py and article_anatomy.py using the uv tool.
[DYNAMIC_EXECUTION]: The skill implements a discovery mechanism in scripts/invoke.py that searches for the vendor's "Manager" engine across a wide range of global directory paths (e.g., ~/.claude/skills, ~/.cursor/skills). The path to the executable script is computed at runtime based on the local filesystem state.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process text artifacts from arbitrary sources, including external URLs via the --brief and operand arguments. This content is subsequently passed to subagents for analysis and repair, creating a surface where malicious instructions embedded in the source text could potentially influence the behavior of the subagents or the main agent.
Ingestion points: Text artifacts and Writing Briefs can be fetched from user-supplied URLs.
Boundary markers: The subagent prompt in references/correction.md uses <text> delimiters to encapsulate the artifact, but lacks explicit instructions to the subagent to ignore potentially malicious content within that text.
Capability inventory: The skill has the capability to execute shell commands via uv, spawn subagents, and invoke other skills (proofread).
Sanitization: The skill does not describe any specific sanitization or filtering of the input text before it is processed by subagents or scripts.
[EXTERNAL_DOWNLOADS]: The skill relies on the uv tool to execute its scripts. Depending on the environment state, uv may attempt to fetch Python interpreters or manage dependencies from external registries.

redline