The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script scripts/build_style_profile.py to perform quantitative analysis on text samples provided by the user. This is a primary functional component used to generate metrics like sentence length, punctuation density, and vocabulary frequency. The command is executed with local paths for input and output as defined in the skill's workflow.
[INDIRECT_PROMPT_INJECTION]: The skill processes external text files (sample articles) to extract stylistic patterns, which introduces an indirect prompt injection surface.
Ingestion points: Sample articles are read from a user-specified directory by both the quantitative Python script and the agent's qualitative analysis step.
Boundary markers: The instructions explicitly frame the ingested content as 'samples' and 'references', which helps the agent distinguish between data and instructions.
Capability inventory: The skill has the ability to read files, execute its own bundled script, and write output files to /tmp/ or a user-defined style directory. It does not possess network access or elevated system privileges.
Sanitization: The quantitative script acts as a security boundary by converting raw text into structured JSON metrics (e.g., character counts, word frequencies) before the agent performs deeper reasoning. The script even includes a 'banned phrase audit' that flags common AI tropes and injection-like patterns (e.g., 'forget previous...').

wechat-style-profiler