The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a local utility script, tools/scan_fingerprints.py, to perform style checks on documents. This script is part of the skill package and performs local regex-based scanning. This use of a provided tool for a specific function is considered safe behavior.
[COMMAND_EXECUTION]: The test file tools/tests/smoke_test_scan_fingerprints.py uses subprocess.run to execute the scanner script. This is a standard testing practice for CLI tools and is benign.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown documents for auditing, which is an untrusted input surface.
Ingestion points: User-provided Markdown files and text snippets entered in the chat (processed by scan_fingerprints.py and reviewed by subagents in multi-model.md).
Boundary markers: The workflow uses 'materialized snapshots' (temporary files) and provides subagents with a strict 'reviewer contract' (found in ref/multi-model.md) that defines their role and limits their reading scope to the provided files.
Capability inventory: Local file reading/writing and execution of the provided scan_fingerprints.py script; no external network operations are present in the provided tools.
Sanitization: The scanning tool includes a mask_nonprose function that filters out code blocks and link targets before analysis, preventing non-prose content from triggering stylistic findings or influencing the scanning logic.

yzr-writing-style