The Agent Skills Directory

[OBFUSCATION]: The scripts/check.py script contains hardcoded invisible Unicode characters (such as Zero-Width Space U+200B, ZWNJ U+200C, and Unicode Tags) within a regular expression.
Evidence: Found in the INVISIBLE variable definition in scripts/check.py.
Purpose: These characters are used as part of a defensive sanitization routine to identify and remove them from user-provided text, which is a security best practice to prevent steganographic attacks.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, user-provided documents as part of its core writing and analysis workflow.
Ingestion points: The skill reads user-provided manuscripts (e.g., 稿件.md) for analysis and cleaning.
Boundary markers: The skill uses a structured multi-stage workflow (fragments, beats, shape) to process text.
Capability inventory: The included scripts/check.py script reads and analyzes local text files.
Sanitization: The clean function in scripts/check.py specifically targets and removes invisible characters often used in prompt injection attacks.
[COMMAND_EXECUTION]: The skill instructions involve running a local Python script (scripts/check.py) provided with the skill package.
Evidence: SKILL.md contains instructions to execute python3 ~/.claude/skills/write/scripts/check.py 稿件.md.
Analysis: The script's logic is dedicated to text pattern matching and statistical analysis of sentence/paragraph structures. It does not perform network requests, access sensitive system credentials, or perform unauthorized file operations.

write