The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted text, creating a vulnerability surface for indirect prompt injection based on the following evidence:
Ingestion points: Text to be processed enters the agent's context through manual pastes, the Read tool for specific file paths, or the current selection in the editor (SKILL.md, Step 1).
Boundary markers: The instructions do not define protective boundary markers (such as specific XML tags) to wrap user-provided text, nor do they include explicit directions to ignore instructions embedded within the processed content.
Capability inventory: The skill is granted access to the Bash tool (used for running scripts/scan.py), and file system tools including Read, Write, Edit, Grep, and Glob.
Sanitization: There is no evidence of pre-processing or sanitization to remove potentially malicious instructions from the input text before it is analyzed by the agent.
[OBFUSCATION]: The file scripts/humanizer_metrics/markers.py contains hidden Unicode characters, including U+200B (Zero-Width Space), U+200C (ZWNJ), U+200D (ZWJ), and U+FEFF (BOM), within a regex pattern. While these characters are intended to detect similar artifacts in AI-generated text, their presence in the script's source code is flagged as obfuscation by static analysis tools.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute a local Python script (scripts/scan.py) for deterministic text metrics. While this execution is part of the skill's primary functionality, it involves running script code included with the skill package.

humanizer