The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection as it is designed to ingest and process external text for rewriting and translation.
Ingestion points: The skill reads external content provided by the user or from files using the Read tool for the purpose of style adjustment and translation as defined in SKILL.md and references/translation-guardrails.md.
Boundary markers: No explicit delimiters or specific boundary instructions are used to separate user-provided text from the agent's core processing logic.
Capability inventory: The skill has access to Read, Write, and Edit tools, allowing it to modify local files based on the processed output.
Sanitization: The skill's extensive style rules (e.g., in references/ai-trace-detector.md) serve as a functional filter. By strictly prohibiting specific words, sentence structures, and 'roadsign' phrases typical of AI models, it is likely to disrupt or remove common prompt injection payloads embedded within the input text.
[SAFE]: The skill includes a local PowerShell script (tools/style-lint.ps1) for text analysis.
The script calculates metrics such as paragraph density and punctuation usage to verify adherence to style guidelines.
It uses Unicode-escaped character literals (e.g., \u800c\u662f for '而是') to maintain consistent pattern matching for Chinese characters across different system encoding environments.
The script performs static analysis and does not utilize dynamic code execution or external network calls.

de-AI-writing