The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data from user-provided novel text or files, which represents a surface for indirect prompt injection attacks.
Ingestion points: Untrusted text is ingested through local file paths (.txt, .md) or direct text pasting as described in Stage 2 of SKILL.md and the 'Pre前置' section of references/style-learn-sop.md.
Boundary markers: The instructions do not specify the use of clear delimiters or instructions for the agent to ignore potentially malicious content within the sample text during analysis.
Capability inventory: The skill allows the agent to write files to the local directory (文风库/文风.md) and execute a Python script via shell commands.
Sanitization: There is no evidence of sanitization or filtering of the input text before it is processed by the LLM for semantic analysis.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to perform statistical analysis using a local Python script.
Evidence: references/style-learn-sop.md instructs the agent to detect a Python interpreter and execute scripts/style_profile.py with specific flags and redirections (e.g., "$PYBIN" {SKILL_DIR}/scripts/style_profile.py --input "$SAMPLE" --json > 文风库/画像.json).
Context: While this is a functional requirement for the skill, executing shell commands based on instructional scripts is an elevated capability that interacts with the host environment.

moshu-style