The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill collects untrusted user input (skill name, usage scene, and notes) and passes them as arguments to a shell command execution.
Ingestion points: User-provided fields skill, scene, satisfaction, and note defined in SKILL.md (Step 1).
Boundary markers: The instructions do not provide specific delimiters or warnings for the agent to sanitize these inputs before constructing the python3 command line.
Capability inventory: The skill executes local Python scripts via shell and performs file-write operations to the local filesystem (06 计划/skill_usage_log.jsonl).
Sanitization: The Python script scripts/log_skill_usage.py uses json.dumps() to safely escape data before writing it to the log file; however, there is no explicit sanitization for the command-line arguments themselves within the skill instructions.

skill-logger