The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional Markdown content and documentation with no executable scripts or system-level commands.
[PROMPT_INJECTION]: The instructions incorporate a mandatory verification step (Step 1) that requires identifying concrete behavioral issues before modification, which prevents the agent from being misdirected by vague or conflicting instructions within the prompts being edited.
[DATA_EXFILTRATION]: No patterns of sensitive data access or unauthorized network communication were detected. Reference links in the documentation point to official vendor documentation.
[SAFE]: (Indirect Prompt Injection Analysis) 1. Ingestion points: Ingests prompt files like CLAUDE.md and AGENTS.md (Step 1). 2. Boundary markers: Uses an instruction-based boundary requiring the agent to ask for concrete issues before acting. 3. Capability inventory: Core capabilities include reading and writing files via standard agent tools. 4. Sanitization: Absent. The skill is assessed as safe because its procedural logic forces a verification and justification step that reduces the likelihood of executing malicious instructions contained within target prompt files.

improving-prompts