The Agent Skills Directory

[SAFE]: No malicious patterns or vulnerabilities were found in the instructions, configuration, or Python script. The skill follows a conservative, diagnostic-focused approach.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it processes externally-sourced error logs and tracebacks.
Ingestion points: scripts/safe_debug.py reads untrusted text from command-line arguments or files specified by the user.
Boundary markers: Instructions in SKILL.md and references/debug-policy.md explicitly direct the agent to wait for human approval before taking any action and to focus solely on diagnosis.
Capability inventory: The skill is restricted to local file reading and writing of analysis results; it has no capabilities for network exfiltration, arbitrary command execution, or automatic code modification.
Sanitization: The input text is categorized via keyword matching and summarized into a short excerpt for human evaluation.

safe-debug