The Agent Skills Directory

[SAFE]: The skill demonstrates strong security awareness by explicitly instructing the agent and user to redact all secrets, tokens, and customer identifiers from documentation examples to prevent accidental data exposure.
[SAFE]: The included Python script (scripts/error-cluster.py) is designed for local execution to analyze error logs and tickets. It does not perform any network operations, use unsafe dynamic execution (like eval), or require external dependencies.
[SAFE]: The skill incorporates a 'verification protocol' (Step 7), requiring that every fix be reproduced and executed before publication. This minimizes the risk of publishing incorrect or harmful instructions.
[SAFE]: The instructions include a refusal mechanism in the 'Interview' section, where the agent is told to decline generating documentation if no real-world failure signals (evidence) are provided, preventing the creation of potentially misleading or hallucinated content.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, untrusted data (support tickets, community threads, and issue trackers). While this data could theoretically contain malicious instructions, the skill mitigates this risk by using a script for structural analysis (normalization and clustering) rather than direct instruction following, and by requiring human-in-the-loop verification of all generated content.

developer-troubleshooting-docs