The Agent Skills Directory

[SAFE]: The skill establishes a robust error-handling protocol with a strong focus on security. It explicitly instructs the agent to redact sensitive information, including credentials, tokens, session identifiers, raw stack traces, SQL, and internal hostnames from public error envelopes.
[SAFE]: The next_actions mechanism, which allows suggesting recovery steps, is implemented with clear security constraints. The instructions mandate that these actions are strictly advisory and that agents must apply their own execution policies. It specifically forbids relaying untrusted input into fields that a client might execute.
[SAFE]: The workflow follows standard development practices for auditing and implementing code within a local repository context. It does not involve suspicious external downloads, unauthorized network operations, or persistence mechanisms.
[SAFE]: The provided JSON schema for the error envelope enforces structured data and machine-readable fields, which helps prevent schema confusion and indirect injection attacks by avoiding free-text imperative instructions.

agent-recoverable-errors