The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted external data such as error logs and reproduction scripts during the diagnostic process.
Ingestion points: Section 3 of SKILL.md specifies the collection of raw error messages, request payloads, and terminal outputs from external sources.
Boundary markers: The skill instructs the use of a structured 'Diagnostic-Session Template' and 'Evidence Ledger' (Section 3 and Section 9) to separate raw observations from analytical assumptions.
Capability inventory: The agent utilizes Read and Grep tools and is instructed to perform investigative tasks like git bisect, profiling, and system instrumentation.
Sanitization: The skill provides a specific 'Evidence safety rule' in Section 3, mandating the redaction or replacement of sensitive data like credentials, session tokens, and personal information before evidence is incorporated into the session context.

diagnosis