The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, which creates a theoretical surface for indirect prompt injection, although this is mitigated by safety instructions.
Ingestion points: Processes bug reports, error logs, stack traces, and local repository files as primary inputs in SKILL.md.
Boundary markers: Defines a mandatory output shape for the "Diagnosis Report" and references an explicit "Secret-safe evidence boundary".
Capability inventory: Capable of reading repository files, executing shell commands (e.g., curl), and running reproduction scripts or tests (SKILL.md).
Sanitization: Implements a detailed sanitization protocol in references/secret-redaction.md that requires redacting credentials and tokens from all displayed or captured data.
[COMMAND_EXECUTION]: The skill involves the execution of reproduction scripts and network commands as part of its core functionality.
Evidence: Step 2 of the process in SKILL.md explicitly directs the agent to build a feedback loop using minimal scripts, commands, or curl requests to observe bug behavior. This is a functional requirement for diagnosis and is not used for malicious purposes.

diagnose