The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses hardcoded Git commands (git status and git log) to inject context into the agent's environment. These commands are restricted to local repository metadata and do not involve network exfiltration or sensitive information disclosure.
[INDIRECT_PROMPT_INJECTION]: The skill ingests potentially untrusted diagnostic data via the $ARGUMENTS variable. The risk is managed by a structured process requiring independent reproduction and tracing. 1. Ingestion points: $ARGUMENTS in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: verify-ui, verify-cli, read-source, verify-pr. 4. Sanitization: Absent.

diagnose-root