The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data from external logs and diagnostics, presenting a surface for indirect prompt injection.
Ingestion points: Workflow instructions in SKILL.md and references/checklist.md guide the agent to read error outputs, stack traces, and console logs.
Boundary markers: The skill does not define specific delimiters to isolate this external data, though its focus is on evidence collection rather than execution of log content.
Capability inventory: The agent can execute reproduction commands and scripts based on its analysis.
Sanitization: No explicit sanitization or filtering of ingested diagnostic content is mentioned in the instructions.
[COMMAND_EXECUTION]: The skill generates and executes temporary scripts and probes as part of its core diagnostic function.
Evidence: references/checklist.md and SKILL.md instruct the creation of "task-owned temporary artifacts" and reproduction harnesses for testing hypotheses.
Context: Dynamic execution is scoped strictly to temporary, non-persistent diagnostic activities.
Mitigation: 'Hard Rules' in SKILL.md explicitly forbid any permanent modification to the repository, tracked files, or git refs, ensuring that all execution remains within a controlled, read-only boundary relative to the target codebase.

diagnose