The Agent Skills Directory

[SAFE]: The skill enforces environment isolation by requiring all exploratory scaffolding and code execution to be confined to the .agents/scratch/ directory or dedicated worktrees, which prevents unauthorized modifications to the primary codebase.
[SAFE]: A 'Thin Tracer Only' invariant is established, restricting the agent to minimal dummy payloads (ping-pong tracers). This limitation prevents the construction of complex or hidden logic during the architectural spike process.
[SAFE]: The execution protocol and domain rubric focus strictly on legitimate architectural verification tasks such as checking service wiring and serialization compatibility, with no evidence of data exfiltration or credential harvesting.

skeleton-spike