Agent Run Inspector

Local, read-only reports for OpenCode session exports — repeated identical tool calls, recovered vs unresolved failures, completion state, tool and usage statistics. Standard-library Python, no network calls, input never modified.

Live sample reports (synthetic data)

• Retry loopA failing tool call repeating with no success — the retry-loop findings view.
• Normal sessionA clean run: no repeats, one recovered retry, completed.
• Provider errorA run that died on a provider error mid-step.
• Interrupted sessionA run that stopped mid-step with unresolved items.

Free runtime guard

Try a reproducible Agent Run Guard repeat-call check. Guard is a separate free MIT tool; Inspector is not required.

Repository agent instructions

Preview the Agent Repo Control Pack workflow. It creates and validates one canonical repository policy for documented Codex, Cursor Agent, supported Copilot, and Claude Code instruction paths.

Links

• Documentation & sample pack (GitHub) — quickstart, how-to-read guide, all samples
• Free demo pack · Individual — $49 · Team — $149
• Repo Control Pack individual — $29 · Team — $79
• Walkthrough: inspect repeated OpenCode tool calls