/debug-loop — debug the agentic loop

The loop is the single most expensive place to be wrong. Before you change anything, instrument.

1. Reproduce deterministically

• Run with FORGE_LOG_LEVEL=debug ./bin/forge.js <command>.
• Capture a session id. Events are in ~/.config/forge/events/<session>.ndjson.
• If the bug only repro's against a real provider, pin to a single model and disable the router's fallback: FORGE_MODEL=... FORGE_ROUTER_FALLBACK=0.

2. Read the event stream, not the terminal

Each turn emits: turn.start, model.call, tool.call, tool.result, validation.result, turn.end. If any stage is missing or out of order, that is your bug. Grep the ndjson before staring at code.

3. Check the usual suspects, in order

1. Mode caps (src/core/mode-policy.ts) — is the agent hitting the turn/token cap and being cut off? The event will say so.
2. Validation gate (src/core/validation.ts) — is the gate rejecting a valid step? Look at validation.result.reason.
3. Tool registry — is the tool the model asked for actually registered? tool.call with class: not_found means no.
4. Permissions — is the call being silently denied? tool.result.denied=true in the stream.
5. State machine (src/persistence/tasks.ts) — is an illegal transition being attempted?

4. Add a failing test before the fix

• Unit-test the smallest failing path with callModel mocked.
• If the bug is in the loop itself, test/unit/executor-loop.test.ts is the template.

5. Fix, then verify

Run /verify. If the test you added doesn't turn green, you fixed the wrong thing.