[rig-claude] Improve Claude dynamic-workflow compatibility for rig

[github-actions]

Compatibility gap addressed

The dynamic-workflow parity describe block in src/workflow.test.ts was missing
test coverage for two of the most fundamental Claude dynamic-workflow primitives:

1. call.text — await agent(prompt) without a schema: The most common
   first call a Claude dynamic-workflow developer writes. The mapping
   (await agent(prompt) → await call.text(prompt)) appears in the conversion
   table and is exercised in the workflow one-off agents block, but was absent
   from the dedicated parity block where a developer porting Claude code would
   look for evidence.

2. log(message) — emits a { type: "log" } event: Claude dynamic workflows
   rely on log(message) for observability. The parity block referenced log in
   comments and tested budget, but never verified that log() surfaces as a
   { type: "log" } event in the onEvent stream.

Why this improves transfer

A developer porting a Claude dynamic workflow reads the dynamic-workflow parity
describe to confirm that rig covers the primitives they rely on. A missing test
for call.text (the no-schema agent call) or for log events creates doubt even
when the behaviour is correct. These two additions make the parity block
exhaustive: every row in the conversion table's core primitives — call.text,
call.json, log, budget, phase override, parallel, pipeline,
call.workflow, non-object schema — now has a test in that block.

Files changed

• src/workflow.test.ts — 30 lines added: two new it(...) cases inside the
  existing describe("dynamic-workflow parity") block.

Validation

npm run typecheck  # tsc --noEmit: no errors
npm test           # 597 passed (up from 595), same 2 pre-existing failing files

The 2 pre-existing failing test files (src/launcher-default-engine.test.ts) and
their 439 failures were already present on main before this change.

Remaining intentional differences

• No effort mapping; prefer a more capable model id.
• No agentType: "Explore" mapping; narrow the tools list and prompt instead.
• No resume journal, worktree isolation, or human checkpoints.
• budget counts agent calls, not tokens.
• parallel requires uniform thunk return types; Promise.all for heterogeneous agents.
• pipeline stage signature is (previous, item, index) vs Claude's (item, index).

All intentional differences are documented in
skills/rig/references/claude-workflow-conversion.md.

Generated by Daily Rig Claude Dynamic Workflow Compatibility · sonnet46 125.1 AIC · ⌖ 8.2 AIC · ⊞ 5.4K · ◷