fix(peers): adopt canonical profile contracts

[drewstone]

Summary

• replace Agent App's local profile and question contract copies with the canonical agent-interface 0.43 values
• preserve deep immutability while converting profile configuration into App execution options
• align the current shared package cohort and require the public Runtime 0.123.1 model-routing fix in the package and both generated-project templates
• refresh generated API documentation and dependency locks

Why

Discovery changes complete AgentProfile values. Agent App must consume that shared contract directly so profiles do not fork into app-local shapes or silently lose model configuration.

Proof

• pnpm run typecheck
• incident tests: 293/293 passed
• full tests: 3,999/3,999 passed; 13 skipped
• NODE_OPTIONS=--max-old-space-size=12288 pnpm run build
• clean default project: typecheck, 4/4 tests, Cloudflare dry run
• clean chat project: typecheck, 5/5 tests, Cloudflare dry run
• pnpm run knip
• clean merge computation against current origin/main

[tangletools]

✅ Auto-approved drewstone PR — ddfecb51

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-08-03T05:08:08Z}

[tangletools]

🟢 Value Audit — sound

Verdict	sound
Concerns	0 (none)
Heuristic	0.0s
Duplication	0.0s
Interrogation	180.7s (2 bridge agents)
Total	180.7s

💰 Value — sound

Removes local harness-extras and interaction-field-widening shims now that agent-interface 0.43 carries them canonically — a clean, in-grain de-duplication.

• What it does: Two de-duplications, both enabled by bumping agent-interface 0.40→0.43. (1) harness/index.ts: drops the EXTRA_HARNESSES=['forge','cursor'] list, its compile-time adoption guard, the canonicalHarness() indirection, and the canonical/non-canonical branching in isModelCompatibleWithHarness/snapModelToHarness/snapHarnessToModel — Harness becomes a plain Exclude<HarnessType, NON_BACKEND_HARNESSES> and
• Goals it achieves: Kill the fork: agent-app held a second harness taxonomy (canonical + extras) and a second field schema (canonical + widened) purely because the pinned agent-interface predates those values. Once 0.43 ships forge/cursor in the enum and allowCustom/maxLength on the field schema, the local copies become drift risk — a harness added upstream would need an edit here, and a field flag the schema adds wo
• Assessment: Coherent and in the grain. The removed code existed for exactly one reason — the pinned peer lacked the values — and the diffs delete precisely that reason's machinery (the extras list, the guard, the canonicalHarness mapper, the widened field unions, the raw-return workaround) and nothing else. parseInteractionRequest moving from return request as InteractionRequestWire to `return validation.da
• Better / existing approach: none — this is the right approach. Searched src/ for AgentProfile local copies (none — it's imported from agent-interface throughout), for residual EXTRA_HARNESSES/canonicalHarness references (none), and for other interaction-field or harness-taxonomy duplications (chat-react re-exports HarnessType from sandbox-ui's picker, which is a different surface — the UI enum — and correctly stays separate)
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 2
• Bridge warning: opencode/kimi-for-coding/k2p7: opencode: opencode error

🎯 Usefulness — sound

Replaces agent-app's local interaction-field widening and harness-extras workarounds with the canonical agent-interface 0.43 contracts now that the upstream schema defines them, and aligns the peer cohort.

• Integration: Fully wired. The changed types (ChatInteractionField, InteractionRequestWire, parseInteractionRequest) are consumed by 6+ downstream modules — web-react/interaction-card-support.ts:55, web-react/chat-stream.ts:208, chat-store/parts.ts:205, chat-routes/sandbox-producer.ts:613, interactions/route.ts:143, interactions/sidecar.ts:105. tsc --noEmit is clean. The harness functions (snapHarnessToModel, i
• Fit with existing patterns: Exactly the codebase's stated grain. AGENTS.md invariant 6: 'Reuse the engine primitive; never hand-roll one the engine already gives you.' The local field widening (allowCustom/maxLength appended to InteractionField) and EXTRA_HARNESSES (forge/cursor held outside the canonical enum with a compile-time guard meant to force exactly this edit) were workarounds for a canonical contract that predates
• Real-world viability: Holds on and off the happy path. The parseInteractionRequest change is safer under edge inputs: a malformed-but-object-shaped request now cannot slip through as a raw cast. The harness simplification removes the canonicalHarness() indirection that short-circuited compatibility for non-canonical ids; since forge/cursor are now canonical, harnessSupportsModel governs them like every other harness —
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

No concerns — sound change, no better or existing approach found. ✅

---

What this audit checks

It judges the change on its merits — not whether it was tasked out in an issue. Unticketed, fast-moving work is fine; the question is whether the change is good and whether a better or existing approach should be used instead.

Pass	What it asks
Heuristic	Vague title? Whitespace-only or cruft-bearing diff? (content signals only)
Duplication	Do added function/class names already exist elsewhere in the repo?
Value Audit	What does it do? What goal does it achieve? Is it good? Better architecture or already-exists?
Usefulness Audit	Does it integrate and fit? Will it hold up in real use and actually get used?

Findings are concerns, not blocks — the human reviewer decides what to do with them.

_{value-audit · 20260803T051130Z}