fix: auto-retry effort-capability 400s on models that predate newer effort tiers (v4.8.58)

[askalf]

What

Same-day smoke catch on the v4.8.57 autodetected catalog: newly-exposed older models hard-400 under a pinned effort the model doesn't support —

claude-opus-4-5-20251101 + DARIO_EFFORT=max →
400 "This model does not support effort level 'max'. Supported levels: high, low, medium."

Fix

Mirrors the existing context-1m / anthropic-beta rejection machinery (one method, not another special case):

• parseEffortRejection() extracts the rejected level + the model's supported set from the 400 body.
• Retry once with output_config.effort clamped to the strongest supported level (bestSupportedEffort: xhigh > max > high > medium > low) — value-only in-place mutation, JSON field order (a fingerprint surface) untouched.
• Supported set cached per wire model (effort support is a model property, not an account property); the body-build path clamps up front on every later request — the 400 round-trip is paid once per model per process.
• If the 400 matches but there's no output_config.effort to clamp, the original upstream error forwards unchanged (the branch consumes the body, so it forwards inline).

fable unaffected: its effort intolerance is a soft refusal (200 + stop_reason:"refusal"), invisible to 400-based machinery — stays on its measured resolveEffort clamp.

Tests

• New test/effort-capability.mjs — 16 assertions on the live-observed wire shape, parser negatives, and clamp preference.
• Full suite 86/86; tsc clean. Lockfile re-synced in-PR.

Post-merge

Will deploy to the box and re-run the prod smoke battery (the opus-4-5 case should flip to PASS).

[github-actions]

Compat test: ✅ PASSED

Ran node test/compat.mjs against dario proxy --passthrough on the self-hosted runner for commit b08aeb6a2e94f8cf970049a8dabcbe1f5e3dcaa7.

Output

============================================================
  dario Compatibility Validation (--passthrough via dario)
  2026-06-10T21:50:38.905Z
============================================================

--- Anthropic Messages API (Hermes) ---
✅ #1 Anthropic non-stream: "COMPAT OK" | thinking_injected=false | service_tier=-
✅ #2 Anthropic stream: 8 events | order=correct | "**STREAM COMPAT**"
✅ #3 SSE framing: event:/data: pairs correctly paired

--- Passthrough Verification ---
✅ #4 No thinking injection: No thinking block in response (passthrough clean)
✅ #5 Client betas preserved: Client-requested thinking honored (status=200)

--- Tool Use (OpenClaw) ---
✅ #6 Tool use: tool=get_weather | input={"location":"Tokyo"}
✅ #7 Tool use stream: 11 events | tool_use=true | input_json_delta=true

--- OpenAI Compat ---
✅ #8 OpenAI non-stream: "OPENAI COMPAT"
✅ #9 OpenAI stream: 3 chunks | [DONE]=true | "**STREAM OPENAI**"

--- Header Visibility ---
✅ #10 Header visibility: request-id=true | ratelimit=true (12 headers)

============================================================
  RESULTS: 10 passed, 0 failed, 0 warnings
============================================================

Full workflow run