feat(routing): route harness-protocol turns up, never down; auto-detect claimed-tool-unavailable by devin-ai-integration[bot] · Pull Request #935 · workweave/router

devin-ai-integration · 2026-08-18T13:10:49Z

Summary

Rewrite of #921 by @makosblade (all credit for the design and implementation to them) — reimplemented on a fresh branch off main to align with internal conventions and fix two bugs found in review.

Behavior (as in the original PR):

New turn types HarnessMeta, SubAgentHarnessMeta, Recovery in internal/router/turntype, with Base() mapping back to the stable policy vocabulary and HarnessEscalation() marking turns that must be served by a strong Claude-family model. applyHarnessEscalation clamps the per-turn decision (runTurnLoop = runTurnLoopInner + clamp) to Anthropic Opus — per-turn only, never written to session pins. Kill switch: ROUTER_HARNESS_ESCALATION_ENABLED.
Claimed-tool-unavailable auto-feedback: post-stream detector scans the captured response for a declared tool name near an "unavailable" phrase and persists a source="auto" negative RouterFeedbackEvent (claimed-tool-unavailable:<tool>), deduped per (session, role, tool) via expiring LRU. Persist-only; never influences live routing.
New helpers: catalog.IsClaudeFamily, translate.ClaudeCodeOnlyToolNames(), translate.ReasonHarnessEscalation, RequestEnvelope.LastUserToolResultErrorText.

Fixes over #921:

NewService now initializes claimedToolTracker (it only initialized spiralTracker, so the nil guard in maybeReportClaimedToolUnavailable made the auto-feedback path dead in production). Regression test constructs via NewService and asserts a row persists.
SubAgentHarnessMeta is gated on translate.FormatAnthropic, mirroring the main-turn gate — OpenAI/Gemini sub-agent dispatches mentioning "plan mode"/"tool schema" now stay on the cheap SubAgentDispatch path instead of escalating to Opus. Regression tests for all three formats.
Comments trimmed to the concise-comment rule (applied workweave-bot's six shortening suggestions; removed narration-only blocks); turntype/AGENTS.md and CLAUDE.md kept verbatim-identical.

make precommit and make test pass.

Link to Devin session: https://app.devin.ai/sessions/6f4da5e234fd42cb83560f40a52e6a2b
Requested by: @steventohme

…ct claimed-tool-unavailable Adds three deterministic turn classifications (harness_meta, sub_agent_harness_meta, recovery) and a post-routing clamp that escalates those turns to claude-opus-5 unless the resolved decision is already a Claude-family TierHigh model. Sub-agent harness turns no longer take the cheap sub-agent hard-pin; usage-bypass, operator hard pins, /force-model pins, and loop escalation all outrank the clamp; kill switch is ROUTER_HARNESS_ESCALATION_ENABLED (default on). The policy sidecar keeps a stable turn-type vocabulary via TurnType.Base() while telemetry records the full new values. Also adds a capture-gated post-stream detector that writes source=auto rating=down router_feedback rows when a response claims a tool is unavailable that the request actually declared - persist-only, LRU-deduped per (session, role, tool). Motivated by a large agentic session where a sub-agent turn "Load EnterPlanMode tool schema" was served by a flash-tier model that could not drive the deferred-tool protocol, silently losing plan mode. Co-Authored-By: Weave Router <router@workweave.ai>

devin-ai-integration · 2026-08-18T13:10:52Z

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

Disable automatic comment, CI, and merge conflict monitoring

greptile-apps · 2026-08-18T13:10:53Z

PR author is not in the allowed authors list.

claude · 2026-08-18T13:11:07Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 9 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

cursor

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 8aad632. Configure here.}

claude · 2026-08-18T13:16:23Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Claude finished @devin-ai-integration[bot]'s task — View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 6 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

claude · 2026-08-18T13:22:05Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Claude finished @devin-ai-integration[bot]'s task — View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 4 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

devin-ai-integration · 2026-08-18T13:27:44Z

Applied the first two rounds of comment-length suggestions (commits 7a42241, 7fa2f2d). Politely declining further rounds: the comments now meet the documented convention (root AGENTS.md — concise, why-only), CI is green, and the advisory check is non-converging by design. Happy to revisit if a human reviewer flags a specific comment.

Written by Devin

claude · 2026-08-18T19:09:04Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Claude finished @devin-ai-integration[bot]'s task — View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 6 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

claude · 2026-08-18T19:15:08Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Claude finished @devin-ai-integration[bot]'s task — View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 3 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

claude · 2026-08-18T19:20:13Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Claude finished @devin-ai-integration[bot]'s task — View job

Gather diff and analyze added comment blocks
Flag verbose comment blocks (3+ lines)
Post review suggestions

Posted 4 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

claude · 2026-08-18T19:24:57Z

Claude finished @devin-ai-integration[bot]'s task —— View job

Run git diff to analyze added comment blocks
Flag verbose 3+ line comment blocks
Post review suggestions if found

Posted 3 committable comment-length suggestions on PR #935. All are advisory; none block merge.

workweave-bot

Advisory only — comment-length nits. Won't block merge.

makosblade and others added 2 commits August 18, 2026 13:05

fix(routing): close harness escalation gaps

8aad632

devin-ai-integration Bot mentioned this pull request Aug 18, 2026

feat(routing): route harness-protocol turns up, never down; auto-detect claimed-tool-unavailable #921

Closed

workweave-bot reviewed Aug 18, 2026

View reviewed changes

cursor Bot reviewed Aug 18, 2026

View reviewed changes

Comment thread internal/proxy/claimed_tool_unavailable.go

Comment thread internal/proxy/service.go

style(routing): shorten harness comments

7a42241

workweave-bot reviewed Aug 18, 2026

View reviewed changes

style(routing): trim remaining harness comments

7fa2f2d

workweave-bot reviewed Aug 18, 2026

View reviewed changes

Comment thread internal/proxy/claimed_tool_unavailable.go Outdated

Comment thread internal/proxy/harness_escalation.go Outdated

Comment thread internal/translate/claudecode_tool_filter.go Outdated

Comment thread internal/translate/lastuser_toolresult.go Outdated

style(routing): shorten remaining review comments

05edf82

workweave-bot reviewed Aug 18, 2026

View reviewed changes

style(routing): shorten round-four review comments

8ae49d6

workweave-bot reviewed Aug 18, 2026

View reviewed changes

Comment thread internal/proxy/claimed_tool_unavailable.go Outdated

Comment thread internal/proxy/claimed_tool_unavailable.go Outdated

Comment thread internal/router/turntype/harness.go Outdated

style(routing): shorten round-five review comments

b521c8d

workweave-bot reviewed Aug 18, 2026

View reviewed changes

Comment thread internal/proxy/turnloop.go Outdated

Comment thread internal/proxy/turnloop.go Outdated

Comment thread internal/proxy/claimed_tool_unavailable.go Outdated

Comment thread internal/proxy/claimed_tool_unavailable.go Outdated

style(routing): shorten round-six review comments

672a2ef

workweave-bot reviewed Aug 18, 2026

View reviewed changes

Comment thread internal/proxy/claimed_tool_unavailable.go

Comment thread internal/router/turntype/harness.go

Comment thread internal/proxy/turnloop.go

Uh oh!

Conversation

devin-ai-integration Bot commented Aug 18, 2026

Summary

Uh oh!

devin-ai-integration Bot commented Aug 18, 2026

🤖 Devin AI Engineer

Uh oh!

greptile-apps Bot commented Aug 18, 2026

Uh oh!

claude Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

claude Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

claude Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

devin-ai-integration Bot commented Aug 18, 2026

Uh oh!

claude Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

claude Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

workweave-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading

claude Bot commented Aug 18, 2026 •

edited

Loading