Summary

Closes #2464 / Area 3 of umbrella #2336.

Removes SessionEntry.thinkingLevel and the gateway protocol schema field, plus every consumer across gateway, auto-reply, agents, status/TUI, ACP translator, HTML export, Mattermost extension, and tests. RemoteClaw does not control reasoning depth — CLI runtimes own it.

Also removes agents.defaults.thinkingDefault (dead after /think directive was gutted in #2466).

Why this is safe

Reader-before-writer ordering:

1. gut(ui): remove dead thinkingLevel display plumbing (#2336 Area 7) #2471 (PR audit(apps+ui): thinkingLevel display-layer classification (#2336 Area 7) #2467) already removed the Web UI reader of chat.history.thinkingLevel and SessionsPatchResult.entry.thinkingLevel. Web UI stopped reading before this PR stops sending.
2. Native apps (Android, iOS/macOS/shared) still reference thinkingLevel as display-only state-drivers per gut(ui): remove dead thinkingLevel display plumbing (#2336 Area 7) #2471's audit. After this PR, the field is absent from gateway responses — apps see it as undefined and degrade gracefully. Full native-app rewrite deferred per umbrella guidance.
3. ACP clients: sessions.patch({thinkingLevel}) now fails schema validation (additionalProperties: false). ACP bridge's own setSessionMode and setSessionConfigOption("thought_level", ...) flows were gutted in the same PR.

Changes

Core type + schema (10 files)

ACP bridge surgery (src/acp/translator.ts)

Thought-level mode presentation gutted entirely:

• thinkingLevel removed from GatewaySessionPresentationRow Pick and GatewaySessionRow union
• formatThinkingLevelName, buildThinkingModeDescription, and the xhigh case in formatConfigValueName removed (all dead after mode removal)
• buildSessionPresentation returns modes: { currentModeId: "", availableModes: [] } — ACP bridge exposes no modes
• setSessionMode is a silent no-op (returns {}, never calls gateway)
• setSessionConfigOption("thought_level", ...) case removed — now rejects with "does not support"
• chat.send _meta simplified to readString(params._meta, ["thinking"]) (single key, no thinkingLevel fallback)
• getGatewaySessionRow no longer reads thinkingLevel
• listThinkingLevels import dropped (no remaining callers in translator)

Auto-reply + agents + status + TUI (9 files)

Extensions + docs

• extensions/mattermost/src/mattermost/monitor-helpers.ts — drop from ResponsePrefixContext
• docs/concepts/session-tool.md, docs/reference/session-management-compaction.md, docs/gateway/configuration-reference.md, docs/gateway/configuration-examples.md — drop stale references
• docs/.generated/config-baseline.json — regenerated

Tests (13 files)

• Where tests used thinkingLevel as a generic field for store-update behavior, swapped to verboseLevel (still schema-valid): src/config/sessions.test.ts, src/gateway/server.config-patch.test.ts, src/gateway/server.sessions.gateway-server-sessions-a.test.ts
• Where tests exercised gutted features, rewrote or removed stale cases:
  • src/acp/translator.set-session-mode.test.ts — rewritten for no-op contract
  • src/acp/translator.session-rate-limit.test.ts — thought-level mode tests replaced with no-op assertions
  • src/commands/agent.test.ts — thinkingDefault / /think-related tests removed
• Fixture cleanups in remaining test files

Fork baseline

.fork-boundary-mock-baseline: 133 → 132 (decrease locked in)

Verification

• git grep "thinkingLevel" in src/ → 2 hits, both in migration-only comments (AC-allowed per issue body)
• git grep "thinkingDefault" in src/ → 0 hits
• pnpm check → PASS (format + tsgo + lint + 2 project-specific linters)
• Fork-integrity gates: zombie-import ✓, stub-debt ✓ (132==132), throwing-stub-callers ✓, obsolescence ✓ (49==49)
• Unit tests on directly-affected files → PASS (69/69 in 4 core files; edited tests pass)
• Broader affected suite → 27 pre-existing failures identical to origin/main baseline (30→27 — my changes fix 3 previously-broken thinking tests). No new regressions.
• CI: build, test, lint, docs, rebrand-gate, zombie-import-gate, stub-debt-gate, throwing-stub-callers-gate, obsolescence-audit-gate (pending this PR's run)

Test plan

• pnpm check locally
• Targeted vitest on each modified test file
• Fork-integrity gates
• LIVE=1 pnpm test:live attempted — middleware smoke tests require local CLI credentials (pre-existing, unrelated to thinkingLevel)
• CI green on all required checks

Context

• Parent umbrella: gut(auto-reply): finish thinking-level and model-selection sweep — follow-up to #2335 #2336
• Sibling shipped: gut(auto-reply): partial thinking-level + model-selection sweep — Areas 1+2+4 of #2336 #2463 (Areas 1+2+4)
• Prerequisites merged: gut(ui): remove dead thinkingLevel display plumbing (#2336 Area 7) #2471 (Area 7 Web UI), gut(docs): remove /think directive references — Area 6 of #2336 (#2466) #2472 (Area 6 docs), test(auto-reply): remove runPreparedReply shield + add real-chain regression test (#2468) #2473 (Area 8 tests), gut(auto-reply): delete dead buildEmbeddedRun* cluster in agent-runner-utils.ts (#2469) #2474 (dead cluster)
• Next in chain: remaining umbrella areas per parent issue

🤖 Generated with Claude Code