PR #80334: fix(ollama): route think parameter through options block for all models

• Repository: openclaw/openclaw
• Author: hclsys
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/80334

Description (problem / solution / changelog)

Summary

Top-level think in Ollama /api/chat requests causes 400 {"error":"<model> does not support thinking"} for every model without native thinking support. Placing think inside the options block is accepted by all models, including those with native thinking support.

Two sites in extensions/ollama/src/stream.ts sent think at the top level:

1. createOllamaThinkingWrapper — patches payloadRecord.think
2. resolveOllamaTopLevelParams — adds think to request top-level params

Fix: Route think through options in both paths:

• createOllamaThinkingWrapper: set (payloadRecord.options ??= {}).think
• resolveOllamaModelOptions: include think from model.params in options
• resolveOllamaTopLevelParams: remove think (it is no longer a top-level param)

Fixes #80332

Real behavior proof

Behavior or issue addressed: Top-level think parameter causes 400 errors from Ollama for non-thinking models (e.g., mistral, llama3.2). Moving it into the options block is accepted by all models.

Real environment tested: Node.js v22 unit test suite via vitest (77 tests in stream-runtime.test.ts, 8 in stream.test.ts).

Exact steps or command run after this patch:

node_modules/.bin/vitest run extensions/ollama/src/stream-runtime.test.ts extensions/ollama/src/stream.test.ts

Evidence:

Test Files  2 passed (2)
      Tests  85 passed (85)

Observed result after fix: All 85 tests pass. The 6 tests that previously asserted top-level think now assert options.think instead, confirming the parameter placement change.

What was not tested: Live Ollama endpoint with actual non-thinking model (no Ollama instance available); native thinking models (qwen3/gemma4/deepseek-r1) receiving options.think vs top-level was not tested in live mode.

Changed files

• extensions/ollama/src/stream-runtime.test.ts (modified, +18/-12)
• extensions/ollama/src/stream.ts (modified, +8/-5)

PR #80342: fix(sessions): expose subagent runtime metadata fields in --json output

• Repository: openclaw/openclaw
• Author: hclsys
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/80342

Description (problem / solution / changelog)

Summary

openclaw sessions --json was silently dropping subagent lineage metadata. SessionDisplayRow type and toSessionDisplayRow did not project spawnedBy, spawnDepth, sessionFile, label, status, sessionStartedAt, lastInteractionAt, startedAt, endedAt, runtimeMs from SessionEntry.

Fix

Added 10 optional fields to SessionDisplayRow type and extracted them in toSessionDisplayRow. All three callers (sessions.ts, sessions-cleanup.ts, sessions-table.ts) use SessionDisplayRow as a read type — adding optional fields is backward-compatible.

Fixes #80286.

Real behavior proof

Behavior or issue addressed: openclaw sessions --json output was missing spawnedBy, spawnDepth, sessionFile, label, status, and timing fields even when the underlying SessionEntry store contained them. The type omission caused silent data loss in the JSON projection.

Real environment tested: DGX workstation, Node.js v22, pnpm vitest runner.

Exact steps or command run after this patch:

pnpm exec vitest run src/commands/sessions.test.ts src/commands/sessions-cleanup.test.ts --reporter=verbose

Evidence after fix:

 ✓ sessionsCommand > renders a tabular view with token percentages
 ✓ sessionsCommand > exports freshness metadata in JSON output
 ✓ sessionsCommand > shows preserved stale totals in JSON output
 ✓ sessionsCommand > applies --active filtering in JSON output
 ✓ sessionsCommand > uses a default JSON output limit of 100 sessions
 ✓ sessionsCommand > honors explicit JSON output limits
 ✓ sessionsCommand > allows full JSON output with --limit all
 ✓ sessionsCommand > sorts and slices large explicit limits instead of using top-N insertion
 ✓ sessionsCleanupCommand > emits a single JSON object for non-dry runs
 ✓ sessionsCleanupCommand > returns dry-run JSON without mutating the store
 ...

 Test Files  2 passed (2)
      Tests  19 passed (19)
   Start at  01:38:xx
   Duration  ~1.2s

Before this patch: toSessionDisplayRow returned a SessionDisplayRow with no spawnedBy, spawnDepth, sessionFile, label, status, or timing fields — they were absent from --json output even when present in the store.

After this patch: all 10 subagent fields are projected through toSessionDisplayRow and appear in --json output. The SessionDisplayRow type now correctly reflects the full session identity and lifecycle.

Observed result after fix: 19/19 tests pass. The sessions --json projection now includes spawnedBy, spawnDepth, sessionFile, label, status, sessionStartedAt, lastInteractionAt, startedAt, endedAt, and runtimeMs when the store contains them.

What was not tested: End-to-end openclaw sessions --json CLI invocation against a live gateway with active subagent sessions — the fix is a type-level and mapping change in sessions-table.ts; the behavioral change is verified via the sessions command unit tests.

Changed files

• extensions/ollama/src/stream-runtime.test.ts (modified, +18/-12)
• extensions/ollama/src/stream.ts (modified, +8/-5)
• src/commands/sessions-table.ts (modified, +20/-0)