Audited the Ollama reference (no wrapper script — it's the OLLAMA.md doc + inline
HTTP-API call pattern) against the live server (Ollama 0.30.8 on GURU-5070):
- /api/chat + think:false + res['message']['content'] confirmed working (clean
output, no thinking leak) -- the core call pattern is correct.
- All referenced models exist on the server (qwen3:8b, qwen3.6:latest, qwen3:14b,
codestral:22b, nomic-embed-text).
Real bug found + fixed: the "Preferred one-liner" auto-detected the endpoint with
`urlopen(...)` used as a truthiness test. urlopen RAISES URLError on a down host
(proven), so the ternary's fallback branch was dead code -- it crashed on a down
localhost instead of failing over to Beast, and it did a per-call probe that
contradicts the doc's own "read endpoint from identity.json, no probe" rule 30 lines
above. Replaced with the identity.json endpoint+model pattern (also swaps the
hardcoded qwen3:14b for the per-machine prose_model). Validated verbatim end-to-end.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The qwen3:8b routing update inserted footnote lines mid-table in both
the "What Ollama owns" and "When to Use Which Model" sections, splitting
each table in half so renderers treated the qwen3.6 rows as paragraph
text. Moved footnotes below the closing table row in both places.
Also updated the bottom "Rule of thumb" line: previously named qwen3:14b
with a "2x faster" claim that's now stale on DESKTOP-0O8A1RL where 8b is
the prose model. Generalized to "the per-machine prose model".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Benchmarked qwen3.6 (36B MoE) vs qwen3:14b and qwen3:32b on 16
representative prompts. qwen3.6 scored 15/16 vs 14b 11/16 and 32b
12/16, winning every strict-format/adherence test (multi-step rules,
weekend-aware scheduling, prompt-injection resistance, word-limit
summary). Single reasoning regression noted for re-check at qwen3.7.
Updated .claude/OLLAMA.md (Models, Documentation Engine, and
When-to-Use tables) and .claude/CLAUDE.md one-line model summary to
route strict-format work to qwen3.6 and keep bulk prose on qwen3:14b
(2x faster). Also removed openclaw npm package + ~/.openclaw data dir
earlier in the session.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Index was dead since 2026-04-19 (watcher not running). Fixes:
- Watcher restarted; scheduled task registered for login persistence
- Removed .md 0.6x penalty — markdown is primary content in this repo
- Added session-logs/ 1.3x, .claude/ 1.2x, /clients/ 1.1x relevance bonuses
- CLAUDE.md: grepai_search is now the first step for any context lookup
- OLLAMA.md: documents config overrides + watcher setup for new machines
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Windows Store python3 stub returns exit 49 instead of running Python.
Replace with: py (Windows launcher) for actual Python code, jq for
simple JSON extraction. Reorder fallback loops to try py first.
Add Bash(py:*) to settings.local.json allowlist.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Extract Ollama docs and PROJECT_STATE locking protocol to on-demand
reference files. Trim Work Mode to detection table only. Remove verbose
anti-pattern examples and credential encryption details.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
