Post
1417
After the Feb/Mar '26 collapse of Claude Code I started building my own framework. The token crunch is still only mitigated, but the reasoning and quality are back ā better than before. For research and LLM
training recipes, though, diverse knowledge and a second point of view are crucial. Pairing my Claude Max sub with an Ollama Pro sub has already saved me from days of botched trainings ā multiple frontier
models helping Claude is next level. Acting as the middleman myself was interesting but inefficient, so I shipped skills that let Claude talk to Ollama models directly.
š claude-hooks v1.1.0 ships two LLM-to-LLM skills.
š¬ /get-advice ā single-shot second opinion. Claude runs a multi-turn conversation with a configured Ollama advisor; the advisor grounds in your project through read_file / grep / glob / list_files /
recall_memory tools. Effort tiers cap fresh-session retries.
š¤ /consultants ā multi-agent council for cross-cutting questions:
š§© planner ā researcher ā critic ā synthesizer
Each role runs its own Ollama model. š¾ Sessions persist to disk (summary.md + transcript.md + SQLite per-role message threads); closed sessions reopen and produce follow-ups š indistinguishable from warm
ones.
šÆ x-tier effort multiplies diversity:
ā¢ xmedium / xhigh ā researcher fans across N models in parallel
ā¢ xmax ā + multi-critic + meta-critic combine; critics anonymized as "Critic 1/2/3" to avoid model-bias
š”ļø Cloud-flap recovery, three layers: 15-attempt / ~15min retry budget; synthesizer failure-fallback model chain; degraded-answer composer surfaces researcher + critic work even when synthesis fails.
š 7 cloud models benchmarked & Claude-graded on locked queries:
ā¢ PROD-READY (P:A R:A C:A S:A): kimi-k2.6, gemma4:31b, glm-5.1
ā¢ Role specialists: minimax-m2.7 (critic), qwen3.5 (planner)
š§ Linux/macOS/Windows. No per-project setup.
š github.com/mann1x/claude-hooks
training recipes, though, diverse knowledge and a second point of view are crucial. Pairing my Claude Max sub with an Ollama Pro sub has already saved me from days of botched trainings ā multiple frontier
models helping Claude is next level. Acting as the middleman myself was interesting but inefficient, so I shipped skills that let Claude talk to Ollama models directly.
š claude-hooks v1.1.0 ships two LLM-to-LLM skills.
š¬ /get-advice ā single-shot second opinion. Claude runs a multi-turn conversation with a configured Ollama advisor; the advisor grounds in your project through read_file / grep / glob / list_files /
recall_memory tools. Effort tiers cap fresh-session retries.
š¤ /consultants ā multi-agent council for cross-cutting questions:
š§© planner ā researcher ā critic ā synthesizer
Each role runs its own Ollama model. š¾ Sessions persist to disk (summary.md + transcript.md + SQLite per-role message threads); closed sessions reopen and produce follow-ups š indistinguishable from warm
ones.
šÆ x-tier effort multiplies diversity:
ā¢ xmedium / xhigh ā researcher fans across N models in parallel
ā¢ xmax ā + multi-critic + meta-critic combine; critics anonymized as "Critic 1/2/3" to avoid model-bias
š”ļø Cloud-flap recovery, three layers: 15-attempt / ~15min retry budget; synthesizer failure-fallback model chain; degraded-answer composer surfaces researcher + critic work even when synthesis fails.
š 7 cloud models benchmarked & Claude-graded on locked queries:
ā¢ PROD-READY (P:A R:A C:A S:A): kimi-k2.6, gemma4:31b, glm-5.1
ā¢ Role specialists: minimax-m2.7 (critic), qwen3.5 (planner)
š§ Linux/macOS/Windows. No per-project setup.
š github.com/mann1x/claude-hooks