|
|
996ef2fe40
|
feat(agent): 8-model Fusion panels with Cohere + GLM + hard_wall type
Revised panel selections (1-8 models per Fusion spec):
- lean_proof: 8 models (Opus, DeepSeek, GPT, Cohere, Gemini, GLM, Kimi, DS-Flash)
- code_bug: 6 models
- architecture: 8 models
- math_novel: 6 models
- infra_debug: 5 models
- hard_wall: 8 models (all, for truly stuck problems)
Added:
- Cohere Command A+ (structured output, tool-use optimized)
- GLM-5.2 (753B, genuinely different architecture)
- DeepSeek V4 Flash (fast iterations)
- hard_wall problem type (stuck/wall/impossible/tried everything)
- Budget cap per type (.50-.50)
|
2026-06-22 15:45:21 -05:00 |
|
|
|
0a79e70ff3
|
feat(agent): break-glass MCP agent for novel problem attacks
Last-resort multi-model panel via OpenRouter Fusion. Only invoked
when all other approaches (local model, single frontier, existing
skills) have failed on the same problem.
Architecture:
- Problem classification (lean_proof, code_bug, architecture, math_novel, infra_debug)
- Per-type model panel selection (Opus + DeepSeek + Gemini for proofs)
- Context assembly (files, build logs, AGENTS.md constraints)
- OpenRouter Fusion call (parallel panel + judge synthesis)
- Structured output (diagnosis, solution, alternatives, risks, cost)
Safety:
- Explicit 'break glass:' trigger required
- Cost gate: /usr/bin/bash.30-.50 per invocation, .00/session cap
- Rate limit: 3 invocations per session
- Audit trail via ContextStream
Files:
- 4-Infrastructure/shim/break_glass_mcp.py (MCP server)
- 6-Documentation/docs/specs/break_glass_mcp_agent.md (design spec)
- .opencode/skills/break-glass/SKILL.md (skill definition)
- .mcp.json (server registration)
|
2026-06-22 15:33:16 -05:00 |
|
