Skip to content

claw: model_route.t27 -- Opus 5.5 orchestrates, Sonnet 5.5 runs sub-tasks (Closes #10492) - #10493

Merged
gHashTag merged 1 commit into
masterfrom
claude/model-route-10492
Oct 18, 2026
Merged

gHashTag merged 1 commit into
masterfrom
claude/model-route-10492

Conversation

@gHashTag

Copy link
Copy Markdown
Owner

Closes #10492

Owner directive 2026-10-18: orchestration on Opus 5.5, sub-tasks on Sonnet 5.5, as a TRI CLAW rule.

  • specs/claw/model_route.t27: roles (orchestrate, plan, review-final -> Opus; implement, port, search, test, docs, fix -> Sonnet), route() escalates a sub-task to Opus only after 2 failed attempts or the owner's flag, is_waste() audits a run. Evidence for the split: epic epic: t27 vs direct Verilog on spec-to-RTL -- close the model-matrix gap (Build-A-Chip-Bench proxy) #10066 (Sonnet 5.5 >= Opus 5.5 on ten bounded tasks at ~40% of the cost).
  • Seal .trinity/seals/claw_ClawModelRoute.json.
  • .claude/settings.json: CLAUDE_CODE_SUBAGENT_MODEL=claude-sonnet-5-5.
  • model: in every .claude/agents/*.md: trinity.md (Queen, orchestrator) on opus, the ten sub-task agents on sonnet.
  • Skill .claude/skills/model-routing/SKILL.md, a pointer in CLAUDE.md, a row in docs/claw/README.md.

Checked locally (t27c 0.5.2 built from master)

  • t27c test-report specs/claw/model_route.t27: 6 of 6 pass, 0 vacuous, invariant proved.
  • Negative control: routing orchestration to Sonnet fails 3 of 6 tests.
  • t27c seal specs/claw/model_route.t27 --verify: all hashes MATCH.
  • t27_check (t27-mcp, the compiler front end in wasm): 0 errors, 0 warnings.

The commit-time tri gates did not run (no tri binary in this worktree); CI runs them.

🤖 Generated with Claude Code

…asks (Closes #10492)

Owner directive 2026-10-18. The law is specs/claw/model_route.t27 (6/6 tests,
0 vacuous, invariant proved; the orchestration-to-Sonnet mutant fails 3).
Claude Code follows it through CLAUDE_CODE_SUBAGENT_MODEL in
.claude/settings.json, model: in every .claude/agents/*.md (Queen on opus,
the rest on sonnet) and the model-routing skill.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-18 15:15:03 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 49
PRs with All Checks Green 1
READY 0
FAILING 49
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 49 + 0 + 0 = 49, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=fad12492937a != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

claw: model_route.t27 -- Opus 5.5 orchestrates, Sonnet 5.5 runs sub-tasks; skill model-routing and agent models

2 participants