[claude] Add canonical fleet routing table with agent verdicts (#836) #889

claude · 2026-04-06T18:08:23Z

claude commented

2026-04-06 18:08:23 +00:00

Fixes #836

What this does

Commits the canonical fleet routing table to fleet/fleet-routing.json. This brings the evaluated Gitea users into the repo as a structured, machine-readable routing artifact that workforce-manager.py (Epic #204) can consume.

Changes

fleet/fleet-routing.json — 10 agents evaluated with:
- Routing verdicts per agent (from Timmy's #836 evaluation)
- do_not_route flags for claw-code (protocol bridge), substratum (no model), and allegro-primus (retired)
- Cost-optimized cascade: free → cheap → prepaid
- Task-type map routing specific work types to specific agents
- Gap analysis with next actions

Routing cascade

Tier	Agents	Use for
Free	fenrir, bezalel, carnice	Issue triage, code review, local code gen
Cheap	kimi, allegro	Small edits (≤3 files), triage/routing decisions
Prepaid	ezra	Complex multi-file, docs, architecture

Gaps documented

substratum — no model, inactive until Epic #196
kimi — empty Gitea profile description needs enrichment
claw-code — removed from routing cascade (infrastructure, not endpoint)
No model scoring yet — wolf evaluation (#195) will fix vibes-based routing

Fixes #836 ## What this does Commits the canonical fleet routing table to `fleet/fleet-routing.json`. This brings the evaluated Gitea users into the repo as a structured, machine-readable routing artifact that workforce-manager.py (Epic #204) can consume. ## Changes - **`fleet/fleet-routing.json`** — 10 agents evaluated with: - Routing verdicts per agent (from Timmy's #836 evaluation) - `do_not_route` flags for claw-code (protocol bridge), substratum (no model), and allegro-primus (retired) - Cost-optimized cascade: free → cheap → prepaid - Task-type map routing specific work types to specific agents - Gap analysis with next actions ## Routing cascade | Tier | Agents | Use for | |------|--------|---------| | Free | fenrir, bezalel, carnice | Issue triage, code review, local code gen | | Cheap | kimi, allegro | Small edits (≤3 files), triage/routing decisions | | Prepaid | ezra | Complex multi-file, docs, architecture | ## Gaps documented 1. substratum — no model, inactive until Epic #196 2. kimi — empty Gitea profile description needs enrichment 3. claw-code — removed from routing cascade (infrastructure, not endpoint) 4. No model scoring yet — wolf evaluation (#195) will fix vibes-based routing ## Next - Wire into `workforce-manager.py` (Epic #204) - Run wolf scoring (#195) - Assign model to substratum (#196)

claude added 1 commit 2026-04-06 18:08:24 +00:00

feat: add fleet routing table with evaluated agent verdicts

CI / validate (pull_request) Waiting to run

Details

4721518ac6

Evaluate all 10 Gitea agents and commit canonical routing table to
fleet/fleet-routing.json. Sourced from ~/.hermes/fleet-routing.json
and enriched with routing verdicts, cost-cascade tiers, task-type map,
gap analysis, and next actions from issue #836 evaluation.

Key decisions:
- claw-code marked do_not_route (protocol bridge, not endpoint)
- substratum marked inactive (no model assigned, blocked on #196)
- allegro-primus marked retired
- Cost cascade: free (fenrir/bezalel/carnice) → cheap (kimi/allegro) → prepaid (ezra)
- Task-type map wires issue types to specific agents

Fixes #836

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>