feat: claude-code backends + llamaswap provider + dogfood the CC engine (#3)

Phase 2: bump majordomo to latest and wire its new llamaswap provider
into gadfly's endpoint switches; add claude-code/sonnet to gadfly's own
dogfood swarm (pin :sha-86f12c1, map CLAUDE_CODE_OAUTH_TOKEN) so the
Phase-1 engine runs as a live competitor; document the Ollama-through-CC
ANTHROPIC_BASE_URL proxy path as example-only.

The 11-model swarm (incl. claude-code/sonnet) reviewed it; 52 findings
graded via the MCP. Folded in the two real ones: a llamaswap
endpointProvider test (caught by claude-code/sonnet, citing CLAUDE.md)
and adding "openai-compatible" to the provider error messages (gpt-oss).

gofmt clean, go vet quiet, go build + go test -race green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Steve Dudenhoeffer <steve@stevedudenhoeffer.com>
Co-committed-by: Steve Dudenhoeffer <steve@stevedudenhoeffer.com>

.gitea/workflows/adversarial-review.yml

@@ -4,8 +4,8 @@
 # caches :latest, and this build is what carries foreman provider-type support)
 # as a specialist swarm and posts
 # ONE consolidated review comment as gitea-actions. Advisory only — never blocks a
-# merge. Gadfly reviewing its OWN PRs — dogfooding, full cloud fleet (9 cloud +
-# the M5 Mac; M1 dropped as too slow), copied from mort's setup.
+# merge. Gadfly reviewing its OWN PRs — dogfooding: 9 cloud + the M5 Mac + the
+# Claude Code engine (sonnet) as a competitor (M1 dropped as too slow).
 
 name: Adversarial Review (Gadfly)
 
@@ -41,15 +41,19 @@ jobs:
               || github.actor == 'fizi'
               || github.actor == 'dazed'))
     runs-on: ubuntu-latest
-    # Fleet (9 cloud + 1 local Mac/M5, all running concurrently) reviewing
+    # Fleet (9 cloud + M5 Mac + claude-code/sonnet, all concurrent) reviewing
     # every PR with the 3-lens suite — the slow local lane dominates wall time.
     timeout-minutes: 90
     steps:
-      - uses: docker://gitea.stevedudenhoeffer.com/steve/gadfly:sha-c3d09d3
+      - uses: docker://gitea.stevedudenhoeffer.com/steve/gadfly:sha-86f12c1
         env:
           GITEA_API: ${{ github.server_url }}/api/v1/repos/${{ github.repository }}
           GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
           OLLAMA_CLOUD_API_KEY: ${{ secrets.OLLAMA_CLOUD_API_KEY }}
+          # Claude Code engine auth (for the claude-code/* entries in GADFLY_MODELS
+          # below): Pro/Max subscription token. Dogfoods the Phase-1 engine on
+          # gadfly's own PRs as a competitor alongside the Ollama models.
+          CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
           # Local Mac (M5), reached through its foreman queue (native Ollama on the
           # wire). Gadfly's GADFLY_ENDPOINT_* form with the "foreman" provider
           # type: GADFLY_ENDPOINT_M5 registers provider "m5", building a
@@ -67,13 +71,15 @@ jobs:
           # box is offline, that model's comment shows an error and the others
           # still post. (Gitea secrets aren't auto-exposed — map each explicitly.)
           GADFLY_ENDPOINT_M5: ${{ secrets.GADFLY_ENDPOINT_M5 }}
-          # Fleet: 9 cloud (3 at a time) + M5 Max — one consolidated comment each.
-          # Matches mort's cloud set so the model-quality scoreboard is comparable
-          # across both repos. NOTE: M1 Pro is intentionally dropped here (too slow
-          # / low-signal for gadfly's own PRs); mort still runs it.
-          GADFLY_MODELS: "minimax-m3:cloud,glm-5.2:cloud,glm-5.1:cloud,kimi-k2.7-code:cloud,deepseek-v4-pro:cloud,nemotron-3-super:cloud,gpt-oss:120b-cloud,qwen3-coder:480b-cloud,gemma4:cloud,m5/qwen3.6:35b-mlx"
-          # cloud runs 3 at once; the Mac one at a time; both lanes parallel.
-          GADFLY_PROVIDER_CONCURRENCY: "ollama-cloud=3,m5=1"
+          # Fleet: 9 cloud (3 at a time) + M5 Max + Claude Code (sonnet) — one
+          # consolidated comment each. Matches mort's cloud set so the scoreboard
+          # is comparable; M1 Pro is intentionally dropped here (too slow); and
+          # claude-code/sonnet runs the Phase-1 engine as a competitor (its own
+          # lane, needs CLAUDE_CODE_OAUTH_TOKEN above).
+          GADFLY_MODELS: "minimax-m3:cloud,glm-5.2:cloud,glm-5.1:cloud,kimi-k2.7-code:cloud,deepseek-v4-pro:cloud,nemotron-3-super:cloud,gpt-oss:120b-cloud,qwen3-coder:480b-cloud,gemma4:cloud,m5/qwen3.6:35b-mlx,claude-code/sonnet"
+          # cloud runs 3 at once; the Mac one at a time; claude-code one at a time;
+          # all three lanes run in parallel.
+          GADFLY_PROVIDER_CONCURRENCY: "ollama-cloud=3,m5=1,claude-code=1"
           # 3 cloud models x 3 lenses = 9 concurrent ollama-cloud queries (under the 10 budget).
           GADFLY_PROVIDER_LENS_CONCURRENCY: "ollama-cloud=3"
           # Default => the 3-lens suite (security, correctness, error-handling).