fix: default to deepseek/deepseek-v4-flash-0731 (drop invalid openrouter/ prefix)

.env.example

@@ -19,7 +19,7 @@ NODE_ID=pr-af
 PR_AF_PROVIDER=aforge
 # Read by the Go node's SDK adapter; the pinned Python SDK always runs `exec`.
 AGENTFIELD_AFORGE_COMMAND=exec
-PR_AF_MODEL=openrouter/moonshotai/kimi-k2.5
+PR_AF_MODEL=deepseek/deepseek-v4-flash-0731
 # Falls back to PR_AF_MODEL when unset
 # PR_AF_AI_MODEL=
 PR_AF_MAX_TURNS=50

Dockerfile

@@ -67,7 +67,7 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
     AGENTFIELD_SERVER=http://agentfield:8080 \
     PR_AF_PROVIDER=aforge \
     AGENTFIELD_AFORGE_COMMAND=exec \
-    PR_AF_MODEL=openrouter/moonshotai/kimi-k2.5 \
+    PR_AF_MODEL=deepseek/deepseek-v4-flash-0731 \
     PORT=8004 \
     HOME=/home/praf \
     PYTHONPATH=/app/src \

README.md

@@ -267,7 +267,7 @@ The key knobs (see `.env.example` for the full list):
 | `AGENTFIELD_AFORGE_COMMAND` | AForge headless command (default `exec`) — read by the Go node's SDK adapter; the pinned Python SDK always runs `exec` |
 | `PR_AF_AFORGE_BIN`          | Path to an aforge-v2 binary (default `aforge`)                 |
 | `PR_AF_HARNESS_BIN`         | Provider-agnostic executable override                          |
-| `PR_AF_MODEL`               | Harness model (default `openrouter/moonshotai/kimi-k2.5`)      |
+| `PR_AF_MODEL`               | Harness model (default `deepseek/deepseek-v4-flash-0731`)      |
 | `PR_AF_MAX_COST_USD`        | Per-run cost ceiling in USD (default `2.0`)                    |
 | `PR_AF_MAX_DURATION_SECONDS`| Per-run wall-clock ceiling in seconds (default `3600`)         |
 | `AGENTFIELD_HARNESS_IDLE_SECONDS` | Harness no-output watchdog window in seconds (default `360`) — harness CLIs in JSON mode emit events only at completion boundaries, so long single completions look silent |

agentfield-package.yaml

@@ -58,7 +58,7 @@ user_environment:
       description: Optional path to the aforge-v2 binary (defaults to aforge on PATH)
     - name: PR_AF_MODEL
       description: Model the harness uses
-      default: openrouter/moonshotai/kimi-k2.5
+      default: deepseek/deepseek-v4-flash-0731
     - name: PR_AF_LABEL
       description: GitHub pull-request label that triggers a webhook review
       default: pr-af

docker-compose.go.yml

@@ -42,7 +42,7 @@ services:
       - AGENT_CALLBACK_URL=http://pr-af-go:8007
       - PR_AF_PROVIDER=${PR_AF_PROVIDER:-aforge}
       - AGENTFIELD_AFORGE_COMMAND=${AGENTFIELD_AFORGE_COMMAND:-exec}
-      - PR_AF_MODEL=${PR_AF_MODEL:-openrouter/moonshotai/kimi-k2.5}
+      - PR_AF_MODEL=${PR_AF_MODEL:-deepseek/deepseek-v4-flash-0731}
       # Optional generic executable override; leave unset for provider defaults.
       - PR_AF_HARNESS_BIN
       - OPENROUTER_API_KEY=${OPENROUTER_API_KEY}

docker-compose.yml

@@ -24,7 +24,7 @@ services:
       - AGENT_CALLBACK_URL=http://pr-af:8004
       - PR_AF_PROVIDER=${PR_AF_PROVIDER:-aforge}
       - AGENTFIELD_AFORGE_COMMAND=${AGENTFIELD_AFORGE_COMMAND:-exec}
-      - PR_AF_MODEL=${PR_AF_MODEL:-openrouter/moonshotai/kimi-k2.5}
+      - PR_AF_MODEL=${PR_AF_MODEL:-deepseek/deepseek-v4-flash-0731}
       - OPENROUTER_API_KEY=${OPENROUTER_API_KEY}
       - GH_TOKEN=${GH_TOKEN:-}
       - XDG_DATA_HOME=/home/praf/.local/share

docker-entrypoint.sh

@@ -10,7 +10,7 @@
 set -e
 
 # Default matches the image ENV / benchmarked model.
-MODEL="${PR_AF_MODEL:-openrouter/moonshotai/kimi-k2.5}"
+MODEL="${PR_AF_MODEL:-deepseek/deepseek-v4-flash-0731}"
 
 # opencode keys models under a provider by the slug *without* the provider
 # prefix, e.g. "openrouter/z-ai/glm-5.2" -> provider "openrouter", key "z-ai/glm-5.2".

go/Dockerfile

@@ -77,7 +77,7 @@ ARG OPENCODE_VERSION=1.17.15
 ENV AGENTFIELD_SERVER=http://agentfield:8080 \
     PR_AF_PROVIDER=aforge \
     AGENTFIELD_AFORGE_COMMAND=exec \
-    PR_AF_MODEL=openrouter/moonshotai/kimi-k2.5 \
+    PR_AF_MODEL=deepseek/deepseek-v4-flash-0731 \
     PORT=8007 \
     NODE_ID=pr-af \
     HOME=/home/praf \

go/README.md

@@ -157,7 +157,7 @@ The node is configured entirely through the environment.
 | `PORT`                      | Listen port (default `8007`)                                   |
 | `PR_AF_PROVIDER`            | Harness provider (default `aforge`; use `opencode` to roll back) |
 | `AGENTFIELD_AFORGE_COMMAND` | AForge headless command — `exec` (default) or `do`             |
-| `PR_AF_MODEL`               | Harness model (default `openrouter/moonshotai/kimi-k2.5`)      |
+| `PR_AF_MODEL`               | Harness model (default `deepseek/deepseek-v4-flash-0731`)      |
 | `PR_AF_LABEL`               | Pull-request label that triggers a webhook review (default `pr-af`) |
 | `PR_AF_MAX_CONCURRENT_REVIEWERS` | Optional webhook review concurrency cap (minimum `1`)     |
 | `PR_AF_MAX_REVIEW_DEPTH`    | Optional webhook sub-review depth cap (minimum `0`)            |
@@ -174,7 +174,7 @@ time from `https://agentfield.ai/downloads/aforge`, pinned by the
 installed and can be selected with `PR_AF_PROVIDER=opencode` without rebuilding.
 
 Note: the code default model is `minimax/minimax-m2.5`, while the Docker image /
-compose / manifest set `PR_AF_MODEL=openrouter/moonshotai/kimi-k2.5`. The env
+compose / manifest set `PR_AF_MODEL=deepseek/deepseek-v4-flash-0731`. The env
 var always wins; both defaults are intentional (they mirror the Python node).
 
 ## Deployment: `af install`

go/agentfield-package.yaml

@@ -53,7 +53,7 @@ user_environment:
       default: exec
     - name: PR_AF_MODEL
       description: harness model
-      default: openrouter/moonshotai/kimi-k2.5
+      default: deepseek/deepseek-v4-flash-0731
     - name: PR_AF_LABEL
       description: GitHub pull-request label that triggers a webhook review
       default: pr-af

go/docker-entrypoint.sh

@@ -10,7 +10,7 @@
 set -e
 
 # Default matches the image ENV / benchmarked model.
-MODEL="${PR_AF_MODEL:-openrouter/moonshotai/kimi-k2.5}"
+MODEL="${PR_AF_MODEL:-deepseek/deepseek-v4-flash-0731}"
 
 # opencode keys models under a provider by the slug *without* the provider
 # prefix, e.g. "openrouter/z-ai/glm-5.2" -> provider "openrouter", key "z-ai/glm-5.2".

go/internal/config/ai.go

@@ -15,7 +15,7 @@ import (
 // AIIntegrationConfig ports config.py AIIntegrationConfig. Env precedence and
 // defaults per design §C.2. The HarnessModel/AIModel fallback is the CODE
 // default "minimax/minimax-m2.5" — deliberately different from the
-// Docker/compose/manifest default (openrouter/moonshotai/kimi-k2.5); env always
+// Docker/compose/manifest default (deepseek/deepseek-v4-flash-0731); env always
 // wins, and both facts ship intentionally (design §B.6). Do not "fix" it.
 type AIIntegrationConfig struct {
 	Provider              string  `json:"provider"`

go/test/functional/compose.functional.yml

@@ -51,7 +51,7 @@ services:
       - AGENT_CALLBACK_URL=http://pr-af:8007
       - PR_AF_PROVIDER=${PR_AF_PROVIDER:-aforge}
       - AGENTFIELD_AFORGE_COMMAND=${AGENTFIELD_AFORGE_COMMAND:-exec}
-      - PR_AF_MODEL=${PR_AF_MODEL:-openrouter/moonshotai/kimi-k2.5}
+      - PR_AF_MODEL=${PR_AF_MODEL:-deepseek/deepseek-v4-flash-0731}
       - OPENROUTER_API_KEY=${OPENROUTER_API_KEY:-}
       - GH_TOKEN=${GH_TOKEN:-}
     ports: