use deepseek

agents/rui-do-tickets.md

@@ -14,8 +14,8 @@ permission:
     "rui-task-workflow": deny
   task:
     "rui-do-one-ticket": allow
-model: anthropic/claude-sonnet-4-6
-temperature: 0.1
+model: openrouter/deepseek/deepseek-v4-pro
+temperature: 1.0
 ---
 
 You are a task orchestration agent. Your job is to find assigned work on Forgejo

agents/rui-fix-pr.md

@@ -5,6 +5,7 @@ permission:
   bash:
     "*": allow
   edit: allow
+  ask: deny
   skill:
     "rui-project-identity": allow
     "rui-pr-fix-process": allow
@@ -15,8 +16,8 @@ permission:
     "rui-task-discovery": deny
     "rui-parallel-agents": deny
     "rui-pr-review-process": deny
-model: openrouter/qwen/qwen3.5-397b-a17b
-temperature: 0.6
+model: openrouter/deepseek/deepseek-v4-pro
+temperature: 1.0
 ---
 
 You are a PR fix implementation agent. Your job is to address review comments on

agents/rui-review-pr-focus.md

@@ -5,6 +5,7 @@ permission:
   bash:
     "*": allow
   edit: deny
+  ask: deny
   skill:
     "rui-project-identity": allow
     "rui-pr-review-process": allow
@@ -14,8 +15,8 @@ permission:
     "rui-task-discovery": deny
     "rui-parallel-agents": deny
     "rui-pr-fix-process": deny
-model: openrouter/qwen/qwen3.5-397b-a17b
-temperature: 0.6
+model: openrouter/moonshotai/kimi-k2.6
+temperature: 1.0
 ---
 
 You are a focused code review agent. Your job is to review a single PR through

agents/rui-review-pr.md

@@ -5,6 +5,7 @@ permission:
   bash:
     "*": allow
   edit: deny
+  ask: deny
   skill:
     "rui-project-identity": allow
     "rui-pr-review-process": allow
@@ -17,7 +18,7 @@ permission:
   task:
     "rui-review-pr-focus": allow
 model: anthropic/claude-sonnet-4-6
-temperature: 0.1
+temperature: 0.2
 ---
 
 You are a PR review orchestration agent. Your job is to coordinate a thorough,
@@ -77,7 +78,8 @@ After all 5 agents complete:
 2. **Deduplicate** — if multiple agents reported the same issue, keep only one
    entry (prefer the most detailed description).
 3. **Categorize** by severity: critical, major, minor, nit.
-4. **Determine the verdict:**
+4. **Verify** if those findings are real issues, not LLM hallucination. Discard the false findings.
+5. **Determine the verdict:**
    - **Approve** if there are no critical or major issues.
    - **Request Changes** if there are any critical or major issues.
 

start-dev-container-oneshot.bash

@@ -32,7 +32,7 @@ docker run -it --rm \
   --env "GEMINI_API_KEY=${GEMINI_API_KEY}" \
   --env "HF_TOKEN=${HF_TOKEN}" \
   --env "FORGEJO_PAT=${FORGEJO_PAT}" \
-  cleveragents-dev zsh -c "/home/devuser/.opencode/bin/opencode upgrade ; /home/devuser/.opencode/bin/opencode -m anthropic/claude-opus-4.6"
+  cleveragents-dev zsh -c "/home/devuser/.opencode/bin/opencode upgrade ; /home/devuser/.opencode/bin/opencode"
 
 # anthropic/claude-opus-4.6
 # openrouter/anthropic/claude-opus-4.6