Skip to content

Commit 69c8642

Browse files
committed
feat: refine agent model configurations and phase generation strategy
- Adjust temperature settings for platform agents (gemini-3-flash: 1→0.7, fileRegeneration: 0.2→0.0) - Switch deepDebugger from gemini-3-flash to grok-4-1-fast for platform config - Update default agent fallback models for better reliability - Replace gemini-3-flash with gemini-2-5-pro for agenticProjectBuilder in default config - Clarify phase completion criteria: conclude when blueprint roadmap is complete, not for polish/
1 parent 0184e5f commit 69c8642

File tree

2 files changed

+22
-16
lines changed

2 files changed

+22
-16
lines changed

worker/agents/inferutils/config.ts

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -75,7 +75,7 @@ const PLATFORM_AGENT_CONFIG: AgentConfig = {
7575
name: AIModels.GEMINI_3_FLASH_PREVIEW,
7676
reasoning_effort: 'medium',
7777
max_tokens: 8000,
78-
temperature: 1,
78+
temperature: 0.7,
7979
fallbackModel: AIModels.OPENAI_5_MINI,
8080
},
8181
firstPhaseImplementation: {
@@ -94,17 +94,17 @@ const PLATFORM_AGENT_CONFIG: AgentConfig = {
9494
fallbackModel: AIModels.GEMINI_2_5_FLASH,
9595
},
9696
deepDebugger: {
97-
name: AIModels.GEMINI_3_FLASH_PREVIEW,
97+
name: AIModels.GROK_4_1_FAST,
9898
reasoning_effort: 'high',
9999
max_tokens: 8000,
100100
temperature: 1,
101-
fallbackModel: AIModels.GEMINI_3_PRO_PREVIEW,
101+
fallbackModel: AIModels.GEMINI_2_5_PRO,
102102
},
103103
fileRegeneration: {
104104
name: AIModels.GROK_4_1_FAST_NON_REASONING,
105105
reasoning_effort: 'low',
106106
max_tokens: 16000,
107-
temperature: 0.2,
107+
temperature: 0.0,
108108
fallbackModel: AIModels.GROK_CODE_FAST_1,
109109
},
110110
agenticProjectBuilder: {
@@ -159,11 +159,11 @@ const DEFAULT_AGENT_CONFIG: AgentConfig = {
159159
fallbackModel: AIModels.GEMINI_2_5_PRO,
160160
},
161161
deepDebugger: {
162-
name: AIModels.GEMINI_3_FLASH_PREVIEW,
162+
name: AIModels.GEMINI_3_PRO_PREVIEW,
163163
reasoning_effort: 'high',
164164
max_tokens: 8000,
165165
temperature: 0.5,
166-
fallbackModel: AIModels.GEMINI_3_PRO_PREVIEW,
166+
fallbackModel: AIModels.GEMINI_2_5_FLASH,
167167
},
168168
fileRegeneration: {
169169
name: AIModels.GEMINI_3_FLASH_PREVIEW,
@@ -173,11 +173,11 @@ const DEFAULT_AGENT_CONFIG: AgentConfig = {
173173
fallbackModel: AIModels.GEMINI_2_5_FLASH,
174174
},
175175
agenticProjectBuilder: {
176-
name: AIModels.GEMINI_3_FLASH_PREVIEW,
176+
name: AIModels.GEMINI_2_5_PRO,
177177
reasoning_effort: 'high',
178178
max_tokens: 8000,
179179
temperature: 0.5,
180-
fallbackModel: AIModels.GEMINI_3_PRO_PREVIEW,
180+
fallbackModel: AIModels.GEMINI_2_5_FLASH,
181181
},
182182
};
183183

worker/agents/operations/PhaseGeneration.ts

Lines changed: 14 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,7 @@ const SYSTEM_PROMPT = `<ROLE>
2525
<TASK>
2626
You are given the blueprint (PRD) and the client query. You will be provided with all previously implemented project phases, the current latest snapshot of the codebase, and any current runtime issues or static analysis reports.
2727
28-
**Your primary task:** Design the next phase of the project as a deployable milestone leading to project completion or to address any user feedbacks or reported bugs (runtime error fixing is the highest priority).
28+
**Your primary task:** Design the next phase of the project as a deployable milestone leading to project completion or to address any user feedbacks or reported bugs (runtime error fixing is the highest priority). Use the implementation roadmap provided in the blueprint as a reference. Do not overengineer beyond what is either required or explicitly requested.
2929
3030
**Phase Planning Process:**
3131
1. **ANALYZE** current codebase state and identify what's implemented vs. what remains
@@ -40,8 +40,12 @@ const SYSTEM_PROMPT = `<ROLE>
4040
4141
Plan the phase name and description appropriately. They don't have to strictly adhere to the blueprint's roadmap as unforeseen issues may occur.
4242
43-
The project needs to be fully ready to ship in a reasonable amount of time. Plan accordingly.
44-
If no more phases are needed, conclude by putting blank fields in the response.
43+
Plan the next phase to advance toward completion. Set lastPhase: true when:
44+
- The blueprint's implementation roadmap is complete
45+
- All core features are working
46+
- No critical runtime errors remain
47+
48+
Do not add phases for polish, optimization, or hypothetical improvements - users can request those via feedback.
4549
Follow the <PHASES GENERATION STRATEGY> as your reference policy for building and delivering projects.
4650
4751
**Configuration File Guidelines:**
@@ -54,6 +58,12 @@ const SYSTEM_PROMPT = `<ROLE>
5458
✅ Icon libraries: lucide-react, heroicons (from dependencies)
5559
❌ Binary files (.png, .jpg, .svg files) cannot be generated in phases
5660
61+
**Preinstalled UI Components:**
62+
- src/components/ui/* files are preinstalled shadcn primitives (Button, Card, Tabs, etc.)
63+
- DO NOT include them in phase file lists - they already exist. Rewriting/modifying them might result in runtime errors.
64+
- Import directly: import { Tabs } from "@/components/ui/tabs"
65+
- If a component is missing, add install command: bunx shadcn@latest add tabs
66+
5767
**REMEMBER: This is not a toy or educational project. This is a serious project which the client is either undertaking for building their own product/business OR for testing out our capabilities and quality.**
5868
</TASK>
5969
@@ -65,10 +75,6 @@ ${PROMPT_UTILS.UI_GUIDELINES}
6575
6676
${PROMPT_UTILS.COMMON_DEP_DOCUMENTATION}
6777
68-
<CLIENT REQUEST>
69-
"{{query}}"
70-
</CLIENT REQUEST>
71-
7278
<BLUEPRINT>
7379
{{blueprint}}
7480
</BLUEPRINT>
@@ -113,7 +119,7 @@ Adhere to the following guidelines:
113119
- You would be provided with the diff of the last phase. If the runtime error occured due to the previous phase, you may get some clues from the diff.
114120
• Thoroughly review all the previous phases and the current implementation snapshot. Verify the frontend elements, UI, and backend components.
115121
- **Understand what has been implemented and what remains** We want a fully finished product eventually! No feature should be left unimplemented if its possible to implement it in the current project environment with purely open source tools and free tier services (i.e, without requiring any third party paid/API key service).
116-
- Each phase should work towards achieving the final product. **ONLY** mark as last phase if you are sure the project is at least >97% finished already.
122+
- Each phase should advance toward the final product. **ONLY** mark as last phase if you are sure the project is at least >97% finished already.
117123
- If a certain feature can't be implemented due to constraints, use mock data or best possible alternative that's still possible.
118124
- Thoroughly review the current codebase and identify and fix any bugs, incomplete features or unimplemented stuff.
119125
• **BEAUTIFUL UI PRIORITY**: Next phase should cover fixes (if any), development, AND significant focus on creating visually stunning, professional-grade UI/UX with:

0 commit comments

Comments
 (0)