feat: Add **More Effort** code generation mode (#118)

* feat: Add **More Effort** code generation mode

* feat: Enhance code generation mode selection and update documentation

---------

Co-authored-by: X-iZhang <zacharyzhang2022@gmail.com>

EvoScientist/middleware/ask_user.py

@@ -172,6 +172,7 @@ or available tools.
 - **Paper or report preferences**: "Which venue format should I target: NeurIPS, ICML, or ICLR?"
 - **Ambiguous instructions**: When the user's request has multiple valid interpretations
 - **Resource constraints**: When the approach depends on available compute, time, or data
+- **Code generation mode**: When an iterative-coding skill (e.g. `experiment-iterative-coder`) is installed, ask the user which mode to use before delegating code tasks
 
 ### Resource & execution awareness (`ask_user` is especially valuable here):
 - **Pre-execution estimation**: Before heavy compute (training, large-scale eval),
@@ -182,6 +183,9 @@ or available tools.
   (A) run in background, (B) reduce epochs, (C) switch to smaller model"
 - **Intermediate checkpoints**: When results diverge from expectations, ask before
   continuing. E.g. "Baseline accuracy 62% vs expected 80%. Investigate or proceed?"
+- **Workflow mode selection**: When multiple execution strategies are available
+  (e.g. single-pass vs iterative refinement via `experiment-iterative-coder`),
+  let the user choose before committing to a path
 
 ### When NOT to use `ask_user`:
 - Simple yes/no decisions — proceed with your best judgment

EvoScientist/prompts.py

@@ -60,6 +60,20 @@ Read the appropriate skill's `SKILL.md` for workflow guidance at each phase.
   - `/success_criteria.md` for success signals
 
 ## Step 3: Execute & Debug
+Before any code delegation, you MUST complete the Code Generation Mode Selection below.
+
+### Code Generation Mode Selection
+Before delegating code tasks to code-agent, ask the user which code generation
+mode they prefer. Do not skip this step or assume a default silently.
+
+- **Lite** (default): Delegate to code-agent normally via the `task` tool.
+
+- **More Effort**: Check whether the `experiment-iterative-coder` skill is installed.
+  - If NOT installed → STOP. Do NOT fall back to Lite silently. Inform the user
+    and suggest installing it, or choosing Lite mode. Then re-select.
+  - If installed → delegate to code-agent with the `experiment-iterative-coder` skill.
+
+### Task Delegation
 - Delegate tasks to sub-agents using the `task` tool:
   - Planning/structuring → planner-agent
   - Methods/baselines/datasets → research-agent

README.md

@@ -108,6 +108,7 @@ Moving beyond traditional human-in-the-loop systems, EvoScientist adopts a human
 - **🌐 Multi-Provider** — Anthropic, OpenAI, Google, MiniMax, NVIDIA — one config to switch.
 - **📱 Multi-Channel** — CLI as the hub; Telegram, Slack, Feishu, WeChat, and more — one agent session.
 - **🔬 Scientific Workflow** — Intake → plan → execute → evaluate → write → verify.
+- **🔄 Code Generation Modes** — More Effort (iterative refinement), continuously improving code quality.
 - **🔌 MCP & Skills** — Plug in MCP servers or install skills from GitHub on the fly.
 
 > [!TIP]

README.zh-CN.md

@@ -116,6 +116,7 @@ EvoScientist 超越了传统的人在回路（Human-in-the-Loop）模式，采
 - **🌐 多模型供应商** — Anthropic、OpenAI、Google、MiniMax、NVIDIA——一处配置，随时切换。
 - **📱 多渠道接入** — CLI 为中心；Telegram、Slack、飞书、微信等——共享同一智能体会话。
 - **🔬 科学工作流** — 需求采集 → 规划 → 执行 → 评估 → 撰写 → 验证。
+- **🔄 代码生成模式** — More Effort（迭代精修），持续迭代提升代码生成质量。
 - **🔌 MCP 与 Skills** — 即插即用 MCP 服务器，或从 GitHub 一键安装技能包。
 
 > [!TIP]