research-loop-resume

Research Loop Resume

Load .research/ state and inject a compact research context into the conversation.

Preconditions

• Current directory is a git repository.
• .research/IDEA.md exists.
• .research/tree.md exists.
• .research/experiments/ exists, possibly empty.

If state is missing, stop and tell the user to initialize a research loop first.

Workflow

1. Read .research/IDEA.md.
2. Read .research/tree.md and extract all hypotheses, status values, and evidence IDs.
3. Scan .research/experiments/*.md. For each experiment, extract only:
   • experiment ID
   • linked hypothesis
   • date
   • status
   • conclusion summary
4. Read .research/DASHBOARD.md if present.
5. Produce a structured context summary of about 1000-2000 Chinese characters.

Output Shape

# 研究上下文

## 研究动机
[IDEA one-liner plus short background]

## 假设树
- H1: [description] (被支持, Evidence: E001)
  - H1.1: [description] (进行中)
- H2: [description] (待验)

## 实验历史
- E001 ([date]): 验证 H1, 结论=[support/refute/uncertain summary]

## Recent Validations
- [date] H1 [被支持]: [summary]

## Next Steps
1. [next step from DASHBOARD or analysis]

## State Files
IDEA:      [absolute path]/.research/IDEA.md
Tree:      [absolute path]/.research/tree.md
Dashboard: [absolute path]/.research/DASHBOARD.md

Rules

• Summarize experiments; do not dump long execution logs.
• Do not modify files.
• If DASHBOARD.md is absent, continue without it and say so.
• If tree.md has invalid status vocabulary, surface the format error instead of guessing.