| name | state-overview |
| description | Load the current research state for a research line and present a structured overview. Use at the start of any research session, when switching lines, or when the user asks "what's the state of X?". |
State Overview
Load and present the current state of a research line from the research knowledge layer. This is the agent's "boot-up" — it must be done before any research reasoning.
Context Budget
MUST load:
research-knowledge/INDEX.md
research-knowledge/lines/<line>/STATE.md (full)
research-knowledge/lines/<line>/hypotheses.md — one-liner summary per hypothesis only (not full evidence chains)
research-knowledge/lines/<line>/open-questions.md — high-priority questions only
MUST NOT load:
- Individual paper insight files (load on demand if a specific paper is referenced)
- Full derivation or experiment records
- methodology files
- Other research lines' STATE files (unless cross-line task)
Rule: The overview should fit in one screen. Don't deep-dive into evidence details — that's for the specific skill that needs them.
When to Use
- At the start of any research session (read CLAUDE.md §0 first)
- When switching between research lines
- When the user asks "现在什么状态", "progress", "overview"
- Before proposing new hypotheses or experiments
Workflow
1. Read INDEX first
Always start from research-knowledge/INDEX.md to understand the global research landscape.
2. Load the target line
Read in order:
research-knowledge/lines/<line>/STATE.md — current facts, hypotheses summary, next actions
research-knowledge/lines/<line>/hypotheses.md — detailed hypothesis tracking with evidence
research-knowledge/lines/<line>/open-questions.md — prioritized open questions
3. Check cross-line connections
If the line has cross-line links (check STATE.md frontmatter and cross-cutting/connections.md), note them — they may be relevant to the current task.
4. Present structured overview
Output a concise summary covering:
## <line-name> — 当前状态
### 已确立
- <3-5 key established facts with sources>
### 活跃假说 (N)
- H1: <one-liner> [evidence_type: theoretical|experimental|mixed] [状态: active|stalled|validated]
### 最优先的开放问题
- Q1: <one-liner> (阻塞因素: <what's blocking>)
### 下一步建议
- <ranked next actions>
5. Act on the overview
After presenting, don't wait — if the research context makes the next action obvious, propose it directly:
- "最自然的下一步是 X,要继续吗?"
- If there's new data or a paper to process, suggest running
paper-to-insight or experiment-analyzer
Key Rules
- Always present state before reasoning — don't make research suggestions without first loading STATE.md
- If STATE.md hasn't been updated in a while, note this but don't block on it
- The overview should fit in one screen — prioritize, don't dump everything
- Don't re-read files you just read in the same session — use what's in context