research-mode

name	research-mode
description	Work mode dispatcher for research workflows. 5 modes: exp (running experiments), daily (reading/notes), explore (deep-research exploration), report (slide generation), publish (academic-pipeline). Commands: /mode exp|daily|explore|report|publish|status|end. All output in English. Triggers: /mode, switch mode, start experiment, begin research session, mode status, end session.
metadata	{"data_access_level":"raw","task_type":"open-ended"}

Research Mode

Declares work context and orchestrates session-end log drafting. Each mode activates a set of recommended skills and a log draft template.

Commands

Command	Description
/mode exp	Experiment mode — running code, recording metrics
/mode daily	Daily research — reading papers, taking notes
/mode explore	Exploration mode — systematic literature search via deep-research
/mode report	Report mode — generating slides for group meetings
/mode publish	Publication mode — full academic-pipeline workflow
/mode status	Show current mode + session activity summary
/mode end	End current mode, trigger auto-draft log entry

---

Mode Activation

When a mode is activated, do three things in order:

1. Print the activation banner (English only):

For /mode exp:

═══════════════════════════════════════
  [EXP] Experiment Mode — Active
═══════════════════════════════════════
  git HEAD : <run git_context.py --head silently>
  Tools    : /research-log add (Full mode)
  End      : /mode end → auto-draft log
═══════════════════════════════════════

For /mode daily:

═══════════════════════════════════════
  [DAILY] Daily Research Mode — Active
═══════════════════════════════════════
  Tools    : freeform notes, /research-log add
  End      : /mode end → lightweight log entry
═══════════════════════════════════════

For /mode explore:

═══════════════════════════════════════
  [EXPLORE] Exploration Mode — Active
═══════════════════════════════════════
  Tools    : deep-research (full / socratic / systematic-review / lit-review / fact-check)
  End      : /mode end → auto-draft log from report
═══════════════════════════════════════

For /mode report:

═══════════════════════════════════════
  [REPORT] Report Mode — Active
═══════════════════════════════════════
  Tools    : /report-slides
  End      : /mode end → log slide deck paths
═══════════════════════════════════════

For /mode publish:

═══════════════════════════════════════
  [PUBLISH] Publication Mode — Active
═══════════════════════════════════════
  Tools    : academic-pipeline (/ars-full, /ars-revision-coach, etc.)
  End      : /mode end → log current pipeline stage
═══════════════════════════════════════

2. Route: after the banner, recommend the primary skill to use next (one sentence).

3. Snapshot (exp mode only): silently run git_context.py to capture HEAD before the user starts work. Store the HEAD value to pre-fill the log entry at /mode end.

GIT_CTX="$(find ~/.claude -path "*/research-log/scripts/git_context.py" | head -1)"
GIT_HEAD=$(python "$GIT_CTX" --head 2>/dev/null || echo "")

Historical Experience Check

After activation routing and before research execution, apply the research-log Historical Experience Check at the decision point defined for the active mode. This adds prior evidence to the decision; it does not require mode activation and does not replace the selected mode's primary workflow.

For direct research intent, use the research-log protocol even when no mode is active. Query history only when the project has a non-empty docs/research_log/ journal and the next action is research planning, recommendation, execution, repetition, or diagnosis. Follow the documented types → search → fetch sequence, disclose query failures, and do not treat an error as empty history.

Use the per-mode decision points in references/routing_guide.md. Keep all banners, prompts, and user-visible research-mode output in English.

---

/mode status

Print a compact summary:

Current mode : [MODE]
Started      : HH:MM (or "unknown")
Session work : <1-2 sentence summary of what has been done this session>
Next action  : <suggested next step>

---

/mode end — Auto-Draft Log

When the user runs /mode end, collect session outputs and draft a log entry. Read references/mode_templates.md for the draft template for the active mode.

Flow

1. Collect data (see per-mode rules below)
2. Fill in the template from references/mode_templates.md
3. Present the draft to the user:

   ─────────────────────────────────────────
    Draft log entry — review and confirm
   ─────────────────────────────────────────
   <draft content here>
   ─────────────────────────────────────────
   Confirm? (yes / edit / skip)
   

4. If confirmed → call /research-log add with the pre-filled content
5. If edit → let user modify inline, then confirm again
6. If skip → discard draft, offer to switch modes

Per-Mode Data Collection

exp:

• git HEAD: from snapshot taken at mode start (or re-run git_context.py)
• Changes: run git_context.py --since-log <last-log-file> and use Suggested Changes bullets
• Ask: "What were the key results? (numbers, metrics)"
• Ask: "What are the next steps?"

daily:

• Ask: "What did you read or work on today? (title / topic)"
• Ask: "What was the key insight or takeaway?"
• Ask: "Any open questions or follow-up items?"

explore:

• Extract from deep-research report (if available in session):
  • Research Question → Goal section
  • Executive Summary bullets → Analysis section
• Ask: "What are the next steps based on this research?"

report:

• Collect slide deck output path from report-slides (if available in session)
• Ask: "Who was the intended audience?"
• Ask: "One-sentence conclusion from the presentation?"

publish:

• Ask: "Which pipeline stage did you complete? (1=Research, 2=Write, 2.5=Integrity, 3=Review, 4=Revise, 4.5=Final Integrity, 5=Finalize)"
• Ask: "Stage result? (PASS / PENDING / needs-revision)"
• Ask: "Key next step?"

---

Notes

• Mode is session-scoped. Starting a new Claude Code session clears the active mode.
• Multiple /mode end calls in one session are allowed — each drafts a new log entry.
• If no mode is active when /mode end is called, ask which mode this session was.
• All output text is English. Never output Chinese characters in banners or prompts.