Current State (auto-injected)

!cat .vibe-science/STATE.md 2>/dev/null || echo "No STATE.md — run /vibe-science:init first"

Tree State

!cat .vibe-science/STATE.md 2>/dev/null | grep -A5 "Tree\|Stage\|Node" || echo "No tree state"

vibe-science:loop

Execute one cycle of the Vibe Science research loop.

When to Use

After /vibe-science:init or to continue an existing session.

Usage

/vibe-science:loop

What It Does

One complete cycle of the research loop:

1. OBSERVE → 2. THINK → 3. ACT → 4. EVALUATE → 5. CHECK STOP

Process

Step 1: Crystallize State

Read current state and update:

Reading .vibe-science/STATE.md...

Current state:
- RQ: RQ-001-transport-integration
- Phase: discovery
- Cycle: 7
- unreviewed_claims_pending: 2

Resuming from: "Search PubMed for data integration methodology details"

Step 2: Search Literature

Execute next planned search:

**Query:** TITLE-ABS-KEY("data integration") AND TITLE-ABS-KEY(methodology OR protocol)
**Database:** Scopus
**Results:** 45

Analyzing results...
- 12 methodology papers
- 8 application papers
- 3 review articles

Flagging for deep-dive:
1. Author A 2020 (original method) - DOI:10.xxxx/example.1234
2. Author B 2022 (improvements) - DOI:10.xxxx/example.5678

Log to queries.log and PROGRESS.md.

Step 3: Analyze Findings

For each relevant paper:

Reading: Author A et al. 2020

**Finding:** Method X produces dataset-wide integration scores

**Type:** MAJOR - directly relevant to research question
**Confidence:** HIGH - primary source, methodology paper

→ STOP: Major finding triggers Reviewer 2

Or:

**Finding:** Sample quality affects measurement reliability

**Type:** MINOR - supporting evidence
**Confidence:** MEDIUM - mentioned in discussion

unreviewed_claims_pending: 5/5 → Batch Reviewer 2 triggered

Step 4: Extract Data (if applicable)

If paper has relevant supplementary data:

Downloading supplementary materials for Author A 2020...

- Table S1: Integration results (CSV) ✓
- Table S2: Quality metrics (XLSX) ✓
- Methods S1: Protocol details (PDF) ✓

Reading Table S1 (complete - 2,847 rows)...
Reading Table S2 (complete - 156 rows)...

Extracted: Integration metric data suitable for analysis

Step 5: Validate

Check data availability:

Validation check:
- [ ] Data exists? YES - Table S1 has count data
- [ ] Accessible? YES - Supplementary materials downloaded
- [ ] Format usable? YES - CSV with clear columns

→ CONTINUE (data available)

Or:

Validation check:
- [ ] Data exists? NO - paper describes method but no public data
- [ ] Alternative? Checking GEO... NOT FOUND

→ PATH BLOCKED - Need alternative data source

Step 6: Check Stop Conditions

Stop condition check:

- Goal achieved? NO - still gathering evidence
- Dead end? NO - promising direction
- Serendipity? NO - staying on track
- Blocked? NO - data available

→ CONTINUE TO NEXT CYCLE

Step 7: Update State

Updating STATE.md:
- Cycle: 8
- Phase: discovery (continuing)
- unreviewed_claims_pending: 0 (cleared by batch review)
- Next action: Deep-dive Tsai 2015 methodology

Appending to PROGRESS.md:
### Cycle 7 - 15:45
- **Action:** Scopus search data integration methodology
- **Result:** Found 45 papers, 3 flagged
- **Decision:** Major finding on count data → Reviewer 2
- **Serendipity:** None

Reviewer 2 Triggers

During the loop, these trigger Reviewer 2:

Trigger	Action
Major finding	STOP loop, invoke `/vibe-science:reviewer2`
5 unreviewed claims	STOP loop, invoke `/vibe-science:reviewer2 --batch`

After Reviewer 2 completes, loop resumes.

Loop Termination

The loop ends when:

SUCCESS: All success criteria in RQ.md met
NEGATIVE: Kill condition triggered
PIVOT: Serendipity warrants new RQ

Output appropriate CONCLUSION.md template.