Litreview — Academic Literature Orientation

Portability: Requires a Consensus MCP connection, Node.js with docx package for document generation, and (in CLI) bash_tool. Works in Claude Code CLI natively. In Claude.ai with Consensus MCP + Code Execution, the workflow is supported.

Produce a launching pad — not a finished literature review, but an orientation document that gives a researcher entering an unfamiliar field everything they need to start reading and searching with confidence. Think: what a generous colleague who knows the field would tell you over coffee.

Agent Integrity Rules (Research-Pack Convention)

Inherited from the research-pack convention; locked verbatim per PR #657's cross-skill consistency audit.

• Source discipline. Only cite Consensus-returned papers from THIS session. Training knowledge labeled [Not from Consensus — model knowledge] and excluded from cited count. Sparse results stated explicitly, never silently filled.
• Counting discipline. Three numbers tracked: searches executed / unique papers received (deduplicated) / papers cited. Every cited paper has a retrievable Consensus URL from this session. Use scripts/citation_tracker.py for deterministic counts.
• Tool constraints. Consensus per-query cap depends on plan tier. Detect at first search, report at checkpoint. Rate limit is 1 query/sec — sequential execution mandatory.
• Retry policy. On failure → wait 3s → retry once → log. After 3 consecutive failures: stop, alert user, share what was collected.
• Plan-tier detection. Parse first-search response for "Showing top 10" / "upgrade" → free tier (10/search). 20 returned → Pro (20/search). Calculate theoretical ceiling and surface at checkpoint so user can recalibrate.

See references/search_budget_allocation.md for the sequential-execution rationale + plan-tier signals.

Error Handling

Phase 0: Grill-Me Intake (3 forcing questions, one at a time)

Each question carries explicit "why I'm asking". Stop condition: max 3 before Phase 1.

Q1 (root) — Research question specificity

State the research question in 1–2 sentences. Specific is better — "How do LLMs perform on clinical reasoning tasks compared to physicians?" beats "AI in medicine". Vague questions produce vague reviews.

Why I'm asking: The reconnaissance search hinges on precise terminology. Vague questions produce thin recon results that don't yield a useful framework breakdown.

Refuse mush. Re-ask once with examples if user is too broad. If still vague, deliver with explicit "broad-scope orientation, not depth review" caveat.

Q2 (depends on Q1) — Framework hint

Framework — pick one or say "you pick":

1. PICO (Population / Intervention / Comparison / Outcome — most clinical questions)
2. SPIDER (Sample / Phenomenon / Design / Evaluation / Research-type — social/qualitative)
3. Decomposition (Problem / Solution / Evaluation / Limitations — technology-focused)
4. Hybrid (you pick which components from which framework)
5. You pick — analyze Q1 and recommend

Why I'm asking: PICO is the default for ~70% of clinical questions but maps poorly to qualitative work or technology evaluation. Picking upfront saves the recon search from suggesting a misaligned framework.

Forcing choice with default ("you pick"). The skill surfaces its own framework recommendation after the recon search so user can override. Use scripts/framework_recommender.py for the heuristic.

See references/framework_selection.md for PICO / SPIDER / Decomposition canon.

Q3 (depends on Q1) — Tentative depth

Tentative depth — pick one. Final confirmation comes after the framework breakdown:

1. Quick scan (5 searches)
2. Standard review (10 searches)
3. Deep dive (20 searches)

Why I'm asking: I ask this twice — once now to calibrate the recon search emphasis, once after the framework breakdown to confirm. Tentative answer affects which sub-areas to surface first; final answer drives search budget allocation.

Forcing choice. Re-asked at the post-Phase-2 checkpoint after the user has seen the framework breakdown.

Stop condition: 3 questions max before Phase 1. The post-Phase-2 checkpoint is its own grill-me moment (framework table + sub-area-adjustment + depth-reconfirmation).

Phase 1: Initial Reconnaissance

One broad Consensus search to map themes, terminology, methodological distinctions.

• Query: broad version of Q1 (terminology variants are okay; first search casts wide)
• Record: citation_tracker.py --action record_search --session NAME --query "..."
• Record received count: citation_tracker.py --action record_papers_received --session NAME --count N
• Detect plan tier from response: "Showing top 10" / "upgrade" → free; 20 returned → Pro

Synthesize for the checkpoint:

• Themes that surfaced
• Terminology variations (e.g., "LLM" vs "large language model" vs "GPT-style model")
• Methodological distinctions (clinical trials vs benchmark eval vs case study)
• Coverage gaps (sub-questions absent from recon results)

Phase 2: Framework Selection + Sub-area Generation

Choose framework (from Q2 OR override based on recon):

• PICO — most clinical questions (~70% default)
• SPIDER — social / qualitative
• Decomposition — technology focus (Problem / Solution / Evaluation / Limitations)
• Hybrid — explicit cross-framework mapping

Generate 4-5 sub-area questions mapped to framework components. Each becomes a targeted Phase 3 search.

Checkpoint (grill-me forcing-options moment)

After Phase 2, halt and present:

3-4 sentence recon summary

• What themes surfaced
• Terminology landscape
• Evidence landscape characterization

Framework breakdown table

Depth re-confirmation (forcing choice)

Surface the practical constraint: detected plan tier + theoretical ceiling.

• Quick scan (5 searches × ~10 results each = ~50 papers max)
• Standard review (10 searches × ~10 = ~100 papers)
• Deep dive (20 searches × ~10 = ~200 papers)

Sub-area forcing options

• "Looks good — proceed with these sub-areas"
• "Adjust: add sub-area on [X]"
• "Adjust: remove and replace [Y] with [Z]"
• "Restart with different framework"

Why I'm asking (the rationale)

A wrong framework or sub-area set wastes the search budget. This is the last cheap moment to correct course.

Wait for user response before Phase 3. Refuse to start Phase 3 without explicit user choice.

Phase 3: Targeted Searches

Sequential (1 query/sec), budget per depth tier. See references/search_budget_allocation.md for full canon.

Quick scan (5 searches)

• 5 sub-area searches (one per sub-area)
• Skip era-gated + review-specific

Standard review (10 searches)

• 5 sub-area searches
• 2 review article searches (top 2 sub-areas): "systematic review [topic]" / `"meta-analysis [topic]"