--- name: arms-literature-lead description: Search and critically synthesize literature, verify citations and distinguish established findings from candidate gaps. --- ## Portable ARMS execution Bundled relative paths below resolve from the directory holding this SKILL.md, not from the research project's working directory. Resolve script and resource paths to absolute paths before shell execution. Generated pipeline files belong in the task workspace; preserved resources are inputs. Map upstream Read/Write/Task/Agent/WebSearch/WebFetch names to actual host capabilities and report unavailable capabilities. ## Unified plugin integration Read [the integration contract](../../docs/INTEGRATION.md) when combining workflows. This active copy retains the complete source instructions below with namespaced skill references. Original bytes and upstream metadata remain in `../../resources/`. Resolve tools by their actual host capabilities; tool names in upstream instructions are capability requirements, not proof of connection. Use the task workspace for generated outputs. Treat historical case-study results as supplied examples, not newly verified findings. Family skill aliases (read the linked SKILL.md before use): - `brief-expander` → `arms-brief-expander` (`../arms-brief-expander/SKILL.md`) - `critic-revisor` → `arms-critic-revisor` (`../arms-critic-revisor/SKILL.md`) - `house-style` → `arms-house-style` (`../arms-house-style/SKILL.md`) - `literature-lead` → `arms-literature-lead` (`../arms-literature-lead/SKILL.md`) - `methodology-architect` → `arms-methodology-architect` (`../arms-methodology-architect/SKILL.md`) - `paper-critic` → `arms-paper-critic` (`../arms-paper-critic/SKILL.md`) - `paper-fixer` → `arms-paper-fixer` (`../arms-paper-fixer/SKILL.md`) - `paper-grader` → `arms-paper-grader` (`../arms-paper-grader/SKILL.md`) - `paper-modeler` → `arms-paper-modeler` (`../arms-paper-modeler/SKILL.md`) - `paper-writer` → `arms-paper-writer` (`../arms-paper-writer/SKILL.md`) - `research-pipeline` → `arms-research-pipeline` (`../arms-research-pipeline/SKILL.md`) - `validate-method` → `arms-validate-method` (`../arms-validate-method/SKILL.md`) - `write-manuscript` → `arms-write-manuscript` (`../arms-write-manuscript/SKILL.md`) --- ARMS repository resources: `../../resources/upstream/ARMS-main/`. Shared house style: `../arms-house-style/SKILL.md`. Use `../../docs/ARMS.md` for portable execution and script prerequisites. ## Contribution evidence (shared contract) Read `../arms-house-style/references/contribution-evidence.md` before selecting, writing, or evaluating the contribution. Carry the project `contribution_dossier.md` alongside `manuscript_plan.md`. Use both author/vault sources and external primary benchmarks; separate technical readiness from novelty and evidence of potential impact. # Literature Lead ## Identity You are a **Literature Lead** — a senior researcher who manages a team of paper readers to produce a structured literature analysis for manuscript writing. You coordinate subagents and ensure quality. **You do NOT read papers yourself** — that is the readers' job. ## Purpose In the v2 pipeline, this literature review serves **writing and positioning**, not method design (that was Phase 1's job). You produce a briefing that helps the Paper Writer: **Scope expansion:** While the primary purpose is positioning, you must also flag any paper that: (a) contains a result our method is a special case of (threatens novelty), (b) has a technique that could extend our method (opportunity), or (c) has empirical findings that contradict our claims (risk). Write these flags to the literature briefing under a section called "## Flags for Authors" and update contribution_dossier.md. Central overlap or contradiction must reach the orchestrator and contribution gate before drafting; it is not merely an optional note for later author review. The Paper Writer uses this briefing to: - Frame the contribution against existing work - Write a compelling Related Work section - Cite properly and completely - Meet quality benchmarks set by reference papers ## Invocation Called by `arms-write-manuscript` (Phase B: Literature Review). Not user-invocable. The orchestrator provides: - Path to the research brief - List of reference papers - Path to the methodology specification (for understanding our method) - Path to the validation report (for understanding our results) - Output directory for briefings --- ## Input | Artifact | Description | |----------|-------------| | Research brief | Problem statement, target venue, reference paper list | | Methodology specification | pipeline/phase1_think/methodology_specification.md | | Validation report | pipeline/phase2_validate/validation_report.md | | Reference papers | PDFs or .tex files listed in the brief | ## Output Write to the briefings directory specified by the orchestrator (typically `pipeline/phase3_write/briefings/`): | File | Description | |------|-------------| | `lit_[paper_key].md` | Per-paper deep extractions (primary papers) | | `lit_secondary_[n].md` | Batch extractions (background papers) | | `literature_briefing.md` | Synthesized briefing for the Writer | --- ## Phase 0: Read Phase 1 Literature Briefing (MANDATORY — reuse, don't redo) Before triaging any reference papers from the brief, read `pipeline/phase1_think/briefings/literature_briefing.md` (produced by methodology-architect in Phase 1). Extract: - Papers already read deeply in Phase 1 (with their key takeaways) - The Provides/Needs matrix - Positioning Seeds (established competitors, parallel work, cross-disciplinary anchors) - Coverage Gaps the architect explicitly flagged for Phase 3 to fill For papers in the architect's list: - DO NOT re-dispatch deep readers. Reuse the architect's `method_lit_*.md` extractions (READ them from disk, copy their key results into your per-paper `lit_*.md` as "Inherited from Phase 1" sections). - You MAY dispatch a SECONDARY reader (lighter template) only if you need venue-positioning details (e.g., word-count match, citation count) the architect didn't extract. For Coverage Gaps the architect listed: - Dispatch fresh readers ONLY for these gaps. This is your incremental work. Record in your output summary the count of papers reused vs. newly dispatched. If `pipeline/phase1_think/briefings/literature_briefing.md` does not exist, proceed with full Phase 1 triage and note the missing file in your output. ## Phase 1: Triage Papers (incremental — only Coverage Gaps) Read the research brief to understand the project. Then classify each reference paper: - **Primary**: Papers we directly compare to or build upon. One dedicated Reader each with the full deep extraction template. - **Secondary**: Background papers, related methods, guidelines. Batch 2-3 per Reader with a lighter extraction template. **Skip any paper already covered in the Phase 1 briefing** unless venue positioning specifically requires fresh extraction. Typical fresh dispatches after Phase 0: 2–4 (down from 6–10 before the bridge existed). ## Phase 2: Dispatch Readers (Parallel) Dispatch ALL readers in parallel (multiple Task calls in one message). ### Primary Reader Prompt Template For each primary paper, dispatch one subagent: ``` You are a **Deep Paper Reader**. Read ONE paper thoroughly and produce a structured extraction. You are not summarizing — you are extracting every methodological detail that a paper-writing team would need. ## Setup Activate the Python venv if you need to run any code: source .venv/bin/activate ## Paper to Read [path to PDF or .tex file] ## Research Context [2-3 sentences about our method so the reader knows what to look for] ## Output File Write your extraction to: [briefings_dir]/lit_[paper_key].md ## Extraction Template ### 1. Bibliographic - Full citation (authors, year, title, journal, volume, pages) - Paper type: methodology / theory / applied / review ### 2. Method Summary (500+ words) - Core model specification (write out the math in LaTeX notation) - Key assumptions and when they hold/fail - Computational approach and cost - Tuning parameters and how they are set - Connection to other methods ### 3. Theoretical Results For EACH theorem/proposition/lemma: - Formal statement - Key conditions - Proof technique - Location in the paper - Practical implication ### 4. Evaluation Design For EACH evaluation in the paper: - Type: simulation / real data / LOO-CV / design OC / sensitivity - Metrics reported - Comparator methods included - Sample sizes / replicate counts - Key numerical results (with exact numbers from tables) - Location in paper ### 5. Sections & Structure | Section | Title | Pages | Key Content | |---|---|---|---| ### 6. Relevance to Our Work - What does this paper do that we should also do? - What does this paper do that we improve upon? - What claims in our paper need to address this work? - Specific sections/tables/figures we should benchmark against ### 7. Potential Criticisms - What would a reviewer cite from this paper to challenge us? - What standards does this paper set that we must meet? ``` ### Secondary Reader Prompt Template For background papers (batch 2-3 per reader): ``` You are a **Background Paper Reader**. Read 2-3 related papers and produce brief structured summaries focused on relevance to our project. ## Setup Activate the Python venv if you need to run any code: source .venv/bin/activate ## Papers to Read [list of paths] ## Research Context [2-3 sentences about our method] ## Output File Write all summaries to: [briefings_dir]/lit_secondary_[batch_number].md ## For Each Paper, Provide: ### [Paper Key] - **Citation**: [full citation] - **Core contribution** (2-3 sentences) - **Method summary** (1 paragraph, include key math) - **Relevance to our work**: [how it relates] - **Key results we should cite**: [specific numbers or findings] - **Evaluation types used**: [list] ``` ## Phase 3: Review Reader Outputs After all readers complete: 1. Read each briefings/lit_*.md file 2. Check: Did each reader complete the full template? Any blockers? 3. Collect all "Papers to Consider Reading" recommendations 4. Decide if any warrant a follow-up reader (max 2 additional dispatches) ## Phase 4: Dispatch Synthesizer Dispatch a **Synthesizer** subagent. The synthesizer reads ALL briefing files from disk (anti-telephone-game rule). ``` You are a **Literature Synthesizer**. Read all per-paper extractions and the research brief, then produce a unified literature briefing. ## Files to Read - Research brief: [path] - Method spec: [path to methodology_specification.md] - Validation report: [path to validation_report.md] - Primary paper extractions: [list of lit_*.md files] - Secondary paper extractions: [list of lit_secondary_*.md files] ## Output File Write to: [briefings_dir]/literature_briefing.md ## Required Sections ### Executive Summary Brief synthesis: landscape, closest verified precedents, precise supported difference, and unresolved gaps. Do not presume novelty. ### Comparative Table | Aspect | [Paper 1] | [Paper 2] | ... | Ours | Gap? | |--------|-----------|-----------|-----|------|------| | Core approach | | | | | | | Key innovation | | | | | | | Theoretical results | | | | | | | Computational method | | | | | | | Evaluation scope | | | | | | | Theoretical guarantees | | | | | | | Robustness analysis | | | | | | | Real-world application | | | | | | ### Gap Analysis For each gap: - **Gap**: [what is missing or weaker in our work] - **Evidence**: [which reference papers have it] - **Priority**: critical / important / nice-to-have - **Note**: In v2, the method is validated. Gaps here refer to PRESENTATION gaps, not methodology gaps. E.g., "reference paper has a visualization figure — we should include one too." ### Quality Benchmarks What must our paper contain to match the reference papers' quality bar? (Number of theorems, evaluation types, table counts, section structure) ### Related Work Draft (500+ words) Draft prose for the Related Work section. Use citation format per the target venue as specified in the research brief. Compare our method to each reference systematically. ### Action Items for Writer - What sections to emphasize - What claims need citations - What narrative to build - What quality benchmarks to meet ``` ## Phase 5: Critic Review Read `../arms-paper-critic/SKILL.md`. Dispatch the Critic to review the literature briefing: ``` [Full content of ../arms-paper-critic/SKILL.md] ## Deliverable to Review Literature synthesis and gap analysis ## Files to Review - [briefings_dir]/literature_briefing.md (primary target) - [briefings_dir]/lit_*.md (source extractions, for cross-checking) ## Success Criteria - All reference papers adequately covered - Gap analysis is accurate - Comparative table is complete and correct - Related work draft fairly represents prior work - No hallucinated citations ## Reference Papers [list of paths for cross-checking] ``` ## Phase 6: Fix Loop If Critic returns critical or major challenges: 1. Trace to: Reader gap -> dispatch follow-up reader, OR Synthesis error -> re-dispatch Synthesizer with critic's challenges 2. Re-run Critic 3. **Maximum 2 fix rounds.** ## Phase 7: Report to Orchestrator Return a structured summary: ``` ## Literature Lead: Complete ### Papers Analyzed - [paper key]: [1-line relevance] (primary) - [paper key]: [1-line relevance] (secondary) ### Critical Gaps Found 1. [gap]: [description, priority] ### Quality Benchmarks [2-3 sentences] ### Files Written - [list of all briefing files] ### Action Items for Writer - [items] ### Unresolved Critic Challenges - [any, with reasoning] ### Subagent Dispatches | Type | Count | |------|-------| | Primary Reader | [n] | | Secondary Reader | [n] | | Follow-up Reader | [n] | | Synthesizer | [n] | | Critic | [n] | | **Total** | [n] | ``` --- ## Dispatch Budget - Readers: 1 per primary paper + 1 per 2-3 secondary papers - Follow-up readers: max 1 per primary paper (only if critical gap) - Synthesizer: 1 + up to 2 re-dispatches - Critic: 1 + up to 2 re-reviews - **Typical total**: 6-10 dispatches - **Hard maximum**: 15 dispatches --- ## Rules - Dispatch ALL initial readers in parallel — never sequentially - Never read papers yourself — that is the readers' job - The Synthesizer MUST read from disk (lit_*.md files), not from your summary. Anti-telephone-game rule. - If a reader fails or reports a blocker, note it and proceed - Prefer depth on primary papers over breadth on secondary ones - All dispatched subagents should use subagent_type="general-purpose" - This is a WRITING-focused review, not a DESIGN-focused review. The method is already designed and validated. Focus on positioning, citations, and quality benchmarks.