Survey SOP

<!-- MAINTENANCE NOTES: This is RC's core research execution skill, covering the entire workflow AFTER papers have been found (search-sop handles finding papers).

Boundary with search-sop: search-sop = "find papers", survey-sop = "read, synthesize, and manage the project lifecycle". Boundary with writing-sop: survey-sop = "integrate information", writing-sop = "turn information into polished documents".

Content sources: - AGENTS.md §8 Research Workflow Phase 2-4 (expanded) - AGENTS.md §6 Cross-Module Handoff (detailed) - research-sop: Paper Evaluation, Citation Integrity, Workspace Integration Last synced: 2026-03-23 (AGENTS.md v3.7) -->

Scope

Covers Phase 2-4 of the research workflow (after papers are found). Does NOT cover: literature search (Phase 1, see Search SOP), writing standards or document production (see Writing SOP), data analysis/coding (see Coding SOP).

Phase 2 -- Deep Reading

Systematic reading extracts structured knowledge from each paper.

Reading Workflow

1. Select papers from library with library_search.

Prioritize by: deadline proximity > user request > relevance score > recency.

1. Set read status to "reading" via libraryupdatepaper(id, read_status: "reading").
2. Read systematically -- for each paper, extract:

- Core findings: key results, quantitative outcomes, effect sizes - Methodology: study design, data sources, analytical approach, sample size - Limitations: stated limitations + unstated weaknesses you identify - Key citations: papers referenced that may warrant Phase 1 follow-up

1. Save reading notes to workspace:

workspace_save("outputs/notes/{paper-short-title}-notes.md", content)

1. Update paper metadata: libraryupdatepaper(id, read_status: "read", notes: "...")
2. Flag gaps: if the paper cites important work not in your library, note it

for a targeted Phase 1 search. Do not silently skip missing references.

Reading Depth Levels

• Skim (triage, large batch): title, abstract, conclusion, figures
• Focused (relevant to current question): methods + results + key citations
• Deep (core project paper): full extraction, reproduce key calculations

Comparative Reading

When reading multiple papers on the same topic, build a comparison matrix (method, dataset, results, limitations). Note contradictions and consensus. Save via workspace_save("outputs/notes/comparison-{topic}.md", ...)

Phase 3 -- Analysis and Synthesis

Transform extracted knowledge into coherent research outputs.

Synthesis Workflow

1. Organize findings by theme, not by paper. Group related results across

multiple papers to identify patterns, contradictions, and gaps.

1. Draft the synthesis following the user's preferred structure and citation

style. Default to thematic organization with APA citations.

1. Save drafts to workspace:

workspace_save("outputs/drafts/{topic}-{type}.md", content) Types: literature-review, summary, analysis, report

1. Generate bibliography: libraryexportbibtex(paper_ids: [...]) or

libraryexportbibtex(paperids: [...], format: 'ris'). Save to: workspacesave("outputs/exports/bibliography-{project}.bib", bibtex)

1. Link output to task if one exists (see Phase 4 Task-File Linking).

Synthesis Quality Gates

Before presenting a synthesis to the user:

• Every claim has a citation traced to a paper in the library
• Contradictory findings are explicitly noted, not silently omitted
• Gaps in the literature are identified and flagged
• Quantitative claims use exact numbers from the source papers

Phase 4 -- Task Management

Structure research work into trackable tasks with deadlines.

Task Lifecycle

task_create → task_link (papers/files) → task_note (progress)
  → task_complete (with summary)

Creating Tasks

taskcreate params: title (action statement), tasktype ("human" | "agent" | "mixed"), priority ("urgent" | "high" | "medium" | "low"), deadline (ISO 8601).

Task-File Linking

When a task produces output: (1) workspacesave the file, (2) tasklinkfile(taskid, filepath), (3) taskcomplete(id, notes: "Output: ..."). For papers: tasklink(taskid, paper_id).

Task Progress Tracking

Use tasknote(taskid, note) for progress updates. Set status: "blocked" with blocker note when stuck. On completion, emit a progress_card.

Paper Evaluation Criteria

Evaluate five dimensions: Venue quality (reputable, peer reviewed?), Citation count (citations-per-year, not raw count), Methodology (sound approach, adequate sample, controls?), Reproducibility (sufficient detail?), Relevance (addresses user's question?).

Rate: high / medium / low. Only add high + medium to library.

Citation Integrity

Full citation rules (4 non-negotiable rules) are in the Writing SOP skill. Key rule for survey phase: before citing any paper, call library_search to confirm it exists locally. If not found, add it first (Cross-Module Rule 6).

Cross-Module Coordination

Six handoff rules keep data consistent across modules:

1. monitorreport -> library: Present papercard per finding. User selects

which to add -> libraryaddpaper. Emit monitor_digest. Save digest to outputs/monitor/monitor-scan-{date}.md.

1. monitor -> cron: After monitor_create, suggest cron scheduling.

Each tick: monitorgetcontext -> scan -> monitorreport -> monitornote.

1. libraryaddpaper + active project: Auto-call tasklink(taskid, paper_id).

Reversible, no confirmation needed.

1. taskcomplete -> progresscard: Output progress_card with papers read,

added, tasks completed, highlights.

1. Phase transitions: After Phase 1 -> progress_card + suggest Phase 2.

After Phase 2 -> summarize findings + suggest Phase 3.

1. Cite -> verify: Before citing, library_search to confirm paper exists.

If missing, libraryaddpaper first. Never cite unverified papers.

Workspace Integration

All research outputs MUST be persisted to workspace -- never only in chat.

Output File Naming

• outputs/drafts/literature-review-{topic}.md -- literature reviews
• outputs/notes/{paper-short-title}-notes.md -- reading notes
• outputs/notes/comparison-{topic}.md -- comparison matrices
• outputs/reports/analysis-{dataset}.md -- analysis reports
• outputs/monitor/monitor-scan-{date}.md -- monitor digests
• outputs/reports/weekly-report-{date}.md -- progress reports
• outputs/exports/bibliography-{project}.bib -- BibTeX exports

Session Continuity

When resuming work, check: workspacehistory (recent changes), workspacelist (current files), task_list (active tasks). Combine with MEMORY.md to reconstruct where the user left off.

RC Local Tools Reference

Key tool call chains for survey and project execution:

Deep Reading Chain

library_search -> library_update_paper(read_status:"reading")
  -> [read & extract] -> workspace_save(notes)
  -> library_update_paper(read_status:"read", notes)

Synthesis Chain

library_search(tags/collection) -> [organize by theme]
  -> workspace_save(draft) -> library_export_bibtex(paper_ids)
  -> workspace_save(bibliography)

Task Management Chain

task_create(title, deadline, priority) -> task_link(paper_id)
  -> task_note(progress) -> workspace_save(output)
  -> task_link_file(file_path) -> task_complete(notes)

Monitor Integration Chain

monitor_create -> cron(schedule) -> monitor_get_context
  -> [scan] -> monitor_report -> monitor_note
  -> library_add_paper (user-selected) -> task_link
  -> workspace_save(digest)

Related Research-Plugins Skills

For methodology and domain guidance beyond this SOP, browse these RP subcategories:

• research/methodology/ -- research design, experimental design, grounded theory, mixed methods
• research/deep-research/ -- systematic review (PRISMA), scoping review, meta-synthesis
• research/paper-review/ -- paper summarization, peer review, quality assessment
• research/funding/ -- grant writing (NSF/NIH), data sharing plans
• research/automation/ -- end-to-end research pipelines, automated workflows
• domains/ -- 16 discipline-specific subcategories (biomedical, cs, economics,

law, humanities, etc.). Browse domains/{discipline}/ for domain-specific research methodology, data sources, and evaluation standards