Scholar Citation Verify

Four-layer citation verification workflow for academic papers. Scans paper files, verifies every citation through WebSearch/Google Scholar/APIs, and produces a verification report with actionable fixes.

Pre-load (before execution)

1. Codebase docs: If .workflow/codebase/ARCHITECTURE.md exists, read for project context
2. Specs: maestro load --type spec --category coding — load coding conventions
3. Wiki knowledge: maestro search "academic writing research paper" --json — top 5 entries as prior context
4. All optional — proceed without if unavailable

Architecture Overview

User: "Verify citations in my paper"
         |
         v
┌──────────────────────────────────────────────────────────────┐
│  SKILL.md (Orchestrator)                                      │
│  Collect preferences → Dispatch phases → Track progress       │
└──────────┬───────────────────────────────────────────────────┘
           |
   ┌───────┼───────────┬──────────────┐
   v       v           v              v
┌──────┐┌──────────┐┌──────────────┐
│ P1   ││ P2       ││ P3           │
│ Scan ││ Verify   ││ Report       │
│ &    ││ 4-Layer  ││ & Fix        │
│Extract││          ││              │
└──┬───┘└────┬─────┘└──────┬───────┘
   │         │             │
   v         v             v
 citation   verified     report.md
 entries    results      + fixed .bib

Key Design Principles

1. Never trust AI-generated citations: AI citations have ~40% error rate. Every single citation must be verified via external search.
2. Verify during writing, not after: Proactive verification integrated into the writing process.
3. Four-layer verification: Format → Existence → Information Matching → Content Validation.
4. Clear failure marking: Unverifiable citations marked as [CITATION NEEDED], never silently skipped.

Interactive Preference Collection

Before dispatching to any phase, collect these preferences:

Questions to ask the user:

1. Paper Location
   "Where are the paper files? (directory path)"
   → paperDir

2. Verification Mode
   Options: Full (all citations) | Incremental (new/changed only) | Spot Check (random sample)
   → verificationMode

3. Strictness Level
   Options: Strict (for submission) | Normal (for draft) | Lenient (quick check)
   → strictness

4. Auto-fix
   Options: Yes (auto-fix format issues) | No (report only)
   → autoFix

Store responses as verifyPreferences context for all phases.

Auto Mode Defaults

When workflowPreferences.autoYes === true:

• Detect paper directory from cwd
• Full verification mode, Normal strictness
• Auto-fix enabled
• Skip confirmation prompts

Execution Flow

COMPACT DIRECTIVE: Context compression MUST check TodoWrite phase status. The phase currently marked in_progress is the active execution phase — preserve its FULL content. Only compress phases marked completed or pending.

TodoWrite Setup

Citation Verification:
- [ ] Phase 1: Scan & Extract — find .tex/.bib files, extract all citations
- [ ] Phase 2: 4-Layer Verify — format, existence, matching, content checks
- [ ] Phase 3: Report & Fix — generate report, apply fixes

Phase Sequence

Phase 1: Scan & Extract
   └─ Ref: phases/01-scan-extract.md
      ├─ Input: paperDir, verificationMode
      └─ Output: citationEntries (list of all citations with metadata)

Phase 2: 4-Layer Verify
   └─ Ref: phases/02-verify.md
      ├─ Input: citationEntries, strictness
      └─ Output: verificationResults (per-citation status + details)

Phase 3: Report & Fix
   └─ Ref: phases/03-report-fix.md
      ├─ Input: verificationResults, autoFix
      └─ Output: report.md + optionally fixed .bib file

Phase Reference Documents (read on-demand when phase executes):

Phase	Document	Purpose	Compact
1	phases/01-scan-extract.md	Scan files, extract citations	TodoWrite driven
2	phases/02-verify.md	4-layer verification	TodoWrite driven + sentinel
3	phases/03-report-fix.md	Generate report, apply fixes	TodoWrite driven

Compact Rules:

1. TodoWrite in_progress → preserve full content, do not compress
2. TodoWrite completed → may compress to summary
3. sentinel fallback → phases marked with sentinel contain compact sentinel; if only sentinel remains, must immediately Read() to recover

Core Rules

1. Every citation must be checked: No citation passes without at least existence verification.
2. Use real APIs: Verify via WebSearch, Google Scholar, Semantic Scholar, CrossRef, arXiv — never trust memory.
3. Mark failures clearly: Use [CITATION NEEDED] for unverifiable references.
4. Preserve user's formatting choices: When auto-fixing, only fix errors, don't restyle.
5. Report honestly: Show verification confidence levels, don't hide partial matches.

Data Flow

Phase 1 ──citationEntries──→ Phase 2
Phase 2 ──verificationResults──→ Phase 3

Data persistence: Results written to paperDir/.verify/
  paperDir/.verify/
  ├── citations-extracted.json    (Phase 1 output)
  ├── verification-results.json   (Phase 2 output)
  ├── verification-report.md      (Phase 3 output)
  └── references-fixed.bib        (Phase 3 output, if autoFix)

Error Handling

Error	Action
No .tex/.bib files found	Ask user for correct directory
API rate limit hit	Wait and retry with exponential backoff
Citation not found in any API	Mark as [CITATION NEEDED], continue
BibTeX parse error	Report unparseable entry, skip and continue
Network unavailable	Fall back to format-only verification

Related Skills

• scholar-writing: Phase 4 (Citation Management) uses these verification principles
• scholar-review: References verification as part of paper quality check