1# Auto-Continuation Protocol
2 
3## When to Use
4 
5Trigger auto-continuation when report exceeds 18,000 words in single run.
6 
7---
8 
9## Strategy Overview
10 
111. Generate sections 1-10 (stay under 18K words)
122. Save continuation state file with context preservation
133. Spawn continuation agent via Task tool
144. Continuation agent: Reads state -> Generates next batch -> Spawns next if needed
155. Chain continues recursively until complete
16 
17---
18 
19## Continuation State File
20 
21**Location:** `~/.claude/research_output/continuation_state_[report_id].json`
22 
23```json
24{
25  "version": "3.0.0",
26  "report_id": "[unique_id]",
27  "file_path": "[absolute_path_to_report.md]",
28  "mode": "[quick|standard|deep|ultradeep]",
29 
30  "progress": {
31    "sections_completed": ["list of section IDs"],
32    "total_planned_sections": 15,
33    "word_count_so_far": 12000,
34    "continuation_count": 1
35  },
36 
37  "artifacts": {
38    "sources_path": "[folder]/sources.jsonl",
39    "evidence_path": "[folder]/evidence.jsonl",
40    "claims_path": "[folder]/claims.jsonl",
41    "run_manifest_path": "[folder]/run_manifest.json"
42  },
43 
44  "research_context": {
45    "research_question": "[original question]",
46    "key_themes": ["theme1", "theme2"],
47    "main_findings_summary": [
48      "Finding 1: [100-word summary]",
49      "Finding 2: [100-word summary]"
50    ],
51    "narrative_arc": "middle"
52  },
53 
54  "quality_metrics": {
55    "avg_words_per_finding": 1500,
56    "citation_density": 5.2,
57    "prose_vs_bullets_ratio": "85% prose",
58    "writing_style": "technical-precise-data-driven"
59  },
60 
61  "next_sections": [
62    {"id": 11, "type": "finding", "title": "Finding X", "target_words": 1500},
63    {"id": 12, "type": "synthesis", "title": "Synthesis", "target_words": 1000}
64  ]
65}
66```
67 
68---
69 
70## Spawning Continuation Agent
71 
72Use Task tool:
73 
74```
75Task(
76  subagent_type="general-purpose",
77  description="Continue deep-research report generation",
78  prompt="""
79CONTINUATION TASK: Continue existing deep-research report.
80 
81CRITICAL INSTRUCTIONS:
821. Read continuation state: ~/.claude/research_output/continuation_state_[report_id].json
832. Read existing report: [file_path from state]
843. Read LAST 3 completed sections for flow/style
854. Load research context: themes, narrative arc, writing style
865. Load source registry from state.artifacts.sources_path — use stable source_ids, assign display numbers via citation_manager.py
876. Maintain quality metrics (avg words, citation density, prose ratio)
88 
89YOUR TASK:
90Generate next batch (stay under 18,000 words):
91[List next_sections from state]
92 
93Use Write/Edit to append to: [file_path]
94 
95QUALITY GATES:
96- Words per section: Within +/-20% of avg_words_per_finding
97- Citation density: Match +/-0.5 per 1K words
98- Prose ratio: Maintain >=80%
99- Theme alignment: Section ties to key_themes
100 
101After generating:
102- If more sections remain: Update state, spawn next agent
103- If final sections: Generate bibliography, verify report, cleanup state
104"""
105)
106```
107 
108---
109 
110## Continuation Agent Quality Protocol
111 
112### Context Loading (CRITICAL)
113 
1141. Read continuation_state.json -> Load ALL context
1152. Read existing report file -> Review last 3 sections
1163. Extract patterns:
117   - Sentence structure complexity
118   - Technical terminology used
119   - Citation placement patterns
120   - Paragraph transition style
121 
122### Pre-Generation Checklist
123 
124- [ ] Loaded research context (themes, question, narrative arc)
125- [ ] Reviewed previous sections for flow
126- [ ] Loaded source registry from artifacts (stable source_ids, not citation numbers)
127- [ ] Loaded quality targets (words, density, style)
128- [ ] Understand narrative position (beginning/middle/end)
129 
130### Per-Section Generation
131 
1321. Generate section content
1332. Quality checks:
134   - Word count within +/-20%
135   - Citation density matches
136   - Prose ratio >=80%
137   - Theme connection verified
138   - Style consistent
1393. If ANY fails: Regenerate
1404. If passes: Write to file, update state
141 
142### Handoff Decision
143 
144Calculate: Current words + remaining sections x avg_words_per_section
145- If total < 18K: Generate all + finish
146- If total > 18K: Generate partial, update state, spawn next agent
147 
148### Final Agent Responsibilities
149 
150- Generate final content sections
151- Generate COMPLETE bibliography from state.citations.bibliography_entries
152- Read entire assembled report
153- Run validation: `python scripts/validate_report.py --report [path]`
154- Delete continuation_state.json (cleanup)
155- Report complete to user
156 
157---
158 
159## User Communication
160 
161After spawning continuation:
162```
163Report Generation: Part 1 Complete (N sections, X words)
164Auto-continuing via spawned agent...
165   Next batch: [section list]
166   Progress: [X%] complete
167```
168