integers (1,2,3) = planned structure
decimals (2.1,2.2) = urgent insertions between integers
Dir format: 02.1-description/, Plan format: 02.1-01-PLAN.md
Validation: X exists+complete, X+1 exists, X.Y doesn't exist, Y >= 1
READ: ~/.claude/write-the-f-paper/templates/section-prompt.md
READ: ~/.claude/write-the-f-paper/references/plan-format.md
READ: ~/.claude/write-the-f-paper/references/verification-layers.md
READ: ~/.claude/write-the-f-paper/references/orchestrator-pattern.md
READ: ~/.claude/write-the-f-paper/references/context-fidelity.md
READ: ~/.claude/write-the-f-paper/references/agent-model-matrix.md
READ: .planning/ROADMAP.md
READ: .planning/PROJECT.md
LOAD: .planning/structure/argument-map.md, outline.md, narrative-arc.md
Create executable section prompt (PLAN.md) via thin orchestrator pattern. Orchestrator validates, loads context, resolves model profile, spawns section-planner agent, then runs plan-checker for quality verification. PLAN.md IS the prompt Claude executes - not a doc that gets transformed.
ORCHESTRATOR: /wtfp:plan-section (this command)
AGENTS: section-planner (creates PLAN.md), plan-checker (validates PLAN.md)
PATTERN: Orchestrator loads context → spawns planner → spawns checker → revise loop if needed
WHY_SUBAGENTS: Fresh context windows for peak quality. Planner gets full project context without orchestrator overhead. Checker validates independently.
Argument-first: Every paragraph serves core argument. Plan prose that advances thesis, not filler.
Evidence-grounded: Plan where citations go before writing. Claims need sources identified.
Reader-aware: Plan for target audience. Technical depth, assumed knowledge, explanation level.
[step:validate_and_resolve_model p=1]
RUN: ls .planning/ 2>/dev/null
RUN: MODEL_PROFILE=$(cat .planning/config.json 2>/dev/null | grep -o '"model_profile"[[:space:]]*:[[:space:]]*"[^"]*"' | grep -o '"[^"]*"$' | tr -d '"' || echo "balanced")
MODEL_LOOKUP{agent,quality,balanced,budget}:
section-planner | opus | opus | sonnet
plan-checker | sonnet | sonnet | haiku
RESOLVE: planner_model, checker_model from MODEL_PROFILE
[/step]
[step:load_project_state]
READ: .planning/STATE.md
PARSE: current_position, accumulated_decisions, open_questions, argument_strength
IF missing+.planning_exists → OFFER reconstruct|continue
[/step]
[step:load_structure_context]
RUN: ls .planning/structure/*.md 2>/dev/null
STRUCTURE_LOAD{section_type → files}:
introduction → argument-map.md, narrative-arc.md
methods → argument-map.md (claims needing methodological support)
results → argument-map.md (claims supported here)
discussion → argument-map.md, narrative-arc.md
related_work → argument-map.md (positioning), literature.md
abstract → all (summary of everything)
TRACK: extracted context for PLAN.md
[/step]
[step:identify_section]
RUN: cat .planning/ROADMAP.md
RUN: ls .planning/sections/
IF multiple_available → ASK which to plan
IF obvious → proceed with first incomplete
REGEX: ^(\d+)(?:\.(\d+))?$ → Group1: integer, Group2: decimal
IF decimal → validate: X complete, X+1 exists, X.Y doesn't exist, Y >= 1
READ: existing PLAN.md or RESEARCH.md in section dir
[/step]
[step:load_context_md]
CRITICAL: Load CONTEXT.md early, pass to ALL agents.
RUN: SECTION_DIR=$(ls -d .planning/sections/${SECTION}-* 2>/dev/null | head -1)
RUN: CONTEXT_CONTENT=$(cat "${SECTION_DIR}"/*-CONTEXT.md 2>/dev/null)
IF CONTEXT.md exists → store for injection into planner prompt as
HONOR: locked decisions, boundaries, vision
EXCLUDE: deferred ideas
ALLOW: discretion areas
IF missing → SUGGEST: /wtfp:discuss-section (simpler) or proceed without
[/step]
[step:mandatory_literature_check]
MANDATORY for sections flagged Research: Likely
LIT_LEVEL{level,desc,action}:
0-skip | internal content only, no external evidence | skip
1-quick | single source verify | quick check, no RESEARCH.md
2-standard | multiple sources, positioning | /wtfp:research-gap → RESEARCH.md
3-deep | core lit review, theoretical framework | /wtfp:research-gap depth=deep → full RESEARCH.md
IF roadmap=Research:Likely → Level 0 not available
[/step]
[step:read_prior_sections]
RUN: for f in .planning/sections/*/*-SUMMARY.md; do sed -n '1,/^---$/p' "$f" | head -30; done
PARSE: section, subsection, key-claims, evidence-used, decisions
AUTO-SELECT sections matching:
- Immediate prior section
- Same argumentative thread
- Mentioned in STATE.md as affecting current
EXTRACT: claims established, evidence cited, patterns, decisions
[/step]
[step:spawn_section_planner]
EMIT:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WTF-P ► PLANNING SECTION {X}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
BUILD planner_prompt with inlined content:
- STATE, ROADMAP, PROJECT, argument-map, outline, RESEARCH, prior summaries
- CONTEXT_CONTENT (from CONTEXT.md)
-
GOOD: Specific paragraphs, clear actions, verifiable output
- "Draft hook paragraph establishing research gap in autonomous systems"
- "Write methods subsection on data collection protocol, ~150 words"
- "Synthesize Smith2023 and Jones2024 findings on intervention efficacy"
BAD: Vague, not actionable
- "Write introduction" / "Add citations" / "Make it better"
TEST: If you can't specify Paragraphs + Action + Mode + Verify + Done → too vague
- No hour estimates
- No "perfect prose" expectations (first drafts are drafts)
- No style committee approvals
- No sub-sub-sub tasks
Tasks = instructions for Claude writing, not editorial board requirements.
- [ ] Model profile resolved, agent models determined
- [ ] STATE.md read, project context absorbed
- [ ] CONTEXT.md loaded early and passed to ALL agents
- [ ] Mandatory literature check completed (Level 0-3)
- [ ] Prior sections, sources, structure synthesized
- [ ] Section-planner agent spawned with full context
- [ ] PLAN file(s) exist with XML structure
- [ ] Plan-checker agent validated plan (or skipped per config)
- [ ] Plan-check-revise loop completed if issues found
- [ ] Each plan: 2-3 tasks (~focused scope)
- [ ] Each task: Type, Mode, Paragraphs (if applicable), Action, Verify, Done
- [ ] PLAN file(s) committed to git
- [ ] User knows next steps