CustomerNewsletter: Agent for Claude Code

Non-Negotiable Execution Rules

1. Do not compress phases. Phase 1A -> 1B -> 1C -> 2 -> 3 -> 4 -> 4.5 -> 4.6 must be materially executed.
2. Phase 1.5 (curator-notes) is conditional mandatory: if notes exist (workspace/curator_notes_*.md or workspace/<Month>.md), execute it before Phase 3.
3. Canonical artifact paths are mandatory. Do not create or rely on fresh_phase* shortcuts.
4. Delegation is allowed only as controlled phase delegation to named agents. Never delegate to generic or "general-purpose" subagents.
5. Use one delegation unit per phase. Delegate only with explicit start/stop boundaries and canonical artifact + receipt requirements.
6. Preferred delegation map:
   • customer_newsletter: Phase 0, 1A, 1B, 2, 3, 4, 4.5, 4.6
   • editorial-analyst: Phase 1C, 1.5
   • skill-builder: skill-authoring tasks only (not newsletter generation phases)
7. Do not delete workspace/newsletter_phase_receipts_<END>.json during an active run.
8. If the user asks for from-scratch generation, run:
   • bash tools/prepare_newsletter_cycle.sh <START> <END> --no-reuse
9. Before reporting completion, run strict validation:
   • bash tools/validate_pipeline_strict.sh <START> <END> --production-artifacts
   • From-scratch: bash tools/validate_pipeline_strict.sh <START> <END> --require-fresh --production-artifacts
   • Benchmark consistency: bash tools/validate_pipeline_strict.sh <START> <END> --require-fresh --production-artifacts --benchmark-mode feb2026_consistency
10. If strict validation fails, continue fixing until it passes or explicitly report the blocker.
11. Record phase receipts immediately after each artifact write:

• bash tools/record_phase_receipt.sh <START> <END> <phase_id> <artifact_path>

1. Do not defer receipts until the end of the run.
2. Use canonical phase_id values only:

• phase0_scope_contract
• phase1a_manifest
• phase1b_github
• phase1b_vscode
• phase1b_visualstudio
• phase1b_jetbrains
• phase1b_xcode
• phase1c_discoveries
• phase1_5_curator_processed
• phase1_5_curator_signals
• phase2_event_sources
• phase2_events
• phase3_working_set
• phase3_curated
• phase4_output
• phase4_5_polishing
• phase4_6_video
• phase4_scope_results
• phase4_editorial_review

1. Curator-note discovery must include both:

• workspace/curator_notes_*.md
• workspace/[A-Za-z]*.md (for example workspace/Jan.md) Exclude generated files: curator_notes_processed_*, curator_notes_editorial_signals_*, and newsletter_*.

1. For the February 2026 benchmark window (2025-12-05 to 2026-02-13), benchmark mode is mandatory before completion:
   • bash tools/validate_pipeline_strict.sh <START> <END> --require-fresh --production-artifacts --benchmark-mode feb2026_consistency
2. If benchmark mode fails, keep iterating until it passes.
3. In Phase 2, generate workspace/newsletter_phase2_event_sources_<END>.json before workspace/newsletter_phase2_events_<END>.md, and record both receipts (phase2_event_sources, phase2_events).
4. For Phase 0, run python3 tools/generate_scope_contract.py <START> <END> before any broad manual scope discovery. If the helper succeeds, treat its output as the Phase 0 artifact.
5. For orchestrated Phase 3 runs, generate workspace/newsletter_phase3_working_set_<END>.md before any broad rereads of discoveries or reference docs, then record phase3_working_set before any Phase 3 curation edits.
6. If workspace/newsletter_phase3_curated_sections_<END>.md is absent during Phase 3, initialize it with python3 tools/init_phase3_curated_sections.py <START> <END> and edit that scaffold in place.
7. When the Phase 3 working set exists and its Missing Data Gate does not flag missing inputs, use it as the primary Phase 3 source. Do not reread raw discoveries, curator notes, LEARNINGS.md, or long reference docs unless the working set explicitly says data is missing.
8. During Phase 3, replace all TODO markers and HTML comment placeholders in the curated scaffold before recording phase3_curated.