pg-essay-audiobook-scope

Resolve Paul Graham Essays Into Approved Source Texts

Use this stage after the intake checkpoint to lock the two requested Paul Graham essays from Paul Graham's website to full-text local captures and to approve the TTS working set before chunking or synthesis.

Inputs

Read:

• workflow/pg_essay_audiobook/intake_checkpoint.json
• workflow/pg_essay_audiobook/continuation_gate.json

Task targets:

• Do Things that don't scale
• Founder mode
• /root/audiobook.mp3

Write Approved Source Texts and TTS Working Set

Create:

• workflow/pg_essay_audiobook/source_texts/do_things_that_dont_scale.txt
• workflow/pg_essay_audiobook/source_texts/founder_mode.txt
• workflow/pg_essay_audiobook/working_set_record.json
• workflow/pg_essay_audiobook/scope_summary.json

Requirements:

• Approve only full-text essay captures from Paul Graham's website.
• Exclude summary-only, truncated, duplicate, off-site, or aggregator candidates.
• Keep selected_essay_sources and non_selected_candidates explicit in working_set_record.json.
• Preserve provider fallback candidates in this priority order and do not choose the final provider yet: elevenlabs_tts, openai_tts, local_tts.
• If a task-local handoff note or route note is already exposed, keep it available as continuation context and do not move it into non_selected_candidates.

Resolve and Verify the Two Essays

1. Confirm the requested essay titles and output path from the intake artifacts.
2. Fetch candidate pages for both essays directly from Paul Graham's website.
3. Save cleaned full-text captures to the two source_texts files so later stages can chunk local text without re-fetching.
4. Reject incomplete or non-canonical candidates and record them in non_selected_candidates.
5. Count words in each approved text file and record brief content_verification_notes.
6. Write the approved working-set and scope-summary records.

Minimal checks:

• Each saved text file contains continuous essay prose and is clearly longer than a summary.
• selected_titles contains exactly the two requested essay names.
• continuation_status is approved_pending_continuation.
• pending_next_skill is pg-essay-audiobook-binder.

Output Contract

workflow/pg_essay_audiobook/working_set_record.json

{
  "selected_essay_sources": [
    {
      "title": "Do Things that don't scale",
      "url": "string",
      "text_path": "workflow/pg_essay_audiobook/source_texts/do_things_that_dont_scale.txt"
    },
    {
      "title": "Founder mode",
      "url": "string",
      "text_path": "workflow/pg_essay_audiobook/source_texts/founder_mode.txt"
    }
  ],
  "selected_tts_provider_candidates": [
    "elevenlabs_tts",
    "openai_tts",
    "local_tts"
  ],
  "non_selected_candidates": [
    "string"
  ],
  "primary_artifact_path": "/root/audiobook.mp3",
  "continuation_status": "approved_pending_continuation"
}

workflow/pg_essay_audiobook/scope_summary.json

{
  "selected_titles": [
    "Do Things that don't scale",
    "Founder mode"
  ],
  "selected_word_counts": {
    "Do Things that don't scale": 0,
    "Founder mode": 0
  },
  "content_verification_notes": [
    "string"
  ],
  "pending_next_skill": "pg-essay-audiobook-binder"
}

Use integer values in selected_word_counts. Keep the selected local text files as the canonical source texts for the next stage.

Continue to Paul Graham Essay Audiobook Binder

Next command:

• Run pg-essay-audiobook-binder with workflow/pg_essay_audiobook/working_set_record.json and workflow/pg_essay_audiobook/scope_summary.json.

Stop Condition

Stop after the approved source texts, working_set_record.json, and scope_summary.json are complete. Do not chunk text, choose the final TTS provider, call a speech API, or write /root/audiobook.mp3 in this stage.