Pipeline

---
description: Runs a recipe or folio through the full publishing pipeline — register, clarity audit, image prompt, format audit, keyword/category pull, glossary pull, hero image check, build, and deploy. Driven entirely by The Manual/recipes.json stage flags. Handles both new recipes (all stages false) and updates (resumes from first false stage). Single entry point after any Claude Desktop Mode 2 session.
---

# Pipeline Workflow

Invoked with: `/pipeline [id]`

Example: `/pipeline 12-23`

---

## Phase 0 — Register

1. Read `The Manual/recipes.json`. Search for an entry with `"id": "[id]"`.
2. **If no entry exists:** run the register-recipe workflow inline to create it, then continue here.
3. Identify all stages currently set to `false`. These are the pending stages.
4. Determine pathway:
   - **New** — no `version` field present (recipe has never been modified by the pipeline)
   - **Update** — `version` field present (content was previously changed by a pipeline run)
5. Initialize a `needs_rebuild` flag to `false`. This flag is set to `true` only when the pipeline writes changes to a folio file or a glossary file. It is never set by a clean audit pass alone.

Print a one-line status before proceeding:
> `[id] — [title] | Pathway: [new/update] | Pending: [list of false stages]`

If all stages are already `true`, report that the recipe is fully complete and stop.

---

## Phase 1 — Clarity Audit → `clarityAudit`

Skip if `clarityAudit: true`.

1. Invoke the **Recipe Clarity Auditor** agent on the folio file.
2. **If clean:** mark `clarityAudit: true`. Do NOT set `needs_rebuild`. Continue.
3. **If issues found:** display each finding with suggested fix.
   - **STOP.** Do not advance until every issue is resolved or explicitly excepted by the user.
   - Once resolved, re-run the audit to confirm clean before marking `clarityAudit: true`.

---

## Phase 2 — Image Prompt

1. Read the folio file. Extract the recipe title, headnote, and key visual elements (primary ingredients, technique, garnish, color, texture).
2. Generate a Gemini hero image brief using this structure:

Recipe: [dish name] Chapter: [chapter name] Description: [headnote — key visual elements, texture, color, sauce, garnish] Cuisine: [cuisine type] Key elements: [3–5 primary visual ingredients or techniques]

1. Output the prompt to the user. No stop point — continue immediately after output.

---

## Phase 3 — Format Audit → `formatAudit`

Skip if `formatAudit: true`.

1. Run: `python3 scripts/audit.py --id [id] --scan-only`
2. **If clean:** mark `formatAudit: true`. Do NOT set `needs_rebuild`. Continue.
3. **If issues found:** display each issue with its fix tag (`[auto-fix]`, `[regex-fix]`, or `[manual]`).
   - Auto-fixable issues: run `python3 scripts/audit.py --id [id]` (interactive mode), show proposed fixes, **prompt to approve** before applying.
   - Manual issues: list them and ask the user to resolve or explicitly flag as an exception. Do not advance past this phase until all issues are resolved or excepted.
4. If any fixes were written to the folio file: set `needs_rebuild = true`.
5. Mark `formatAudit: true`.

---

## Phase 4 — Keyword & Category Pull → `keywordPull`

Skip if `keywordPull: true`.

1. Read the folio file. Check for `## Keywords` and `## Category` sections.
2. **If both present:** mark `keywordPull: true`. Do not set `needs_rebuild`. Continue.
3. **If either missing:** generate via Ollama using `python3 scripts/audit.py --id [id]`, show the proposed content, **prompt to approve** before writing.
4. If content was written to the folio: set `needs_rebuild = true`. Mark `keywordPull: true`.

---

## Phase 5 — Glossary Pull → `glossaryPull`

Skip if `glossaryPull: true`.

1. Extract all bullet entries from the `## Glossary` section of the folio.
2. For each term, grep across all files in `The Manual/Glossary/` for an existing entry with the same name.
3. Present findings in two groups:
   - **New terms** — show proposed definition and target letter file
   - **Existing terms** — show current definition vs proposed; flag if the new version meaningfully adds to the existing one
4. Before writing, strip any recipe-specific context from definitions (e.g., "used here for...", "in this recipe..."). Definitions in the main glossary must be universal.
5. **Prompt to confirm** the full list before writing anything.
6. Write approved entries to the correct letter files in `The Manual/Glossary/`, inserting in alphabetical order within the file.
7. Set `needs_rebuild = true`. Mark `glossaryPull: true`.

---

## Phase 6 — Hero Image → `heroImage` + `heroImageOptimized`

Skip if `heroImage: true` AND `heroImageOptimized: true`.

1. Check `site/public/images/[id].webp`.

   **File found:**
   Run a spec check using `sips`:

   ```bash
   sips -g pixelWidth -g pixelHeight -g fileSize "site/public/images/[id].webp"

• Warn (non-blocking) if dimensions ≠ 1920×1080
• Warn (non-blocking) if file size > 500 KB
• Mark heroImage: true and heroImageOptimized: true.
• If deployed is still false at this point: set needs_rebuild = true. A recipe that has never been deployed must be deployed regardless of whether this pipeline run wrote any content changes.
• Continue.

File not found: STOP. Output:

Hero image missing for [id]. The image prompt was provided at Phase 2. Place [id].webp in site/public/images/ then re-run /pipeline [id]. All completed stages are saved — the pipeline will resume from this point.

Do not proceed to build or deploy.

---

Phase 7 — Pre-Deploy Review

Run only if needs_rebuild = true.

1. Print a full summary of every change made during this pipeline run (folio edits, glossary additions/updates, registry flag changes).
2. Flag any items that warrant judgment before going live:
   • Keyword count outside 8–15 range (or 8–20 for collection folios)
   • Unresolved manual audit issues marked as exceptions
   • Glossary terms