feat(publish): strip pipeline scaffolding from body before WP publish

.gitignore

@@ -32,3 +32,4 @@ tmp/
 # Playwright / browser caches
 playwright-report/
 test-results/
+.worktrees/

TODOS.md

@@ -22,30 +22,65 @@ Generated by /plan-ceo-review on 2026-04-04. Updated by reviews.
 
 ---
 
-### CLI-1: --auto-approve flag silently ignored
+### CLI-1: --auto-approve flag silently ignored — RESOLVED
 
-**What:** `brewpress draft --auto-approve` is accepted by the parser and stored in `args.auto_approve` but never passed to `Orchestrator.draft()`. Users expecting scripted end-to-end runs will be confused when the command stops at the review step.
+**Status:** ✅ Fixed by feat(stack-8). `auto_approve` is now passed to `Orchestrator.draft()` (cli.py:275). When critic verdict = "pass", the job is auto-approved to APPROVED_STEP_1. Fixed by /qa on feat/ga-release, 2026-04-08.
 
-**Why:** Broken UX for CI/scripted pipelines that depend on this flag.
-**Effort:** S (CC: ~20 min)
-**Priority:** P1
-**Depends on:** Decision on what auto-approve should do (skip review step? auto-approve-content?)
+---
+
+### CLI-2: reject() blocked at APPROVED_STEP_2 — RESOLVED
+
+**Status:** ✅ Fixed. `brewpress reject --force` flag added (forces rejection at any state including APPROVED_STEP_2).
+**README:** Add one-liner: "`brewpress reject --force` — discard a job stuck at any approval state."
 
 ---
 
-### CLI-2: reject() blocked at APPROVED_STEP_2 with no escape hatch
+### CLI-3: Score transparency after `brewpress draft` — RESOLVED
 
-**What:** A job stuck at `APPROVED_STEP_2` (e.g., WP publish failed, user wants to discard) cannot be rejected — the state machine blocks rejection at this state. The only escape is manual deletion of `~/.brewpress/last_draft.json`.
+**Status:** ✅ Fixed by feat(stack-8). `DraftResult.pipeline_summary` printed to stdout when non-empty (cli.py:283-284). Format: `[pipeline] N rounds | seo: X, clarity: X, tech_accuracy: X, readiness: X`. Fixed by /qa on feat/ga-release, 2026-04-08.
 
-**Why:** Users who hit publish errors have no CLI-level way to reset and start fresh.
-**Fix:** Either allow `brewpress reject --force` at APPROVED_STEP_2, or add a `brewpress reset` command that clears state.
-**Effort:** S (CC: ~15 min)
-**Priority:** P1
+---
+
+### PIPE-1: BREWPRESS_PIPELINE_MODE feature flag — INVALIDATED
+
+**Status:** ❌ Invalidated. DraftAgent (v1) was deleted in chore(stack-8) — no v1 code path exists to fall back to. The 4-agent pipeline IS the only pipeline. This TODO is no longer actionable. Closed by /qa on feat/ga-release, 2026-04-08.
 
 ---
 
 ## P2 — Medium Priority
 
+### GAP-5: Code-based MarkdownFixer for StructurerAgent
+
+**What:** Replace StructurerAgent's LLM call with deterministic markdown transformations:
+- Heading level normalization (H3 before H2 → demote H3 to sub-H2)
+- H1 dedup (multiple H1s → demote 2nd+ to H2)
+- CTA injection at end of body if `BlogJob.cta` is set but absent from markdown
+
+**Why:** Structure issues are rule-based, not creative. Eliminates 0–1 LLM calls per run.
+The LLM call is the only one in the pipeline with zero creative upside.
+**Effort:** M (human: ~3 hr / CC: ~30 min)
+**Priority:** P2
+**Depends on:** Nothing (self-contained)
+
+---
+
+### GAP-6: Code-based SEO patcher (pre-LLM pass in SEOAgent)
+
+**What:** Before calling `think()` in `SEOAgent.optimize()`, run a deterministic patcher:
+- Inject `primary_keyword` into the first paragraph if missing
+- Trim title to ≤60 chars or expand to ≥50 chars via slug-based suffix
+- Trim/expand meta_description to 120–160 chars
+
+After patching, re-run `seo.full`. If score ≥ 85 now, skip the LLM call entirely.
+
+**Why:** Eliminates 0–1 LLM calls per run for posts that are "almost there" SEO-wise.
+Most SEO issues are mechanical (length, keyword presence) and don't need generation.
+**Effort:** M (human: ~3 hr / CC: ~30 min)
+**Priority:** P2
+**Depends on:** Nothing (SEOAgent already calls seo.full)
+
+---
+
 ### WP-1: 'Already wrote about this?' duplicate detection
 
 **What:** Before generating, query WP REST API for posts with matching slug or
@@ -76,6 +111,19 @@ fetches diff from GitHub API directly.
 
 ---
 
+### SEO-1: Consolidate boost_eval.py into tools/seo.py
+
+**What:** Once SEOAgent uses `seo.full` from `tools/seo.py` directly, `boost_eval.py` and its `run_checks()` function become redundant. Post-GA cleanup: merge the deterministic check logic into `tools/seo.py` and update CriticAgent to use `seo.full` tool via the tools registry.
+
+**Why:** DRY violation — two modules doing the same SEO analysis. Maintenance burden as SEO rules evolve.
+**Pros:** Single source of truth for SEO checks. Simplifies CriticAgent's fast path.
+**Cons:** Breaking change to CriticAgent imports; requires updating test_critic_agent.py.
+**Effort:** S (human: ~2 hr / CC: ~20 min)
+**Priority:** P3
+**Depends on:** SEOAgent shipped and proven in production
+
+---
+
 ## P3 — Low Priority / Vision
 
 ### WATCH-1: CLI watch mode