@koda-sl/baker-cli 0.164.0 → 0.166.0

package/README.md

@@ -296,7 +296,7 @@ Read it before interpreting any null. It does not fire on the `--out` path (rows
 | Preset                 | Description                              | Default date range |
 |------------------------|------------------------------------------|--------------------|
 | `campaign-performance` | Campaign metrics overview                | LAST_30_DAYS       |
-| `keyword-analysis`     | Keyword performance with match type      | LAST_30_DAYS       |
+| `keyword-analysis`     | Keyword performance per ad group         | LAST_30_DAYS       |
 | `positive-keywords`    | Positive (targeting) keywords only       | ALL_TIME           |
 | `negative-keywords`    | Negative (blocking) keywords only        | ALL_TIME           |
 | `negative-keyword-lists`   | Shared negative lists + their terms  | ALL_TIME           |
@@ -582,6 +582,8 @@ Command groups: `budgets`, `campaigns`, `ad-groups`, `keywords` (add/update/remo
 
 **Batch keyword adds** — `keywords add`, `negative-keywords add`, and `keyword-lists add` take a whole batch in one command: comma-separate `--text` entries and/or pass `--file <list.txt>` (one keyword per line). A `:EXACT`/`:PHRASE`/`:BROAD` suffix per entry overrides the `--match-type` default. Batches stage all-or-nothing as one request (limit 500); each keyword still lands as its own draft op, so it stays individually removable/amendable.
 
+**Pausing or removing a keyword returns a per-ad-group hint** — `keywords update --status PAUSED|REMOVED` and `keywords remove` come back with hints telling you to base the decision on `keyword_view` rows carrying `ad_group.id` + `ad_group_criterion.criterion_id`, never on campaign totals. A keyword's criterion id is derived from its text + match type, so every copy of the same keyword across ad groups shares one id and any aggregate blends the copy holding the history with the copies that never served. The `keyword-analysis` and `positive-keywords` presets select both ids so one query identifies each copy. A bid or URL edit gets no hint — only a pause/removal can strand history.
+
 ---
 
 ### Caching
@@ -4812,7 +4814,7 @@ baker landing critique spring-offer summer-offer claude  # score three landings
 
 - **Takes several slugs at once.** `data.landings` is always an array with one entry per slug (`{ slug, overall, counts, dimensions, findings }`). For a single slug those fields are also spread at the top of `data`, so existing one-landing readers are unaffected. A wrong slug anywhere in the batch fails the whole call before any landing is scored.
 - **Advisory by design — findings never fail it.** A finding is guidance, not a gate; the command errors (exit 2) only on a wrong slug (`NOT_FOUND` / `INVALID_SLUG`, with `error.fix.availableSlugs`).
-- **BRAND.md is the allowlist.** A font, hue, or ground that the client's `src/brand/BRAND.md` (mirrored into `src/styles/global.css` `@theme`, hex or oklch) commits is a decision, not a tell — matched values downgrade to advisory `brandDrift` so the critic never fights a real brand.
+- **The brand is the allowlist.** A font, hue, or ground the client's brand commits is a decision, not a tell — matched values downgrade to advisory `brandDrift` so the critic never fights a real brand. The allowlist reads `src/brand/tokens.json` (the canonical machine-readable brand token file) when present; workspaces that predate it fall back to scanning the `@theme` block in `src/styles/global.css` (hex or oklch) plus `src/brand/BRAND.md` prose.
 - **Records a freshness snapshot** to `.cache/landing-critique/<slug>.json` (the source hash it scored). The publish quality gate reads it to enforce that every changed landing was critiqued and ships no block-tier tell. Re-run after edits so the snapshot stays fresh.
 
 - **AI writing tells are scored in the `copy` family, in English and Spanish.** Beyond buzzwords and em-dash density, the critic detects the constructions models over-produce: the negation pivot ("it's not just X, it's Y" / "no se trata solo de X, sino de Y"), fake-candid openers ("Honestly?", "La verdad es que"), participles hung off a comma (", ensuring…", ", permitiendo…"), copula avoidance ("serves as", "cuenta con"), vague attributions, signposting, filler, hollow closers, false ranges, staccato runs, Title Case headings, and chatbot framing that leaked into the page. Most need a **cluster** (≥2 on the page) before they fire — the source guidance is explicit that isolated tells prove nothing — with two deliberate exceptions: chatbot framing fires at one, and curly quotes are not detected at all, since every editor inserts them automatically. Nothing in this family is block-tier, so it never fails the publish gate. Pattern set adapted from Wikipedia's "Signs of AI writing" (WikiProject AI Cleanup).