feat(cms): fill a site's editorial brand context from its own blocks

[aka-sacci-ccr]

What

Adds an Autonomous content collection to the Content tab (Blog group), holding the brand context every generated blogpost will be written against: tone of voice, generation dos and don'ts, audience, values, blog categories, competitors.

Persists to the site's own .deco/blocks/blog-manager-brand.json as plain JSON — no new table. Same path and field names Spire writes, so a site it already set up opens here populated.

BLOG_BRAND_EXTRACT fills the empty fields by reading the site's own CMS blocks.

No scraper, no new credential

The first draft called Firecrawl. That was wrong: FIRECRAWL_API_KEY is an instance-wide env var, so someone on the hosted Studio can neither enable it nor bring their own key, and the cost lands on a shared quota. Reading a URL server-side would also have needed an SSRF guard we don't have (httpUrlSchema only rejects non-http schemes).

The blocks are already in the browser's decofile, so the tool takes them as input and only calls the model. Inference runs on the org's own smart tier via resolveTier — same path as suggest-commit-message and judge-requires-review — so it bills to the org, and the org picks the model.

Two things the real data forced

Tested against a real storefront's 1969 blocks:

• Serialized block size is a bad proxy for prose. One institutional page was 15KB of JSON carrying two sentences — the rest asset URLs, signed video links and device matchers. Ranking 1018 pages by size surfaced product-listing stubs and buried the institutional pages that hold the brand's values. extractBlockProse walks a block and keeps only prop: phrase lines (a phrase = contains a space), deduped, which separates alt: "92% de funcionárias" from site/sections/Layout/Flex.tsx and 20px without a prop allowlist.
• That site has 1018 pages and zero posts, so "read the existing posts" cannot be the only path. Tiers degrade: posts → categories → pages, and the prompt says so instead of treating a blogless site as the exception.

The prompt also encodes traps found in that data: internal asset annotations ([LP … ] [carrossel …]) are not copy; shipping/return/payment fine print is not editorial voice; brand-specific vocabulary (a site calling the shopping bag mochila) is the most valuable thing to capture verbatim; and casing is part of the voice, including when it's inconsistent.

Won't clobber hand-written rules

Re-running the extract writes only into fields that are still empty, so it can never wipe dos/don'ts someone typed. The toast reports how many it filled, or that there was nothing to fill.

Testing

62 unit tests in blog-data.test.ts cover the prose walker (drops URLs/identifiers/dimensions, keeps phrases and prop names, dedupes exact repeats, preserves casing variants) and the evidence ranking (tier order, prose-density ordering, char budget, empty site, missing page keys).

bun test apps/api/src/tools/ reports 25 failures — pre-existing: the same 25 fail on a clean tree (verified via git stash -u), and each file passes in isolation, so it's cross-file test pollution unrelated to this change.

tsc --noEmit clean in both workspaces, knip clean, lint 0 errors, fmt applied.

Not in this PR

• The extract has not been exercised end-to-end against a live model yet — org-fs mounts break the daemon's write route off-cluster, so local verification needs DISABLE_ORGFS_MOUNTS=1. Unrelated to this diff.
• selfhost/examples/dev-hybrid/.env.example still ships STUDIO_SANDBOX_PREVIEW_URL_PATTERN set, which 502s every daemon call off-cluster. Out of scope here; worth its own fix.
• No sub-navigation inside the new tab — one section today, so a tab bar with one item would be noise. Planner, ideas and drafts land under the same heading later.

🤖 Generated with Claude Code

---

Summary by cubic

Infers a site’s editorial brand context from its own CMS blocks and adds an “Autonomous content” collection to edit it, so generated posts match the site’s tone and guardrails. Previously there was no editor and rule fields were string arrays; now rules are objects with a name and markdown body.

• Adds Autonomous content to Content with Planner/Ideas (placeholders) and a Library editor (Context/Formats) that saves to .deco/blocks/blog-manager-brand.json with a sticky rail (Basics, Generation rules, Guardrails, Extra). dos, avoid, values, and competitors now use { name, value }; the markdown editor disables attachments for these text-only fields.

• The Library’s Fill action calls BLOG_BRAND_EXTRACT to read the site’s blocks and fill only empty fields; the UI shows reading/inferring/searching phases.

• BLOG_BRAND_EXTRACT (category “Blog”) runs on the org’s smart tier; input capped to 60 blocks and 12k chars per block. Evidence sampler prioritizes posts > categories > pages, with a 60k total and 4k per block. Returns sources and searchedCompetitors.

• Competitors are enriched via the org’s web_search tier only when none are named in blocks; failures or missing tier return [].

• Tests cover rule normalization, prose extraction, evidence ranking, budgets, and empty/missing cases.

• Rollout

  • No DB/env changes. Backward compatible: legacy string[] rules load and save in the new shape; no migration required. Extraction is idempotent on non-empty fields.
  • Billing: brand inference bills the org’s smart tier; competitor enrichment uses web_search when configured.

^{Written for commit 22104fe. Summary will update on new commits.}