rzen
64bad7091b
One written standard for every audience-facing word: concise, to the point, inspiring delight through specificity — with word budgets per copy type (reviews had drifted from a ~370-word median to 500-1000+). All writing skills now reference the charter, and daily-pipeline gains a stage-4 copy-edit pass: a subagent that didn't write the day's copy tightens it against the charter, prose only, diff reviewed before commit. Digest gains a copy-edit line. Claude-Session: https://claude.ai/code/session_01WZaczDJjL3xZ3u5spsN5AL
28 lines
3.2 KiB
Markdown
28 lines
3.2 KiB
Markdown
# unique.rzen.dev
|
||
|
||
A daily product-discovery site (Astro, static). The editorial pipeline is skill-driven and designed to run unattended.
|
||
|
||
## Pipeline
|
||
|
||
- `/daily-pipeline` — the whole daily chain: capture → bundle → reviews (several per run, judgment call) → pick of the day (explicit decision every run) → build → commit → push → digest. This is the entrypoint for scheduled/headless runs; scheduled runs auto-deploy afterwards via the lab CI `publish-unique-daily` job.
|
||
- `/daily-finds` — capture-only sweep of `sources.json`; writes `src/content/find/<yyyy>/<mm>/<slug>/`, marks the day's shortlist (`shortlist: true`, up to 12), adds `amazonLink` for confidently-matched physical products, and applies mechanical source maintenance.
|
||
- `/build-bundle [auto|weekly|<theme>|<slugs>]` — one themed roundup referencing find slugs. `weekly` = date-based catch of unbundled shortlist finds.
|
||
- `/cluster-bundles [auto]` — batch-cluster the backlog into 4–7 bundles.
|
||
- `/build-review <slug> [auto]` — graduate a find to a full review (fact-check against the product's own site is mandatory in auto mode).
|
||
|
||
Unattended runs never pause for confirmation — they publish, log to `daily-finds.log.md`, and report a digest; the user vetoes after the fact. Only two things stay user-only: vetting `candidate-sources.md` nominees into `sources.json`, and scheduling.
|
||
|
||
## Conventions that bite
|
||
|
||
- **`EDITORIAL.md` is the voice charter** — concise, to the point, inspiring delight (through specificity, never ad copy); word budgets per copy type. All audience-facing prose follows it; the pipeline runs a fresh-eyes copy-edit stage over each day's new copy.
|
||
- **No tobacco or alcohol content** — hard editorial exclusion at every stage: never captured as finds, never selected into new bundles, never graduated to reviews. A few legacy alcohol finds (May–July 2026) stay published and keep their existing bundle spots, but they're ineligible for anything new.
|
||
- Review heroes are auto-discovered from `src/content/reviews/<slug>/hero.<ext>` (`src/lib/hero.ts` glob) — name the file `hero.*`; no registration.
|
||
- The homepage "pick of the day" is `src/data/pick.json` (`{"hero": "<review-slug>"}`); falls back to the latest review.
|
||
- Find slugs are globally unique although folders are date-bucketed; URLs stay flat at `/find/<slug>/`.
|
||
- Finds are reachable through bundles and through tag pages (a secondary "Finds" list under the entry grid on `/tags/<tag>/`) — an unbundled find is still invisible on the home feed, grids, and category pages.
|
||
- `source` frontmatter must exactly match a `name` in `sources.json`.
|
||
- `topics` frontmatter (finds and reviews) must come from the controlled vocabulary in `src/lib/topics.ts` (`TOPIC_META` keys) — the schema enum fails the build on anything else. Tags are free-form; topics are curated. "Giftable" is a tag (`giftable`), never a topic.
|
||
- Amazon links: store canonical `https://www.amazon.com/dp/<ASIN>` with no query params; `src/lib/affiliate.ts` adds the tracking ID, badge, and disclosure at render time.
|
||
- Commit style: see `git log` — `content: …` for capture runs, `site: …` for everything user-visible; one commit per pipeline stage.
|
||
- `npm run build` (astro check + build) must pass before pushing.
|