Asked for 128, diigo returned 100 (the short-page guard caught it on the
first real fill), so PAGE_SIZE = 100: ~168 pages for 16.7k bookmarks.
The same run showed a fill job outliving a page-size change: progress
counted in 24s ("page 25 of 698") resumed as if counted in 128s. Stored
state reinterpreted, so: FillStatus records pageSize (additive; jobs
saved without it counted in 24), and a job whose size differs from
PAGE_SIZE fails with "page size changed …; start the fill again"
instead of resuming; starting again replaces it. Tested with the old
shape in storage.
Generated-by: Claude Opus 5.5
diigo-cache
Cloudflare Worker cache over the webapp factory:
session-cookie reads of diigo bookmarks, cached in Durable-Object SQLite
storage. Evolved from the diigo-dump worker (provenance:
mm-diigo projects/diigo-bak/); the
static API-key path is dead (see the capsite checklist,
goal 2 preamble).
What it holds
- The create flow, on one port:
/_connectrenders webapp's own mint form (diigo-prefilled, cookie blank) — wrangler multi-config dev doesn't serve aux workers' HTTP, so/_mintrelays the form opaquely into the WEBAPP binding (the cookie crosses; nothing on this side opens or logs it). Mint →/_wire(webapp id pre-filled, mode public|private) → the created page, where the user saves the ids. - Two independent page caches (
src/cache.ts), one behavior: a page hit is served from storage with no diigo call; a miss asks the webapp instance in the cache's wiring and etches the page.DiigoCache(bindingCACHE_DO): every item. Addressed by its cache id.PublicCache(bindingPUBLIC_DO): shared items only — private rows are dropped before etching (publicOnly), so its storage holds nothing private. Same cookie webapp as the private cache. Public page n is diigo page n, filtered (may show fewer thanPAGE_SIZE, 128)./public/<username>resolves through theidFromName("index")registry.
- Pages: home's dump form (cache id + page number, POST) shows raw
dump rows;
/public/<username>?page=nshows an HTML list with a page form (JSON forAccept: application/json). Both saycache: HITorMISS; every page read also logs{event: "cache", kind, page, outcome}. - Publish later:
POST /_publish(cache id in the body) wires a public companion to a private-only cache. - The webkey page and Cap'n Web:
/cache/#<cache id>is a skeleton page, the same for every key; its script reads the key from the fragment and trades it in-band for a capability over Cap'n Web at/api(CacheApi.cache(key)→ aCacheCapfacet: fill, status, the public companion;src/api.ts). The worker serves capnweb's browser build itself at/_capnweb.js. - Fill job: walks every diigo page into a cache in the background —
DO alarms, a few pages per step, progress stored in the DO
(
startFill/fillStatus/stopFill), driven from the webkey page; the public page shows progress. - The
DiigoBridgeentrypoint for other workers (its cache-id source is the goal-2 wiring TODO).
State surface
Ids are webkeys (the pattern): 256-bit bearer secrets that never ride a query string or path. The webkey page carries the cache id in the URL fragment (never sent to the server) and passes it in-band over Cap'n Web; forms carry ids in POST bodies; logs and pages show only a fingerprint (a prefix of the id's one-way hash).
- The cache id: the cache's read capability (bearer; not logged). Held by the user — saved to the password manager under a petname they choose (the username field; never sent to the server) — and passed per POST.
- The webapp instance id: the diigo-read capability; lives in each cache DO's wiring, arrives from the mint step.
- The cookie: lives only in webapp's DO storage; crosses the wire once, at mint time, as an opaque relay.
- The public cache: no capability —
/public/<username>is public by design (a guessable URL; hiding the link doesn't hide the page).
Stored state is meant to be durable: see the breaking-change rule in capsite AGENTS.md.
Deployment concern: public pages spend your diigo session
Anyone who can reach /public/<username> can make the cache fetch from
diigo through your cookie session: each request may spend up to
FETCH_BUDGET (40) diigo fetches on uncached pages (goal 1d), paced at
MS_PER_CALL. Cached pages cost nothing. Before exposing a deployment:
- rate-limit the public routes and
/api(Cap'n Web pipelining makes calls cheap to send) — on Cloudflare, the Workers rate-limit binding (keyed by client and/or username); on other workerd hosts, a rate limit in the proxy in front; - run the fill job (goal 1e) so most pages are cached before strangers arrive.
RPC use (from another worker, same account)
// wrangler.jsonc of the caller
"services": [{ "binding": "DIIGO", "service": "diigo-cache", "entrypoint": "DiigoBridge" }]
// cache ids ride goal-2's wiring; see the capsite checklist
const ndjson = await env.DIIGO.dumpNdjson(4); // 4 pages of PAGE_SIZE
Test
pnpm types # regenerate worker-configuration.d.ts (after wrangler.jsonc changes)
pnpm lint # fails if those types are stale; tsc over src/ and tests/
pnpm test # vitest
Tests run the real worker and both DOs in the vitest pool, with
WEBAPP bound to a fake webapp serving
diigo-shaped items (wrangler.tests.jsonc).
Run, deploy
How to run / connect / deploy is maintained in the capsite checklist (single source; this README keeps only worker-level facts).
Status: create, dump, publish, and public pages exercised against live diigo in dev (2026-10-07); not yet deployed.