Upstream issues: every workaround, and the issue it waits for
mise run upstream:status shows the state of each issue that has one. When one closes, do what the row says, then mise run dev:test. Seen with EmDash 1.2.0, Astro 7.3.5 or 7.3.6 and wrangler 4.147.0 or 4.148.0, unless a row says otherwise. “Not filed” means nobody has reported it upstream yet.
| Issue | Problem | Workaround | When fixed |
|---|---|---|---|
| emdash#3993 | With Cloudflare Access configured, the CLI cannot sign in to the dev site: /_emdash/api/auth/dev-bypass is not there | The emdash task takes a token from /_emdash/api/setup/dev-bypass?token=1 (scripts/emdash.mjs) | Take that block out |
| emdash#3994 | emdash whoami does not send EMDASH_HEADERS, so it fails behind Cloudflare Access | signin:token -- --live also writes EmDash’s sign-in store; the emdash task answers whoami itself | Take both out |
| emdash#3995 | emdash whoami exits 0 when no site is running | site:start ends on emdash schema list | Use whoami |
| emdash#3996 | The CLI’s sign-in file is written without a lock: two sign-ins at once lose one | signin:token keeps a file per site; the test gives each site a config folder. With --live it writes EmDash’s own file too, with no lock either | Share one config folder in the test |
| EmDash, not filed | No setup and no CLI sign-in without a browser; emdash login always opens one | signin:token writes the administrator and the token into the database. signin:passkey puts a do-nothing open first on the PATH for emdash login (scripts/quiet/); on Windows the browser does open | Use EmDash’s way |
| EmDash, not filed | Under Cloudflare Access a machine cannot sign in by any documented route | signin:token -- --live | Use the documented route |
| EmDash, not filed | A local sign-in is kept per project folder and outlives the database | signin:token does not use emdash login; site:reset removes the saved token with the database | |
| EmDash, not filed | The welcome dialog has no setting | site:start and signin:token close it with the call its own button makes (scripts/core/site.mjs, scripts/core/signin.mjs) | Set the option |
| EmDash, not filed | Its tables are made while it answers its first request: a site just started or deployed answers no such table | On a Cloudflare site signin:token tries up to six times, five seconds apart (scripts/core/signin.mjs) | Take the loop out |
| EmDash, not filed | A Node build takes itself to be on port 4321 | site:preview sets EMDASH_SITE_URL | |
| EmDash, not filed | emdash export-seed and emdash doctor cannot reach a Cloudflare template’s local database | model:sync records the model with emdash types; the seed is not refreshed | Refresh the seed too |
| EmDash, not filed | A seed is applied only on the first request to an empty database, and emdash seed writes to a SQLite file: neither reaches a Cloudflare site’s database once it runs | site:seed applies the file through the site’s HTTP API (scripts/core/seed.mjs) | Use EmDash’s |
| EmDash’s HTTP API and client, not filed | No request makes a block type from a seed file’s versions or names a byline’s avatar, and the client gives a term no language | site:seed says each in a skip line (Seed files) | Apply them |
| EmDash, not filed | A site just made or emptied answers 500 to lists while it sets itself up over its first requests | A read that site:seed and site:demo send themselves is asked again, up to eight times, two seconds apart (scripts/core/api.mjs); one made through EmDash’s client is asked once. A request that changes something is never sent twice | Take the loop out |
| EmDash’s OpenAPI document, not filed | GET /_emdash/api/openapi.json of EmDash 1.2.0 and the site differ in 14 places, each asked of a running site. 12 requests the site answers are not in it: GET and POST /relations, POST /taxonomies, GET and POST /admin/bylines, GET /content/{collection}/{id}/revisions, POST /content/{collection}/{id}/preview-url, GET and POST /admin/api-tokens, GET and PUT /settings/backups, POST /settings/backups/archives. One method is wrong: it has PUT /settings, which the site answers with its 404 page; the site takes POST. One answer is shaped wrongly: GET /schema/collections/{slug} is given as a bare collection, and the site answers { item }, with commentsEnabled and commentsModeration in it | Each is written down as a type in scripts/core/emdash-requests.d.ts (Unlisted, Corrected, Refused), so the requests site:seed and site:demo send are still type-checked | Take each out when the document has it |
EmDash’s client (emdash/client), not filed | Its type declarations (EmDash 1.2.0) differ from what the site answers in two ways the code records: a value the site answers as null is declared as one that may be left out, and a taxonomy’s terms come as a tree, each with its children, which the declarations do not say | The shapes the tasks read are written by hand in scripts/core/emdash-client.mjs | Use the package’s own types |
| EmDash, not filed | emdash migrate cannot use wrangler’s sign-in | No migration step: EmDash migrates on the first request | |
| EmDash, not filed | The skills create-emdash writes cannot be refreshed | emdash:update says they are not updated | Refresh them in emdash:update |
EmDash’s templates and astro check, not filed | Not every template has typescript. Without it astro check asks to install it, checks nothing and ends as a success; and it says it does not work with TypeScript 7 yet | site:check and emdash:update add typescript@6 to the site first, and say so (the state site:typescript in scripts/core/site.mjs) | Take the state out; once TypeScript 7 works, add the newest |
| Astro’s dev server, not filed | Packages changed under a running dev site leave it answering 500 from its next request on | Packages are installed only when the lockfile is not the one last installed from. A running dev site is stopped while its packages change and started again after; one that answers 500 is started again (scripts/core/site.mjs) | |
| pnpm, not filed | In a site with a local plugin (file:./plugins/<name>), pnpm exec installs everything again first, every time: seconds a command, and a lockfile date that changes | The site’s own tools (astro, emdash, wrangler) are run by Node directly (tool in scripts/core/world.mjs). Not everywhere yet: How to help | |
| pnpm on Windows, not filed | pnpm add file:./plugins/<name> writes the folder into package.json with a backslash and into the lockfile with a slash; with a frozen lockfile, as on a CI runner, every pnpm command after it refuses: ERR_PNPM_OUTDATED_LOCKFILE | plugin:new and plugin:add write it with a slash (scripts/plugin-sandbox.mjs) | Take the rewrite out |
| EmDash, not filed | emdash-plugin init fails on Windows beside a running dev server | plugin:new stops the site first | |
| EmDash, not filed | Nothing checks a project before its first deploy | live:check is wrangler’s dry run | |
| emdash discussion 3999 | No command installs, removes or lists a registry plugin | plugin:install sends the admin’s two requests (scripts/plugin-install.mjs) | Use the command |
| EmDash, not filed | create-emdash --sandboxed-plugins does not switch the sandbox on | plugin:sandbox writes both settings | |
| EmDash, not filed | emdash-plugin info <handle> <slug> fails when the publisher’s own host is down | plugin:install asks the registry’s aggregator | |
| EmDash, not filed | On a Node site EmDash stops workerd’s launcher, not workerd: the sandbox process outlives a restart of the sandbox and the site itself, keeps its port, and the next one fails with bind(): Address already in use | plugin:sandbox allows workerd’s install script (allowBuilds in pnpm-workspace.yaml), which puts the program in the launcher’s place; site:stop stops one left over; plugin:install and plugin:remove restart a Node site | Drop all three |
| EmDash, not filed | A Node site’s sandbox takes fixed ports from 18788, so two Node sites on one machine cannot both run sandboxed plugins (seen 2026-10-08, macOS: the second one’s workerd ends with bind(): Address already in use) | None: one Node site with sandboxed plugins at a time | |
| EmDash, not filed | On a Node site one plugin that cannot start takes every sandboxed plugin down | plugin:works reports it; plugin:remove takes it out | |
| EmDash, not filed | Nothing lists what an installed registry plugin declares | plugin:install keeps what the site answered when it verified the release, for plugin:works and plugin:update | Ask the site |
| EmDash’s plugin sandbox on Cloudflare, not filed | It lists a plugin’s keys with id LIKE '<key>%', and D1 refuses a LIKE pattern over 50 bytes (“LIKE or GLOB pattern too complex”). Contact Forms 0.2.0 counts a visitor’s messages by a key of some 118 characters: on a Cloudflare site with EmDash 1.2.0 it takes one message, and answers every later one, to any of its forms, “Please try again later.” | None. plugin:demo sends one message and proves that one (scripts/core/plugin-demo.mjs) | Prove a second message |
| EmDash’s plugin sandbox on Cloudflare, not filed | It gives a plugin 10 subrequests a call (DEFAULT_LIMITS.subrequests in src/sandbox/runner.ts of @emdash-cms/cloudflare). Cloudflare’s Worker Loader holds a plugin to that only when deployed, never on this machine, and EmDash 1.2.0 passes no limits to its runner, so a site has no setting for it. On a deployed site a plugin’s admin page is refused now and then with 400 Too many subrequests by single Worker invocation, a different plugin’s each run. Seen 2026-10-09, in two runs of four: Bulletin 0.1.1’s three admin pages, then Link Guardian 0.1.0’s dashboard, whose two other pages answered 200 | None. plugin:works -- --live says each in a FAIL line and fails. Whether asking again succeeds has not been tried | |
| EmDash on a deployed Cloudflare site, not filed | A plugin’s admin page answers 404 Plugin route not found now and then, for a plugin that is listed as active and whose route answered 200 in the same check. Seen 2026-10-09, in one run of four: Instant Indexer 1.0.0’s two admin pages, which answered 200 the run before | None: the same FAIL line. Whether asking again succeeds has not been tried | |
| EmDash on a deployed Cloudflare site, not filed | Straight after an install, a plugin that GET /_emdash/api/admin/plugins lists as active is not in the admin’s manifest. Seen 2026-10-09, in one run of four: four plugins at once (Instant Indexer 1.0.0, Image Alt Text Manager 0.2.1, Comment Notify 0.1.0, Bulletin 0.1.1) | None: plugin:works -- --live says active, NOT in the admin's manifest and fails. Whether asking again succeeds has not been tried | |
| EmDash, not filed | A page asked for by a visitor gets a locals.emdash with no handlePluginApiRoute (EmDash 1.2.0), which the block of @masonjames/emdash-contact-forms 0.2.0 asks for its form through: a visitor reads “This form is currently unavailable.” | site/src/middleware.ts gives a visitor’s page that one question. A site of your own needs the same file | Delete the file |
| EmDash, not filed | With a prefix on the site’s first language, the admin’s pages answer 404 | This repo’s site/ gives English no prefix, and French /fr (site/astro.config.mjs) | |
| A local Cloudflare site, not explained, not filed | Now and then a GET gets no answer, and gets one when asked again | The plugin tasks ask a read twice before calling it unanswered (scripts/plugin-api.mjs, scripts/core/plugin-demo.mjs). A write is never sent twice | |
| EmDash’s guide, not filed | The backup guide’s SQL dump fails on a site with search: D1’s export refuses the search tables | live:backup takes a Time Travel bookmark and a content package | Add the dump |
| wrangler, not filed | wrangler types --check fails on a freshly scaffolded template | Not used as a check | Add it to site:check |
| Cloudflare’s Vite plugin, not filed | Two sites started in the same moment take the same debugger port (9229), and one answers 500 to everything | Sites start one at a time, machine-wide (scripts/core/site.mjs) | Drop the lock |
| Coywolf Pack 0.35.11, not filed | Its Redirects admin page asks GET /_emdash/api/plugins/coywolf-pack/redirects/removed, which answers 404 while Redirects is off, and the browser logs it as a console error (seen 2026-11-07, EmDash 1.4.0) | None. plugin:works -- coywolf-pack says it in a FAIL line; every other check of the pack passes | Prove it passes |
| Miniflare before November 2026, not filed | With Cloudflare’s route cache in a site (cacheCloudflare(), which plugin:coywolf adds when a site has none), every write to the built site on this machine answers 500: cache.purge is not a function (seen with wrangler 4.147.0, miniflare 5.20261001.0-alpha; gone with wrangler 4.158.0 and miniflare 5.20261104.0-beta, 2026-11-07) | Update wrangler and the Cloudflare adapter | |
| Cloudflare Worker Previews (beta) | A preview’s database, bucket and session store are not made for it, and a preview has no address until preview addresses are on | live:preview makes them and switches the addresses on (scripts/live-preview.mjs) | Let wrangler do it |