Skip to content

Releases

v2.0.0

Choose a tag to compare

@joeblew999 joeblew999 released this 09 Oct 03:39

Changes since v1.2.1

  • the release's own check could not pass: check runs the test, the test rewrote each group's last-run.txt, and the docs check that follows found the generated tests page stale (seen in mise run release -- v2.0.0 -dry-run). Only a run of every step, the level all, is recorded now; a smoke or fast run changes no file git knows, so check changes none. The pages that say what is recorded say so
  • the seed tasks refuse --live, plugin:install asks a publisher's server again, the test's own faults that GitHub and a deployed site showed, and the task descriptions that had gone stale. site:seed and site:demo did not stand on site:local-only: asked with --live they signed this machine in to the deployed site and then worked on the local one — the state is the site's own now and all three stand on it. plugin:install: the site verifies a plugin at its publisher's own server, which once did not answer (400 RECORD_VERIFICATION_FAILED, The direct PDS request failed) and failed two steps of the level all: that refusal alone is asked again, twice, and said. The step for plan read mise's echo of the command line as a state: mise cuts that line to the terminal's width here and prints it whole on GitHub, where the level all failed on that step alone on Linux and macOS (run 37877972102) — it reads the lines that are a state's name now. The descriptions: plugin:new and plugin:add no longer say to start the site again (the task does), live:ship's sentence is whole, site:check says it adds typescript, dev:test says the kept site stays running, dev:hooks does not say the test turns the hook on; TEMPLATE here is the template site/ was made from; the sign-in note names the file that exists. The docs: the three counts of the deployed-site runs, a third thing a deployed site does after an install (active, not in the admin's manifest), and what Windows did in today's run. On this Mac, this code: the level all from nothing, 71 steps, no failure (site 3 min 17 s, signin 4 min 0 s, plugin 4 min 42 s); the level fast 52 s; the live group on a deployed Worker, 43 steps, 8 min 14 s
  • the docs say what the code does at this commit, and where it stops short. A task is a state to reach: plan, site:status, the line each state prints, sites left running (Getting started). New guide, Seed files: site:seed, what it changes in a site, what it does not apply (the skip lines, and three things it drops without one) and what a seed file cannot say; site:demo. plugin:demo's seven and what its description leaves out; signin:mcp for an agent (in Sign in); the test's levels, its kept site, what writes last-run.txt and what GitHub has run; how it is built, and that design's limits (the older scripts have no unit test, only site:seed and site:demo send typed requests, pnpm exec is still in three places). Gaps, each next to the claim it limits: on GitHub only the smoke level has run, and the last run of every step on Windows failed (#6); site:seed, site:demo and plugin:demo are this machine's built site only, and only plugin:demo refuses --live; LIVE_PREVIEW is read only with a workers.dev LIVE_URL, and never by model:sync; plugin:works --live on every plugin does not hold on a deployed site (admin pages refused now and then, 2026-10-09); a running dev site does not load a plugin installed through the built one; French on site/ is the home page and the posts; no step makes a site from cloudflare:blog. upstream.md: rows for the sandbox's ten subrequests and the 404 on a deployed site, D1's LIKE limit under Contact Forms 0.2.0, the 14 places EmDash's OpenAPI document differs from the site and the two in its client's declarations, astro check with no typescript, pnpm exec with a local plugin, a visitor's page with no handlePluginApiRoute, a first language with a prefix; the registry-command row names discussion 3999. Favourite plugins: the five, the two more plugin:demo uses (LinguaDash at 0.2.1), and what a deployed site did. Settings: ADMIN_NAME, SITE_DATABASE and four variables read from the environment. Gone: dev:test live, "running a test turns on the commit check", plugin:favourites without --yes, the rule about steps, and commands no test step runs where one that a step does run says the same
  • the count in the step's reason: three runs out of four
  • a site whose server ended after it had started is not waited for, dev:ci runs the branch you are on, and what a deployed site does to the plugin check. The kept test site, made again from a changed site/, had its built server say running and then be gone: the first sign-in step waited three minutes on a closed port and the fast level failed in 3 min 39 s (the log that said why was emptied by the next start; the Mac's swap was full). Starting a dev or built site is one piece of code now: when nothing answers and the process Astro noted is gone, it is said, the site's own log is shown, and it is started once more; a site that is there and answers nothing still fails, and now with its log (world.alive; 2 unit tests, 90 in all). dev:ci had GitHub run main whatever branch you were on: it passes the branch. The plugin steps were run on a deployed site for the first time, four times: install, favourites, update and remove pass there; plugin:works on every plugin failed three times out of four, a different way each time (not in the admin's manifest, an admin page refused with Too many subrequests, Plugin route not found) — EmDash's on a deployed site, said in the step, which is this machine's only now. live:logs asks five times at addresses of its own: once in four runs the log did not show the one request in time. On this Mac: the level all from nothing, 71 steps, no failure, 5 min 45 s; the level fast, 28 steps, 111 s. On GitHub: smoke on Linux 60 s, macOS 100 s, Windows 208 s
  • a task leaves both sites as it found them: the built site too is started again when a task that rebuilt it left it stopped
  • site:check really type-checks: a site with no typescript had astro check ask to install it, check nothing and end well — it is added (version 6, which astro check works with), said, with a running dev site stopped and started around it. The demo site: 22 files, 0 errors
  • a read is asked again, a few times, while a site that was only just emptied is still setting itself up (it answered 500 to lists, and site:demo run straight after site:reset failed two items). The full cycle on the demo site — emptied, site:demo, plugin:demo — 79 s, no failure; both again: nothing made
  • installing packages stops a running dev site first and starts it again after: changed under it, it answers 500 from its next request on
  • plugin:demo: what the plugins do is on the public pages of the site, and each line proves what a visitor is given — and it is seven plugins now. Contact Forms' form is on /pages/contact and Forms' on /pages/feedback (the page is made, or the form's block added under what a page contact already says; published, in the primary menu, nothing twice); the French copy LinguaDash makes of the post is given its French words and published at /fr/posts/chaque-lien-est-une-promesse, which the English page names (hreflang) and LinguaDash's own switcher leads to; Instant Indexer's key address answers a visitor, and its line quotes the plugin saying it sends nothing from localhost. The post reads as a post ("Every link is a promise", with a photograph asked for at Unsplash) and is the first card of the home page; one an earlier run made is brought up to date. A recipe asks the site's manifest first: where the site's config cannot draw a form's block, or has no French, the line says what the site lacks and nothing is made that no visitor could reach — so the task still passes on a site from EmDash's template. site/: the two packages that draw the forms and LinguaDash's for its switcher, i18n (en, and fr under /fr: src/pages/fr), the home, post and search pages ask for content in the language of their address, and src/middleware.ts, because EmDash 1.2.0 gives a visitor's page no handlePluginApiRoute and Contact Forms' block told every visitor "This form is currently unavailable." World.download: a file as bytes. 24 unit tests on a made-up EmDash (12 before), and the step at the level all asks the pages too. On a copy of site/ from nothing, after site:start: 54 s the first time (the task 41.6 s, installing seven plugins), 0.2 s the second. Seen and not changed, written where the recipe is: on a Cloudflare site Contact Forms 0.2.0 takes one message and then answers every other "Please try again later." — the sandbox lists a plugin's keys with LIKE, and D1 refuses the plugin's 118-character pattern once there is a key to try it on; the same on the site as it was before this
  • site:seed and site:demo speak to EmDash with what EmDash gives, not with requests written by hand. EmDash's own client (emdash/client, loaded from the site's packages by scripts/core/emdash-client.mjs, so it is the client of the EmDash that site runs) for what it has a call for: a collection and its fields, taxonomies, terms, entries (get, list, update, publish, unpublish, schedule), uploads, a menu to read, search. Its requests go through World: the last interceptor it is given answers each by asking World, so a test still sees every request and none reaches a network. What the client has no call for, or cannot say (a menu item, a widget, a section, a byline, a redirect, a comment, a setting, a token, a backup; an entry made with its terms and bylines, a collection with its addresses, a field that is searched) is a request typed from EmDash's own list of its requests: mise run dev:api-types writes scripts/core/emdash-api.d.ts with openapi-typescript from scripts/core/emdash-openapi.json (EmDash 1.2.0), and a path, a method or a body EmDash does not have fails dev:types. Where that list and the site differ (twelve requests it lacks, one it names PUT that is POST, one answer it shapes wrongly) is written down in emdash-requests.d.ts. A see...
Read more

v1.2.1

Choose a tag to compare

@joeblew999 joeblew999 released this 08 Oct 02:35

Changes since v1.2.0

  • a check task of this repo's own (test steps for every task, a green record, repo:ci), and how to release with charter
  • charter 0.16.0: releases are cut with charter's release task, which runs this repo's check here and does not wait for GitHub; the changelog is the releases' notes
  • charter 0.15.3: its workflow installs the mise this repo asks for
  • charter 0.15.2: one task, repo:ci, is what the repo-check workflow runs — the same here and on GitHub
  • charter 0.15.0: the tasks page is charter's own (docs-tasks, from tasks.toml), its repo-check workflow runs the docs and repo checks, and every file charter writes says so
  • charter for everything but the tasks themselves: docs in charter's layout, rules in docs/rules.md, AGENTS.md and CLAUDE.md as pointers, the generated pages declared in docs/_generated.toml, issue forms and labels and this repo's own tasks from charter 0.14.0 (tasks/repo); plans are issues

v1.2.0

Choose a tag to compare

@joeblew999 joeblew999 released this 08 Oct 01:44

Plugins from EmDash's registry, and previews. The full test on a Mac: 133 steps, every one of the
35 tasks, three sites at once and a deployed site with a preview. Not yet seen on Linux or Windows.

  • live:preview deploys the site as a preview: an address of its own, with its own database,
    bucket and sessions, beside the live site (Cloudflare's Worker Previews). LIVE_PREVIEW=<name>
    makes every task that takes --live act on that preview. signin:access covers the preview
    addresses.

  • Sites start one at a time. Two Cloudflare sites started in the same moment took the same
    debugger port, and one answered 500 to everything.

  • signin:token waits for a new site's tables instead of failing on a site that has just
    been started or deployed.
    Plugins from EmDash's registry: find one, install it with no clicking, and know it works. Run on
    macOS on a Cloudflare and a Node site; not yet against a deployed site, nor on Windows or Linux.

  • plugin:install -- <publisher>/<slug> installs a registry plugin as the admin's Install
    button does, with the token signin:token saved. From a site that is not even running it
    switches the sandbox on, builds, starts, signs in and installs. plugin:remove takes one out.

  • plugin:favourites installs five chosen plugins in one go — or a project's own list
    (PLUGINS). The list, why, and what was rejected: docs/favourite-plugins.md.

  • plugin:works says whether a plugin works, one line per check: builds, starts, listed,
    routes, admin pages in a real browser, log, sandbox. With no name: every plugin in the site.

  • plugin:sandbox makes the edits a sandboxed plugin needs (Cloudflare: deploying then needs
    the Workers Paid plan).

  • plugin:new and plugin:add write the plugin's lines in astro.config.mjs — nothing is
    left to edit by hand.

  • site:stop on a Node site also stops the sandbox process EmDash leaves running, which
    otherwise blocks the next start. Six new entries in docs/upstream.md.

  • One source for each thing in the docs. What a task does is its description in tasks.toml:
    docs/tasks.md (written by mise, in the order you use the tasks) and every list of tasks in the
    README are filled from it. The README says only what is not a task.

v1.1.0

Choose a tag to compare

@joeblew999 joeblew999 released this 08 Oct 00:23

From the first real deployment (the Remy-Sport site) and from running several sites at once. The
full test on a Mac: 101 steps, every task, three sites side by side and a deployed site, mise
2026.10.4. Not yet seen on Windows at the full level (the 1.0.1 run hit its time limit with no
failing step).

  • site:ports gives a project two free ports of its own, in mise.local.toml — so several
    sites, or several agents, run on one machine without meeting. The full test now runs three sites
    at once, and holds a lock while it uses the deployed test Worker.
  • EmDash's welcome dialog no longer appears: site:start and signin:token close it for the
    user they make, with the call the dialog's own button makes. EmDash has no setting for it.
  • A site set to Cloudflare Access works with the CLI on this machine again. After
    signin:access and its auth: access(…) line, site:start failed and every mise run emdash
    on the dev site said "Not authenticated": EmDash leaves out the address its CLI signs in at. The
    emdash task now takes the dev sign-in's token instead.
  • Emptying the local database removes the token saved for it (site:reset, site:admin).
    Left behind, the CLI sent it to the new database and was told "Invalid or expired token".
  • docs/tasks.md is written by mise itself (mise generate task-docs): every task with its
    description, arguments and flags, rebuilt on every test run, by mise run docs and by the
    commit check.
  • Working on emdash-run: mise run hooks turns on a commit check — a task without a test step,
    a test step for a task that is gone, or stale generated pages, and the commit is refused.

v1.0.1

Choose a tag to compare

@joeblew999 joeblew999 released this 07 Oct 11:46

Proven on macOS, Linux and Windows (the stages workflow), and by the full test on a Mac: 91 steps,
every task, both templates and a deployed site.

  • Windows: mise run emdash -- … with a quoted argument, and signin:token, were broken in
    1.0.0 — a shell took the argument apart. EmDash's CLI and wrangler are now run by Node itself.
  • Every task is safe to run again. site:new leaves an existing site alone, site:delete with
    no site has nothing to delete, plugin:new does not scaffold twice. The test runs them twice.
  • plugin:search is its own task.
  • tasks.toml is in order: the tasks by group, then every hidden step.
  • The test: two levels (test, test:full); one task alone (mise run test -- <task>); it
    refuses to run unless every task has a step; the README's table of tasks is written from it.
  • The docs site: https://joeblew999.github.io/emdash-run/. Issue forms and labels.

Use it: in your repo's mise.toml

[settings]
experimental = true

[tools]
node = "26"
pnpm = "12"
fnox = "1.36.0"

[task_config]
includes = ["git::https://github.com/joeblew999/emdash-run.git//tasks.toml?ref=v1.0.1"]

If you are on v1.0.0: change the tag to v1.0.1. On Windows this is required.

What works, task by task: https://github.com/joeblew999/emdash-run/blob/v1.0.1/docs/status.md

v1.0.0

Choose a tag to compare

@joeblew999 joeblew999 released this 07 Oct 09:49

The first release with signing in, deploying and a test whose result is a file:
docs/status.md says what works, task by task.

  • Signing in, four ways side by side: signin:token (a machine, no browser: an administrator
    and an EmDash API token written to the site's database — D1 or a Node site's SQLite file),
    signin:access (Cloudflare Access in front of a deployed site's admin, plus a pass for
    machines), signin:passkey and signin:open (EmDash's real wizard, and a signed-in window;
    the only tasks that need Playwright).
  • One rule for where a task acts: this machine by default, the deployed site (LIVE_URL) with
    -- --live — on signin:*, model:sync and emdash. emdash -- … --preview is this
    machine's built site. Every task prints where it is acting first.
  • Deploying: live:ship (checks, deploys with the site's .env as secrets, waits for the site
    to answer), live:undo, live:logs, live:backup. No SQL dump: wrangler d1 export refuses
    an EmDash database.
  • site:preview: the built site on this machine, which signs in the way a deployed one does.
  • emdash:update: the site on the newest EmDash, type-checked and built.
  • The test, and a status that matches the tasks. Two levels: mise run test (quick) and test:full
    (everything — both templates, then the deployed-site tasks). Each runs
    as another developer would — a clean environment, an empty config folder. Every step is
    recorded against the task it tests; docs/status.md has one row for every task in
    tasks.toml, and says NOT TESTED where there is none. The CI workflow runs the same thing, by
    hand or on a tag.
  • plugin:search is its own task.
  • The build is skipped when nothing it is made from has changed. Nothing is downloaded to delete
    a folder. A task run where there is no site says so.
  • emdash whoami behind Cloudflare Access is handled two ways, because EmDash's does not send
    the Access pass: the token is put in EmDash's own sign-in store, and the emdash task answers
    it when the token is only in the environment.
  • fnox is in the setup snippet. Only signin:access needs it — for a Cloudflare token that can
    change Access — and reads it by itself: no task needs a fnox exec prefix.
  • Gone: the nushell harness (reference/), its lint configs, plugins/README.md.

Use it: put this in your repo's mise.toml

[settings]
experimental = true

[tools]
node = "26"
pnpm = "12"
fnox = "1.36.0"

[task_config]
includes = ["git::https://github.com/joeblew999/emdash-run.git//tasks.toml?ref=v1.0.0"]

Docs: https://joeblew999.github.io/emdash-run/ · What works, task by task: https://github.com/joeblew999/emdash-run/blob/v1.0.0/docs/status.md

Tested on macOS. Linux and Windows: see the stages workflow run for this tag.

emdash-run 0.6.1

Choose a tag to compare

@github-actions github-actions released this 06 Oct 08:10

The first published release of the 0.6 line; 0.6.0 was tagged and never published.

  • Two sites side by side. SITE_PORT (default 4321) sets the dev server's port; give a second
    checkout its own in mise.local.toml.
  • A release is fast. A tag runs the fast check and publishes — about a minute. The full
    cross-platform matrix is a manual workflow, for when the cross-platform layer changes.
  • The installers trust the new config before changing it: choosing a template (| sh -s -- blog)
    failed on any machine that was not CI.
  • verify installs a fresh clone's site instead of assuming it.

emdash-run 0.5.0

Choose a tag to compare

@github-actions github-actions released this 06 Oct 07:48
  • Upgrading EmDash is one line and one command. Change EMDASH_VERSION, run mise run dev:
    it copies the local database to run/backups/, re-pins and installs in place, re-pins your
    plugins, and carries on. The plugin CLI, plugin-test and the Node sandbox runner versions are
    derived from the EmDash release — there is nothing else to keep in step.
  • One setup call instead of five requests. EmDash's own dev-bypass migrates, seeds through the
    site's storage and returns the admin token; the harness now just makes that call.
  • restore -- <package> --wipe --confirm works — the wiped site comes up empty, as an import
    requires. verify -- --restore proves the round trip.
  • Seeded images work on every template, and an unchanged seed is no longer re-applied on each
    dev (which overwrote admin edits and duplicated media).
  • A cold dev server is no longer starved. Readiness probes are patient and never overlap; that
    was the "hang" that blocked 0.4.0 and 0.4.2.
  • doctor --url fails on a pending migration (emdash migrate --check); rollback says that it
    does not undo migrations.
  • An encryption key from the start. dev generates EMDASH_ENCRYPTION_KEY into your gitignored
    .env and the site reads it; deploy refuses a Worker that has none.
  • A new project stays on the free plan's defaults. The Worker Loader binding (Workers Paid plan)
    is switched on by your first plugin:new, which says so — not by setup.
  • A seed's redirects, sections, blockTypes, relations, bylines, menus and widgetAreas
    are merged from both seeds.
  • Leaner: the task shape is declared once; about 100 lines of repeats and dead branches removed.
  • A push runs only mise run check on three OSes; the full matrix runs on a release tag.
  • mise run source -- emdashcms.com clones EmDash's own production site as a reference, beside the
    templates and the EmDash source; status lists the reference checkouts.
  • New docs: docs/emdash.md (EmDash, read from its source), docs/emdashcms-com.md (what a
    production EmDash site does that no template shows) and docs/agents/lessons.md.

emdash-run 0.4.1

Choose a tag to compare

@github-actions github-actions released this 06 Oct 06:51
  • Seeded images work on Cloudflare templates. emdash seed writes a seed's media to ./uploads,
    which a Cloudflare site never reads, so on blog-cloudflare every seeded image was broken locally
    and snapshot failed. dev now moves them into local R2. Known and upstream: each dev still
    adds duplicate media rows for a seed that has images.
  • upgrade takes the latest release (not main) and runs check afterwards, as it claimed to.
  • rollback refuses to run without a DEPLOY_URL instead of verifying the local site.
  • The linters and the skills CLI run through pnpm dlx: mise installs five tools, not nine.
  • Docs audited line by line against the code and corrected.

emdash-run 0.3.0 — every OS, Cloudflare or Node.js

Choose a tag to compare

@joeblew999 joeblew999 released this 06 Oct 06:23

Runs on macOS, Linux and Windows — on Cloudflare or plain Node.js. CI proves all six combinations with the same mise run verify you run locally.

curl -fsSL https://raw.githubusercontent.com/joeblew999/emdash-run/main/install.sh | sh              # Cloudflare template
curl -fsSL https://raw.githubusercontent.com/joeblew999/emdash-run/main/install.sh | sh -s -- starter # plain Node.js, no Cloudflare

Already installed? mise run upgrade.

  • Not locked into Cloudflare. The harness runs on the plain Node.js templates (starter,
    blog, …) as well as the *-cloudflare ones: a SQLite file and local uploads instead of D1 and
    R2. Same tasks. Sandboxed plugins work on both — under workerd on Node.js.
  • Windows works, with macOS and Linux — proven by CI, which runs the same mise run verify
    a dev runs, on all three, on a Cloudflare template and a Node.js one.
  • Portable by construction. The harness runs only programs mise installs: curl, find,
    printenv, open and friends are replaced by nushell's own commands, and check fails on any
    other program, on /dev/null, and on a Windows-unsafe glob.
  • New: mise run verify (does it work on this machine), verify:template -- <template>,
    verify:linux (a clean container).
  • Task arguments work on Windows. mise does not append a task's arguments to its command there,
    so plugin:new -- demo arrived with no demo. Every task now receives them through mise's
    usage mechanism, which behaves the same on all three OSes — and check sends spaces, JSON,
    a quote and a flag through and requires them back intact.
  • CI answers fast. check runs alone per OS (about 10–30 seconds); the site verifications run
    in parallel beside it. All of it is mise run … — nothing is CI-only.
  • install.sh takes a template: … | sh -s -- starter.
  • Not yet run by anyone: the PowerShell installer script itself (install.ps1). The harness it
    installs is proven on Windows by CI.