Add Prettier formatting checks

.claude/skills/wdl-deploy/SKILL.md

@@ -53,31 +53,30 @@ enable flags, `legacy_error_serialization`, and
 (`wrangler.json`, then `wrangler.jsonc`, then `wrangler.toml`). Both JSON
 filenames use Wrangler's JSONC syntax, including comments and trailing commas.
 The control plane is canonical for unsupported runtime shapes such as
-unsupported workerd compatibility flags and WDL-reserved injected module
-names. The CLI still fails fast for cheap local cases such as Python Workers
-modules, unmapped top-level or selected-env Wrangler runtime/deploy keys
-(`[site]`, `pages_build_output_dir`, `observability`, `limits`,
-`placement`, etc.), and ambiguous runtime `env` name collisions between
-`[vars]`, explicit bindings, and the implicit `ASSETS` binding.
-For an operator-enabled routed Worker, explicit `workers_dev = false` keeps its
-pattern routes active while disabling the default platform-domain URL; it
-requires at least one `route` / `routes` pattern and is not inferred.
-The deploy summary prints every active route-pattern URL hint, preserving the
-trailing `*` on prefix patterns, and includes the platform-domain URL only while
-it is enabled. Cloudflare's separate `preview_urls` field is unsupported and
-rejected by the CLI.
-WDL-only `[[exports]]`, `[[platform_bindings]]`, `[[triggers.schedules]]`, and
+unsupported workerd compatibility flags and WDL-reserved injected module names.
+The CLI still fails fast for cheap local cases such as Python Workers modules,
+unmapped top-level or selected-env Wrangler runtime/deploy keys (`[site]`,
+`pages_build_output_dir`, `observability`, `limits`, `placement`, etc.), and
+ambiguous runtime `env` name collisions between `[vars]`, explicit bindings, and
+the implicit `ASSETS` binding. For an operator-enabled routed Worker, explicit
+`workers_dev = false` keeps its pattern routes active while disabling the
+default platform-domain URL; it requires at least one `route` / `routes` pattern
+and is not inferred. The deploy summary prints every active route-pattern URL
+hint, preserving the trailing `*` on prefix patterns, and includes the
+platform-domain URL only while it is enabled. Cloudflare's separate
+`preview_urls` field is unsupported and rejected by the CLI. WDL-only
+`[[exports]]`, `[[platform_bindings]]`, `[[triggers.schedules]]`, and
 `[[services]].ns` are parsed by the CLI and removed from Wrangler's temporary
-bundle config; other fields retain their existing Wrangler passthrough
-behavior. Wrangler's object-shaped declarative `exports` config is unsupported.
-The dry-run child hides Wrangler's banner (and its normal update check) and
-disables anonymous telemetry. Wrangler may still consult the configured npm
-registry when reporting an unknown configuration field; project build hooks
-retain their normal network access.
-For `[[services]]` and `[[exports]]`, read `docs/deploy.md`: tenant JSRPC may
-delegate service or Durable Object class stubs as opaque capabilities, but the
-receiver cannot rewrite their host-authored caller properties. Keep delegated
-stubs in memory; long-term irrevocable stub storage is unsupported.
+bundle config; other fields retain their existing Wrangler passthrough behavior.
+Wrangler's object-shaped declarative `exports` config is unsupported. The
+dry-run child hides Wrangler's banner (and its normal update check) and disables
+anonymous telemetry. Wrangler may still consult the configured npm registry when
+reporting an unknown configuration field; project build hooks retain their
+normal network access. For `[[services]]` and `[[exports]]`, read
+`docs/deploy.md`: tenant JSRPC may delegate service or Durable Object class
+stubs as opaque capabilities, but the receiver cannot rewrite their
+host-authored caller properties. Keep delegated stubs in memory; long-term
+irrevocable stub storage is unsupported.
 
 Never recommend setting `CONTROL_CONNECT_HOST` outside local development: it
 overrides the TCP target the admin token connects to (Host header + TLS SNI

.github/workflows/ci.yml

@@ -30,6 +30,7 @@ jobs:
 
       - run: npm ci
       - run: npm audit --audit-level=moderate
+      - run: npm run format:check
       - run: npm run lint
       - run: npm run typecheck
       - run: npm test

.github/workflows/release.yml

@@ -25,6 +25,7 @@ jobs:
 
       - run: npm ci
       - run: npm audit --audit-level=moderate
+      - run: npm run format:check
       - run: npm run lint
       - run: npm run typecheck
       - run: npm test

.prettierrc.json

@@ -0,0 +1,22 @@
+{
+  "embeddedLanguageFormatting": "off",
+  "printWidth": 120,
+  "proseWrap": "always",
+  "quoteProps": "preserve",
+  "trailingComma": "es5",
+  "overrides": [
+    {
+      "files": "**/*.md",
+      "options": {
+        "printWidth": 80
+      }
+    },
+    {
+      "files": "**/*-zh.md",
+      "options": {
+        "printWidth": 80,
+        "proseWrap": "never"
+      }
+    }
+  ]
+}

AGENTS.md

@@ -19,6 +19,8 @@ newer target.
 
 - `npm install`: install dependencies from `package-lock.json`.
 - `npm link`: expose the local `wdl` binary for manual testing.
+- `npm run format`: format supported source and documentation files.
+- `npm run format:check`: verify formatting without modifying files.
 - `npm run lint`: run ESLint over the CLI, docs helpers, and examples.
 - `npm run typecheck`: run TypeScript's JavaScript-aware `tsc --noEmit` check.
 - `npm test`: run Node's built-in test runner against
@@ -31,9 +33,9 @@ newer target.
   configured — the CLI has no default endpoint.
 
 Use Node.js 22 or newer for local work. The GitHub Actions workflow uses Node
-22, runs `npm ci`, `npm audit --audit-level=moderate`, `npm run lint`,
-`npm run typecheck`, `npm test`, and `npm pack --dry-run`, then checks workflow
-syntax with actionlint.
+22, runs `npm ci`, `npm audit --audit-level=moderate`, `npm run format:check`,
+`npm run lint`, `npm run typecheck`, `npm test`, and `npm pack --dry-run`, then
+checks workflow syntax with actionlint.
 
 ## Coding Style & Naming Conventions
 
@@ -47,14 +49,14 @@ dependency injection for testable command behavior, as seen in
 types rather than `any`, and use `unknown` plus narrowing for values validated
 at runtime.
 
-Markdown wrapping is bilingual by design, normalized with Prettier
-(`--embedded-language-formatting=off`; code blocks are hand-formatted) and kept
-by editing habit rather than a linter: English prose hard-wraps at 80 columns
-(`--prose-wrap always`), while Chinese prose never hard-wraps inside a sentence
-(`--prose-wrap never`, one line per paragraph) because CommonMark renders a soft
-break as a space and CJK text would pick up spurious mid-sentence spaces.
-Tables, code blocks, and long URLs are exempt; there is no line-length lint,
-since table rows would trip it.
+Prettier formats supported files with a 120-column source width. Markdown
+wrapping is bilingual by design (`--embedded-language-formatting=off`; code
+blocks are hand-formatted): English prose hard-wraps at 80 columns
+(`--prose-wrap always`), while Chinese `*-zh.md` prose never hard-wraps inside a
+sentence (`--prose-wrap never`, one line per paragraph) because CommonMark
+renders a soft break as a space and CJK text would pick up spurious mid-sentence
+spaces. Tables, code blocks, and long URLs are exempt; run
+`npm run format:check` to enforce the repository configuration.
 
 ## Testing Guidelines
 
@@ -75,26 +77,27 @@ defaults change. The per-feature docs are bilingual pairs — `docs/<name>.md`
 (English) and `docs/<name>-zh.md` (Chinese) — and both languages are
 authoritative: update the pair in the same change. Agent-facing references
 (`templates/AGENTS.md`, the wdl-deploy skill, generated projects) point only at
-the English set. Before packaging, re-run the audit, test, and `npm pack
---dry-run` checks from Build, Test, and Development Commands.
+the English set. Before packaging, re-run the audit, test, and
+`npm pack --dry-run` checks from Build, Test, and Development Commands.
 
 ## Release
 
-Releases are tag-driven. `.github/workflows/release.yml` re-runs audit, lint,
-typecheck, and tests, verifies the tag matches `version` in `package.json`, and
-runs `npm pack --dry-run` — all before any publish, so a broken release fails the
-tag's check job and never publishes. It then publishes `@wdl-dev/cli` to npmjs
-(with provenance) and to GitHub Packages (authenticated with the workflow's own
-`GITHUB_TOKEN`), and creates a GitHub Release for the tag: final releases take
-their notes from the matching `CHANGELOG.md` section, pre-releases fall back to
-generated notes and are marked Pre-release. Do not run `npm publish` by hand.
+Releases are tag-driven. `.github/workflows/release.yml` re-runs formatting,
+audit, lint, typecheck, and tests, verifies the tag matches `version` in
+`package.json`, and runs `npm pack --dry-run` — all before any publish, so a
+broken release fails the tag's check job and never publishes. It then publishes
+`@wdl-dev/cli` to npmjs (with provenance) and to GitHub Packages (authenticated
+with the workflow's own `GITHUB_TOKEN`), and creates a GitHub Release for the
+tag: final releases take their notes from the matching `CHANGELOG.md` section,
+pre-releases fall back to generated notes and are marked Pre-release. Do not run
+`npm publish` by hand.
 
 Published npm versions are immutable (no reuse; unpublish only within 72 hours),
 but the check job gates every publish, so most releases tag the final version
 directly. This project ships documented breaking removals in 1.x minors (called
 out in the CHANGELOG) — do not hold or re-version a release for generic SemVer
 reasons. Stage a pre-release only for the narrower risk an RC actually guards:
-the *published artifact* differing from what the check job validated — packaging
+the _published artifact_ differing from what the check job validated — packaging
 changes (the `files` allowlist, entry points, the bundle/publish pipeline) or a
 large release you want to smoke-test as a real `@next` install. For an RC, set
 `version` to e.g. `2.0.0-rc.1`, write the CHANGELOG entry, commit, and tag
@@ -128,15 +131,15 @@ drops a `.env`-supplied endpoint when the effective token is not from the same
 `.env`), then the global token store (`~/.config/wdl/credentials`, managed by
 `wdl token`). The store is trusted (home directory, same-source token +
 endpoint) and not subject to the guard; a project `.env` is not. The namespace
-itself follows the same shape — `--ns > shell WDL_NS > project .env WDL_NS >
-store default (base WDL_NS)` — so the store's default namespace is the lowest
-selector, materialized into `env.WDL_NS` before the per-key gap-fill. Keep that
-ordering and the guard intact when touching `loadCliControlEnv` or
-`lib/token-store.js`; `--no-token-store` / `WDL_TOKEN_STORE=off` (via
-`tokenStoreReader`, read from the process env, not a project `.env`) must keep
-opting the store out of resolution entirely. Do not commit tenant tokens or
-generated secrets; read credentials from the environment and keep example
-configuration generic.
+itself follows the same shape —
+`--ns > shell WDL_NS > project .env WDL_NS > store default (base WDL_NS)` — so
+the store's default namespace is the lowest selector, materialized into
+`env.WDL_NS` before the per-key gap-fill. Keep that ordering and the guard
+intact when touching `loadCliControlEnv` or `lib/token-store.js`;
+`--no-token-store` / `WDL_TOKEN_STORE=off` (via `tokenStoreReader`, read from
+the process env, not a project `.env`) must keep opting the store out of
+resolution entirely. Do not commit tenant tokens or generated secrets; read
+credentials from the environment and keep example configuration generic.
 
 ### Deploy runs project code as you
 

CHANGELOG.md

@@ -6,17 +6,17 @@
 
 ### Added
 
-- Routed Workers with at least one custom route can set `workers_dev = false`
-  to disable their platform-domain URL. Deploy summaries print the active
-  platform URL and route-pattern URL hints, omitting the disabled URL.
+- Routed Workers with at least one custom route can set `workers_dev = false` to
+  disable their platform-domain URL. Deploy summaries print the active platform
+  URL and route-pattern URL hints, omitting the disabled URL.
 
 ## 1.5.1
 
 ### Security
 
-- Bump the bundled `wrangler` to 4.114.0, which vendors a patched sharp
-  (0.35.2) and clears a high-severity libvips advisory (GHSA-f88m-g3jw-g9cj)
-  reachable only through the miniflare dev server, which the CLI never runs.
+- Bump the bundled `wrangler` to 4.114.0, which vendors a patched sharp (0.35.2)
+  and clears a high-severity libvips advisory (GHSA-f88m-g3jw-g9cj) reachable
+  only through the miniflare dev server, which the CLI never runs.
 
 ## 1.5.0
 
@@ -41,11 +41,10 @@
 
 ### Fixed
 
-- `wdl deploy` no longer passes WDL-only `[[exports]]`,
-  `[[platform_bindings]]`, `[[triggers.schedules]]`, or `[[services]].ns` to
-  Wrangler's temporary bundle config. This avoids the `exports` schema
-  collision in Wrangler 4.107+ while preserving the fields in the WDL deploy
-  manifest.
+- `wdl deploy` no longer passes WDL-only `[[exports]]`, `[[platform_bindings]]`,
+  `[[triggers.schedules]]`, or `[[services]].ns` to Wrangler's temporary bundle
+  config. This avoids the `exports` schema collision in Wrangler 4.107+ while
+  preserving the fields in the WDL deploy manifest.
 
 ## 1.4.0
 
@@ -54,8 +53,8 @@
 - `wdl deploy` now selects `wrangler.json`, then `wrangler.jsonc`, then
   `wrangler.toml`, parses both JSON filenames as JSONC, and resolves Wrangler
   from explicit/project/package/`PATH` sources (`npx` remains opt-in). It
-  rejects Python modules, unmapped runtime fields, and binding collisions
-  before upload; runtime/workerd bundle policy remains canonical in control.
+  rejects Python modules, unmapped runtime fields, and binding collisions before
+  upload; runtime/workerd bundle policy remains canonical in control.
 - New projects use `compatibility_date = "2026-06-17"` unless a feature requires
   newer behavior.
 - Control requests now identify the CLI version, cap buffered JSON responses at
@@ -73,17 +72,17 @@
 
 ### Fixed
 
-- Control connection failures, invalid 2xx JSON, unreadable project `.env` or
-  D1 input files, and unexpected arguments now use escaped CLI diagnostics
-  instead of raw Node errors or silently ignored input.
+- Control connection failures, invalid 2xx JSON, unreadable project `.env` or D1
+  input files, and unexpected arguments now use escaped CLI diagnostics instead
+  of raw Node errors or silently ignored input.
 - Deploy failures now retain control warnings and actionable secret,
   environment-budget, and caller-secret guidance.
 - D1 migrations reject symlinked SQL files, invalid `.assetsignore` patterns
   identify the offending rule, and project `.env` parsing tolerates unrelated
   dotenv extensions.
-- Secret mutation failures explain environment-budget, contention, and
-  envelope errors, including when no mutation was written; token list safely
-  renders stored labels and endpoints.
+- Secret mutation failures explain environment-budget, contention, and envelope
+  errors, including when no mutation was written; token list safely renders
+  stored labels and endpoints.
 
 ### Security
 
@@ -103,14 +102,14 @@
 
 ### Changed
 
-- `wdl d1 execute` now requires exactly one SQL source (`--sql` or `--file`)
-  and rejects empty SQL locally before contacting control. Even `--sql ""`
-  conflicts with `--file`.
+- `wdl d1 execute` now requires exactly one SQL source (`--sql` or `--file`) and
+  rejects empty SQL locally before contacting control. Even `--sql ""` conflicts
+  with `--file`.
 - `wdl deploy` now rejects more malformed Wrangler config locally instead of
   silently dropping invalid input: non-array `kv_namespaces`, malformed KV
   entries, present-but-non-table `[assets]`, and several validated fields that
-  previously reached the manifest with loose types. Wrangler local-dev KV
-  fields `preview_id` and `remote` remain allowed but are ignored by deploy.
+  previously reached the manifest with loose types. Wrangler local-dev KV fields
+  `preview_id` and `remote` remain allowed but are ignored by deploy.
 
 ## 1.2.2
 
@@ -120,8 +119,8 @@
   world-writable store directory: a 0600 file there can still be deleted,
   replaced, or symlink-swapped by another user, so the write fails with a
   `chmod 700` hint instead. POSIX only.
-- `wdl r2 object head --json` parses custom metadata without prototype
-  pollution — an `x-amz-meta-__proto__` header stays an own key, and an empty
+- `wdl r2 object head --json` parses custom metadata without prototype pollution
+  — an `x-amz-meta-__proto__` header stays an own key, and an empty
   `x-amz-meta-` header is dropped.
 
 ### Changed
@@ -140,8 +139,8 @@
   dropped.
 - A local deploy prints a direct `http://<ns>.<domain>:8080/<worker>/` URL
   instead of a `curl -H 'Host: ...'` hint, and a control plane reached via a
-  `.test` / `.local` host is recognized as local (previously only
-  localhost / 127.0.0.1).
+  `.test` / `.local` host is recognized as local (previously only localhost /
+  127.0.0.1).
 - KV docs (GUIDE and `docs/kv.md`) describe the 512-byte key / list-prefix cap
   the platform now enforces; they previously said it was not checked. Documented
   that `wdl d1 execute --mode exec` takes no `--params`.
@@ -182,9 +181,9 @@
   values already were. A malicious or compromised control plane could put
   terminal control bytes (ESC / OSC / C1) in a JSON error property name and have
   them written unescaped to stderr (OSC 52 clipboard writes, display spoofing).
-- Control-plane responses now abort the connection when the body exceeds the
-  10 MiB cap, instead of rejecting the result while continuing to read the
-  stream — the cap bounds resource use, not just the returned value.
+- Control-plane responses now abort the connection when the body exceeds the 10
+  MiB cap, instead of rejecting the result while continuing to read the stream —
+  the cap bounds resource use, not just the returned value.
 - The trusted-publishing release job pins the npm CLI to an exact reviewed
   version instead of installing `npm@latest`, so a compromised npm release can't
   run in the job that holds the npm OIDC token and publish a tampered,
@@ -225,8 +224,8 @@
 
 ### Changed
 
-- `wdl init`'s `--ns` is now optional. With `--ns`, the scaffolded `npm run
-  deploy` keeps `wdl deploy . --ns <ns>`; without it the script is
+- `wdl init`'s `--ns` is now optional. With `--ns`, the scaffolded
+  `npm run deploy` keeps `wdl deploy . --ns <ns>`; without it the script is
   `wdl deploy .` and the namespace is resolved at deploy time (`--ns` / `WDL_NS`
   / project `.env` / a `wdl token` default). `init` also no longer autoloads
   control credentials, so a corrupt token store cannot block scaffolding.
@@ -244,9 +243,9 @@
 - `wdl secret put` no longer echoes the typed secret on a TTY: input is read in
   raw mode (hidden), and fails closed — it errors rather than echo if the
   terminal cannot hide input.
-- `.env` values containing literal backslash escape sequences (e.g. a token
-  with a backslash followed by `n`) now round-trip correctly instead of being
-  decoded as control characters.
+- `.env` values containing literal backslash escape sequences (e.g. a token with
+  a backslash followed by `n`) now round-trip correctly instead of being decoded
+  as control characters.
 
 ### Security
 

CONTRIBUTING.md

@@ -58,9 +58,11 @@ reaching the terminal**.
 
 ## Checks
 
-All of these run in CI on every pull request and must pass:
+Run `npm run format` to apply the repository formatting rules. The following
+checks run in CI on every pull request and must pass:
 
 ```bash
+npm run format:check
 npm run lint
 npm run typecheck
 npm test