@theholocron/cli 5.0.0-alpha.38 → 5.0.0-alpha.380

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -17,39 +17,122 @@ holocron --help
17
17
  ## Interactive mode
18
18
 
19
19
  Run `holocron` with no command, a parent command with no subcommand
20
- (`holocron auth`, `holocron skills`, `holocron upgrade`), or a leaf command
21
- missing a required positional (`holocron deploy`, `holocron secret set`), and
22
- you get a prompt instead of a `--help` dead end:
20
+ (`holocron auth`, `holocron repo`, `holocron repo upgrade`, …), or a leaf command
21
+ missing a required positional (`holocron deploy`, `holocron secrets set`), and
22
+ you get a prompt instead of a `--help` dead end.
23
+
24
+ The menu mirrors the command tree: each level shows only the next word, and a
25
+ group (marked `›`) opens its own submenu, as deep as the commands nest. Nothing
26
+ is shown as a `<command> <subcommand>` path — `config show` is `config ›` then
27
+ `show`. Every submenu has a `← back` item, and a command that is also a group
28
+ (`repo sync`, `deploy`) has a `(run it)` item at the top of its submenu. Typing
29
+ searches the whole tree — `upgrade` at the top finds the `repo` group — with
30
+ name matches ranked above description matches.
31
+
32
+ ```console
33
+ $ holocron
34
+ ? What would you like to do? › repo
35
+ ❯ repo ›
36
+ ? repo — choose a subcommand: › upgrade
37
+ ? repo upgrade — choose a subcommand: › node
38
+ ? Target Node.js major version: › 24
39
+ ```
23
40
 
41
+ For a command that declares options (so far: `package publish` — more are
42
+ rolled out one command group at a time), it also offers those: any required
43
+ option is prompted like a positional, and everything else goes through one
44
+ "Set any options?" checkbox, prompting only what you picked:
45
+
46
+ ```console
47
+ ? Set any options? (space to select, enter to skip) › ◉ npm dist-tag ◯ One-time password (2FA)
48
+ ? npm dist-tag (blank = latest): › beta
24
49
  ```
25
- $ holocron
26
- ? What would you like to do? › dep
27
- ────────────────────────────────────────────────
28
- ❯ deploy Trigger a deployment via the configured `deployment` capability
29
50
 
30
- ? Branch to deploy: › main
51
+ `package publish` specifically has three workflows that don't mix — a
52
+ steady-state publish, a one-shot `--initial` bootstrap, and a `--sync-trust`
53
+ maintenance run — so it asks which one first, then only offers that
54
+ workflow's own options; nothing from a different workflow is ever shown:
55
+
56
+ ```console
57
+ ? What would you like to do? › Initial publish (--initial)
58
+ ? Set any options? (space to select, enter to skip) › ◉ One-time password (2FA)
59
+ ...dry-run output...
60
+ ? Dry run complete — proceed with the real run? › yes
61
+ ? One-time password (2FA), if npm requires it for this publish: › ***
31
62
  ```
32
63
 
64
+ Each workflow's label ends with its real flag in parentheses when it has
65
+ one, and a longer caveat ("use only for a package's very first publish")
66
+ shows as helper text under the highlighted choice rather than bloating the
67
+ label itself.
68
+
69
+ A secret option (`--otp`) is masked as you type and delivered to the spawned
70
+ command through its environment, never its argv — `ps` and shell history
71
+ both expose a child's argv, never its env. A time-sensitive one like `--otp`
72
+ is checked off early but not actually typed until right before the real run
73
+ — after the dry-run preview and the "proceed?" question, never before —
74
+ since a dry run never checks it anyway and a code typed any earlier risks
75
+ going stale while something else is still being answered.
76
+
77
+ Whether an option is _checkbox-optional vs. always included_ and _collected
78
+ upfront vs. collected last_ are independent choices. `package publish`'s
79
+ `--otp` is always collected last, but whether it's checkbox-optional depends
80
+ on `HOLOCRON_NPM_ENFORCE_2FA`: unset (the default), it's opt-in like any other
81
+ checkbox pick; set in your own shell profile (never checked into the repo),
82
+ it's always included — for an operator whose own npm account enforces 2FA
83
+ on every publish, so the box doesn't need re-checking every single run.
84
+ Setting it has no effect on anyone else publishing from the same repo.
85
+
33
86
  The picked command spawns as a normal `holocron <command> …` invocation —
34
87
  `--token`, `--org`, `--cwd`, and `--dry-run` on the original invocation carry
35
88
  through. A non-interactive shell (CI, a script, a pipe) gets the old
36
89
  hard-failure message instead of hanging on a prompt.
37
90
 
91
+ For a command whose handler honours `--dry-run` (most of them — everything
92
+ except `auth *` and `config show`), the menu always previews it first — a
93
+ dry run doesn't mutate anything, so there's nothing to ask permission for —
94
+ then asks the one real decision:
95
+
96
+ ```console
97
+ ...dry-run output...
98
+ ? Dry run complete — proceed with the real run? › yes
99
+ ```
100
+
101
+ A preview that reports nothing would actually change (e.g. `clone` finding
102
+ every repo already cloned) skips that question too — there'd be nothing for
103
+ the real run to do either. `--skip-dry-run` opts out of the preview
104
+ entirely, straight to the real run, for when you already know what you want.
105
+
38
106
  ## Execution contexts
39
107
 
40
108
  Every command is tagged with how much of a repo it needs:
41
109
 
42
- | Context | Needs | Examples |
43
- | ------------ | -------------------------------------------------------- | -------------------------------------------------------------------------------------------- |
44
- | `global` | just the CLI binary | `version`, `clone`, `new`, `upgrade node`, `auth set` / `check`, `plugin create`, `skills …` |
45
- | `repo-aware` | `./holocron.config` in cwd, no plugins | `run`, `ci`, `config show`, `sync-readme` |
46
- | `workspace` | the `@theholocron/holocron-plugin-*` packages resolvable | `doctor`, `setup`, `sync`, `secrets sync`, `deploy` |
110
+ | Context | Needs | Examples |
111
+ | ------------ | -------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
112
+ | `global` | the CLI binary alone | `version`, `clone`, `new repo` / `plugin`, `repo upgrade …`, `auth set` / `check`, `package …`, `run lint commit-msg`, `skills …` |
113
+ | `repo-aware` | `./holocron.config`, walking up from cwd, no plugins | `run`, `run ci`, `config show`, `repo sync readme`, `deploy on-release` |
114
+ | `workspace` | the `@theholocron/holocron-plugin-*` packages resolvable | `doctor`, `repo setup`, `repo sync`, `repo cleanup branches`, `secrets …`, `deploy`, `sync github`, `sync issues` |
47
115
 
48
116
  A `workspace` command run from a bare global install (no plugins next to the
49
117
  CLI) fails with one actionable line — install the plugins as devDependencies or
50
118
  use `pnpm exec holocron <cmd>`. `auth check` degrades to "token present,
51
119
  verification skipped" instead. `holocron --help` prints the grouping; see
52
- [Execution Contexts](https://theholocron.github.io/holocron/execution-contexts).
120
+ [Execution Contexts](https://docs.theholocron.dev/holocron/execution-contexts).
121
+
122
+ ### Viewing the resolved config
123
+
124
+ `holocron config show` prints the resolved config. On a TTY it opens a foldable
125
+ tree by default; piped or redirected, it always emits plain, valid JSON
126
+ (`holocron config show | jq`).
127
+
128
+ ```console
129
+ holocron config show # TTY: foldable tree (↑↓ move, →/⏎ expand, ← collapse, E/C all, q quit)
130
+ holocron config show --depth 2 # static coloured tree, anything nested deeper than 2 levels collapsed
131
+ holocron config show --no-interactive # static coloured tree, fully expanded
132
+ ```
133
+
134
+ Passing `--interactive` explicitly needs a TTY on stdin and stdout and fails
135
+ with a one-line message otherwise; the default falls back to static output. It has no extra dependencies — nothing to install.
53
136
 
54
137
  ## Config file
55
138
 
@@ -101,20 +184,92 @@ in at load time:
101
184
  A minimal config — for repos with a `package.json` and a GitHub remote
102
185
  — only needs `providers`:
103
186
 
187
+ ### The `.holocron/` directory
188
+
189
+ A big declaration can live in a file next to the config instead of inline, so `holocron.config.ts` stays a short index
190
+ of choices. `.holocron/<area>.{ts,js,mjs,cjs,json}` is read by name, and the areas are a closed list (a file not on
191
+ the list is never read):
192
+
193
+ | File | Where it lands |
194
+ | --------------------------- | ---------------------------------------------------------------------------------- |
195
+ | `.holocron/notifications.*` | the `layout` option of the `providers.notifications` entry (which keeps `guildId`) |
196
+ | `.holocron/repo.*` | keys of `repo` (`defaultBranch` stays inline: Sentinel needs it first) |
197
+ | `.holocron/org.*` | provider options the config leaves out (`teamId`, `accountId`), see below |
198
+
199
+ The file's default export is the value (`.ts` keeps its comments; `.json` needs no `import`, for repos with no
200
+ `package.json`). If both exist for one area, `.ts` wins over `.json`. A missing directory changes nothing. Setting the
201
+ same thing inline and in a file is an error that names both (`the notifications layout is set in providers.notifications
202
+ and in .holocron/notifications.ts`, `repo.topics is set in the config and in .holocron/repo.json`): there is no merge and
203
+ no precedence. An area file may `import` a sibling in `.holocron/`. Never put a secret in the directory: it is public in
204
+ a public repo.
205
+
206
+ To type the layout, import the type and use `satisfies` (there is no `defineLayout()` helper; the type-only import keeps
207
+ runtime code out of a file Sentinel evaluates from the default branch):
208
+
209
+ ```ts
210
+ import type { NotificationsLayout } from "@theholocron/cli";
211
+
212
+ export default {
213
+ // comments stay
214
+ testChannel: "holonet",
215
+ } satisfies NotificationsLayout;
216
+ ```
217
+
218
+ This repo's own Discord layout lives in `.holocron/notifications.ts`; `holocron.config.ts` keeps only `guildId`.
219
+
220
+ `RULES.md` and the webhook avatar live in `.holocron/` too (`.holocron/pinned/rules.md`, `.holocron/media/webhook-avatar.png`),
221
+ but they are not areas: the layout names them with an explicit repo-relative path (`pinned: [{ channel, file }]`,
222
+ `webhookAvatar`). `.holocron/projects/<name>.json` is the project-board template folder (SPEC-0020), read by its own
223
+ default and not placed into the config.
224
+
225
+ **Sentinel** reads the directory too, so its PR config check, property sync and role sync see the same merged config as
226
+ the CLI. It lists `.holocron/` at the config's ref and fetches the whole tree (any depth, so an area may import a
227
+ sibling), within caps: 50 files, 100 KB per file, 512 KB in all. Past a cap, or on an area file that does not load,
228
+ the check reports a load error. Only what the config or an area file imports is evaluated. The design and the decisions
229
+ are in `docs/wiki/specifications/tech-holocron-directory.spec.md` (SPEC-0069).
230
+
231
+ **Keys that never move out of `holocron.config.ts`.** `repo.defaultBranch`, because the config is the one file read
232
+ before the directory, and the default branch is what says which branch holds `.holocron/` and the config. Every other
233
+ top-level key (`name`, `org`, `domain`, `providers`, `tasks`) also stays, but only because the closed list of areas has
234
+ no entry for it: Sentinel's `validateConfig()` reads none of them before it fetches `.holocron/` (it needs the config
235
+ file's path and the `ref`, nothing from inside it). Keys that a preset spreads into `repo` (`protection`, `properties`,
236
+ which `syncPropertiesFromConfig()` reads) stay inline too, since a key set both inline and in the file is an error. There
237
+ is no per-key precedence: the loader does not let an inline key win, it refuses the duplicate and names both places. This
238
+ repo's `.holocron/repo.ts` holds `teams`, `topics`, `variables` and `secrets`, names and identifiers only, typed with
239
+ `satisfies RepoConfig` (a type-only import).
240
+
241
+ **The `org` area** (#1820). The organization's constants that are provider options and that CI does not need:
242
+ `.holocron/org.ts` is `{ domain, cloudflare: { accountId }, vercel: { teamId }, checkly: { accountId }, sentry: { org } }` (typed with
243
+ `satisfies OrgConfig`). `domain` is the org's DNS zone: the loader sets `domain` on a config that leaves it out (set in
244
+ both is an error), and a file that builds a hostname while it is evaluated imports it (`import org from "./org.js"`). The loader fills each of these into every provider entry of that name that leaves the option
245
+ out, so the config can say `deployment: ["vercel", { domain }]`, `monitoring: "checkly"` and `errors: "sentry"`. The same option set inline
246
+ and in the file is an error that names both (`vercel.teamId is set in providers.deployment and in
247
+ .holocron/org.ts`), with no merge and no precedence. A provider key or an option the area does not know is an error; a
248
+ provider the config does not use is ignored. The org area is shared: a config in a subdirectory (`apps/sentinel`)
249
+ also reads the repository root's `.holocron/org.ts` (the nearest ancestor with a `.git`), and its own
250
+ `.holocron/org.ts` wins for a key set in both, option by option; that is not an error, so an identifier written in the
251
+ root file needs no copy in the app. The other areas (`repo`, `notifications`) describe the repository and are not
252
+ shared. Anything CI needs (`CLOUDFLARE_ACCOUNT_ID`,
253
+ `TURBO_TEAM`, `HOLOCRON_AXIOM_DATASET`) stays out: it lives in 1Password and syncs to a GitHub variable (`repo.variables.sync`),
254
+ and nothing else carries a copy. `.holocron/org.md` (the agent contract the CLI embeds) is not an area: only
255
+ `.ts`, `.js`, `.mjs`, `.cjs` and `.json` are read.
256
+
104
257
  ### repo options
105
258
 
106
- Additional `repo` fields recognised by `holocron setup`:
259
+ Additional `repo` fields recognised by `holocron repo setup`:
107
260
 
108
- | Field | Type | Description |
109
- | ----------------- | --------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
110
- | `repo.teams` | `Array<string \| { slug, permission }>` | GitHub teams granted repo access. String shorthand defaults to `push` (Write). `holocron setup` also writes `.github/CODEOWNERS` for teams with `push`/`maintain`/`admin`. |
111
- | `repo.topics` | `string[]` | GitHub topics set on the repository. |
112
- | `repo.protection` | `"balanced" \| "strict" \| "none"` | Branch-protection preset applied by `holocron setup`. For `"strict"`, the required status checks are derived from the task manifest — see below. |
113
- | `repo.properties` | `RepoProperties` | Org-level custom property values synced to the GitHub dashboard. |
261
+ | Field | Type | Description |
262
+ | ----------------- | --------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
263
+ | `repo.teams` | `Array<string \| { slug, permission }>` | GitHub teams granted repo access. String shorthand defaults to `push` (Write). `holocron repo setup` also writes `.github/CODEOWNERS` for teams with `push`/`maintain`/`admin`. |
264
+ | `repo.topics` | `string[]` | GitHub topics set on the repository. |
265
+ | `repo.protection` | `"balanced" \| "strict" \| "none"` | Branch-protection preset applied by `holocron repo setup`. For `"strict"`, the required status checks are derived from the task manifest — see below. |
266
+ | `repo.properties` | `RepoProperties` | Org-level custom property values synced to the GitHub dashboard. |
267
+ | `repo.variables` | `{ repo?, org?, sync?, prune? }` | GitHub Actions variables created and updated by `holocron repo setup`. Undeclared ones are left alone unless `prune` is on. `sync: [{ environment, keys, scope?, visibility? }]` names the vault environment `holocron secrets sync` fills Actions variables from (a separate environment from the secrets one; names only, a blank value is skipped, `doctor variables` expects the names). A name given as a literal and in `sync`, or `prune` with `sync`, is a config error. |
268
+ | `repo.secrets` | `{ repo?, org?, environments?, sync? }` | Names of the secrets that must exist, checked (never written) by `holocron doctor secrets`. Names only. The notification layout's secrets and a Chromatic project token are expected without being listed. `sync: [{ environment, keys?, scope? }]` declares the vault environments `secrets sync` fills from (it follows these entries, one step per scope, unless `--github-secret` is given; a blank value is skipped); the doctor expects their keys (every key the `vault` provider lists when `keys` is omitted, names only) at the `scope` (`repo`, `env=<name>`, `org=<name>`). |
114
269
 
115
270
  ### Custom properties synced to GitHub
116
271
 
117
- `holocron setup` and `holocron sync` both call `syncProperties()` with two
272
+ `holocron repo setup` and `holocron repo sync` both call `syncProperties()` with two
118
273
  kinds of fields — always one-way (`holocron.config.ts` → resolved →
119
274
  properties; properties are never a second editable source of truth):
120
275
 
@@ -139,7 +294,7 @@ properties; properties are never a second editable source of truth):
139
294
  | `holocron_compliance` | `"compliant" \| "non-compliant"` against a minimal `source` + `ci` baseline |
140
295
 
141
296
  Field definitions and derivation heuristics:
142
- `.notes/tech-holocron-platform.spec.md` → "Custom-properties sync — field
297
+ `docs/wiki/specifications/tech-holocron-platform.spec.md` → "Custom-properties sync — field
143
298
  definitions (#677)".
144
299
 
145
300
  The four derived fields' pure computation —
@@ -160,20 +315,27 @@ posts, instead of reporting pass/fail.
160
315
 
161
316
  ### Required status checks (`protection: "strict"`)
162
317
 
163
- `holocron setup` builds the branch-protection required-check list from the
318
+ `holocron repo setup` builds the branch-protection required-check list from the
164
319
  manifest, not a hand-maintained array:
165
320
 
166
- - `"DCO"` (hard-prepended), then
321
+ In the repo that holds the reusable workflows (holocron itself) the list is `ci / Required checks` and `DCO`. That is the aggregate
322
+ check (`ci/workflows/ci.yml`, written by `repo setup` / `repo sync` as `.github/workflows/ci.ci.yml`): it waits on
323
+ every `{ required: true }` task, the hand-written checks in `extraRequiredChecks` and its `codecov/*` statuses, and its
324
+ job summary lists each of them with a link, failing ones first. `DCO` stays its own required context. In any other
325
+ repo, which has no `ci` caller yet, the list is:
326
+
327
+ - `"DCO"`, then
167
328
  - every `{ required: true }` task's workflow check context (`Lint / Conclusion`,
168
- `Test / Conclusion`, `Typecheck / Conclusion`, `audit / Conclusion`), walked
329
+ `Test / Conclusion`, `Typecheck / Conclusion` while the repo is on the frozen reusables; `test:unit / Conclusion`,
330
+ `typecheck / Run tsc --noEmit`, … once it moves to the flat ones), walked
169
331
  in CI order, then
170
332
  - the top-level `extraRequiredChecks` (codecov gates, a bundle-build check, …).
171
333
 
172
334
  ```ts
173
335
  export default defineConfig({
174
336
  tasks: [
175
- { name: "lint", required: true },
176
- { name: "test", required: true },
337
+ { name: "lint:code", required: true },
338
+ { name: "test:unit", required: true },
177
339
  { name: "typecheck", required: true },
178
340
  ],
179
341
  extraRequiredChecks: ["codecov/patch", "codecov/project"],
@@ -186,11 +348,45 @@ The generated ruleset also sets `require_extra_approval_for_unattributed_changes
186
348
  false` — with `required_approving_review_count: 0` and a single maintainer, the
187
349
  GitHub default (`true`) is an unsatisfiable block on any PR with an
188
350
  unverified/unattributed commit. Repository-admin bypass still covers the real
189
- risk. And the `codecov.yml` `holocron setup` writes carries `if_not_found:
351
+ risk. And the `codecov.yml` `holocron repo setup` writes carries `if_not_found:
190
352
  success` on every status default, so an infra-only PR (dep bump, config,
191
353
  lockfile — no coverable change) passes instead of hanging on a `codecov/*`
192
- required check that never posts. A re-run of `holocron setup` backfills both into
193
- an existing repo.
354
+ required check that never posts. It also ignores `**/*.mdx`, so a docs-only PR that
355
+ adds pages is not judged to change testable files and left with an empty upload.
356
+ A re-run of `holocron repo setup` backfills all of these into an existing repo.
357
+
358
+ ### Default branch (`repo.defaultBranch`)
359
+
360
+ ```ts
361
+ export default defineConfig({
362
+ repo: { protection: "strict", defaultBranch: "alpha" },
363
+ providers: { source: "github" },
364
+ });
365
+ ```
366
+
367
+ `holocron repo setup` syncs GitHub's repo-level default branch (`HEAD`) — what a
368
+ fresh clone checks out, what `gh pr create`/the compare UI target by default,
369
+ and what Sentinel's `validateConfig()` reads (it only ever reads config from
370
+ the repo's default branch, never a PR ref — a deliberate security boundary).
371
+ Omit this field to leave GitHub's current setting untouched; most repos never
372
+ need it. It exists for a repo using a `main`/`alpha` prerelease-channel split
373
+ where active development happens on a non-`main` branch: set this to that
374
+ branch so Sentinel validates against what's actually being developed, then
375
+ change it back once that branch merges to `main` for a stable cut and stops
376
+ being the active branch.
377
+
378
+ ### Doc-comment coverage
379
+
380
+ ```ts
381
+ export default defineConfig({
382
+ repoValidation: { docsCoverage: "required" }, // "off" (default) | "warn" | "required"
383
+ });
384
+ ```
385
+
386
+ Controls `holocron probe docs patch` (skipped, reports without failing, or fails for an export a PR adds or changes
387
+ without a doc comment) and whether `holocron repo setup` adds the `documentation` flag and its statuses to
388
+ `codecov.yml`, informational for `"warn"`. It never touches the required checks. See the
389
+ [config docs](https://docs.theholocron.dev/holocron/config/#doc-comment-coverage).
194
390
 
195
391
  ### Git hooks
196
392
 
@@ -202,7 +398,7 @@ export default defineConfig({
202
398
  });
203
399
  ```
204
400
 
205
- `holocron setup` writes `.husky/pre-push` (runs `holocron ci` before every
401
+ `holocron repo setup` writes `.husky/pre-push` (runs `holocron run ci` before every
206
402
  push) and `.husky/pre-commit` (runs `gitleaks protect --staged` + `lint-staged`
207
403
  before every commit — scoped to staged changes, not the whole history, so a
208
404
  past false positive can't permanently block future commits) and sets
@@ -213,20 +409,22 @@ setup --hooks` / `--no-hooks` override per run. Bypass one push with `git push
213
409
 
214
410
  ### Task scripts
215
411
 
216
- `holocron sync` reconciles `package.json#scripts` from the `tasks` manifest
412
+ `holocron repo sync` reconciles `package.json#scripts` from the `tasks` manifest
217
413
  (merge, never clobber — only the managed keys are touched):
218
414
 
219
415
  ```ts
220
416
  export default defineConfig({
221
- tasks: ["test", "lint", "typecheck"],
417
+ tasks: ["test:unit", "lint:code", "typecheck"],
222
418
  // syncScripts: false, // disable the step
223
- // holocronScript: "node packages/cli/dist/cli.mjs", // override the "holocron" entry
419
+ // holocronScript: "node cli/cli/dist/cli.mjs", // override the "holocron" entry
224
420
  providers: { source: "github" },
225
421
  });
226
422
  ```
227
423
 
228
- Writes `"holocron": "<holocronScript ?? 'holocron'>"` plus one
229
- `"<task>": "holocron run <task>"` per runnable task. `syncScripts: false`
424
+ Writes `"holocron": "<holocronScript ?? 'holocron'>"` plus one script per runnable task, named by the task id
425
+ (`test:unit`) and one per suite (`test`, `lint`, `scan`): `turbo run <task>` in a pnpm workspace with a `turbo.json`
426
+ (or the tool's own command for a task that runs once at the root), `holocron run <task> --` in a single package.
427
+ `holocron probe generated check` fails on a root script that differs from what is generated. `syncScripts: false`
230
428
  opts out entirely.
231
429
 
232
430
  ### Splitting the task manifest — `astromech.config.ts`
@@ -240,7 +438,7 @@ The `tasks` key (plus `syncScripts` / `holocronScript` / `hooks` /
240
438
  import { defineConfig } from "@theholocron/astromech/config";
241
439
 
242
440
  export default defineConfig({
243
- tasks: [{ name: "audit", required: true, with: { "run-knip": true } }, "wiki"],
441
+ tasks: [{ name: "lint:unused", required: true }, "publish-wiki"],
244
442
  syncScripts: false,
245
443
  });
246
444
  ```
@@ -248,38 +446,56 @@ export default defineConfig({
248
446
  `@theholocron/astromech`'s `loadTasksConfig` merges it over
249
447
  `holocron.config`'s `tasks` — **task arrays concatenate**
250
448
  (`holocron.config` first), the dedicated file **wins on scalars**,
251
- `extraRequiredChecks` concatenate. `holocron run` / `holocron ci` /
252
- `holocron setup` / `holocron sync` all read the merged result, so the
449
+ `extraRequiredChecks` concatenate. `holocron run` / `holocron run ci` /
450
+ `holocron repo setup` / `holocron repo sync` all read the merged result, so the
253
451
  split is transparent. Useful for keeping the capability/provider config
254
452
  and the "what this repo runs" manifest in separate files (this repo does).
255
453
 
256
454
  ### `holocron run`
257
455
 
258
456
  ```bash
259
- holocron run <task> [job] [--required] [--filter <pkg>] [--dry-run] [-- <passthrough>]
457
+ holocron run <task | suite> [qualifier | job] [--check] [--required] [--filter <pkg>] [--dry-run] [-- <passthrough>]
260
458
  ```
261
459
 
262
- Runs one task locally — resolving turbo vs the package manager vs the tool
263
- itself. A `job` argument targets a sub-job of tasks that have them: `holocron
264
- run audit performance` runs Lighthouse CI, `holocron run audit knip` runs Knip,
265
- `holocron run audit` (no job) runs every audit sub-job in declared order. A
266
- sub-job whose tool isn't installed, or that has no local equivalent
267
- (`bundle-size`), is skipped with a note — `--required` makes that a failure. For
268
- a task with no sub-jobs the `job` slot is the first passthrough arg
269
- (`holocron run build src/`).
270
-
271
- **CI runs the identical command.** The reusable `typecheck` / `test` / `audit`
272
- workflows call the `holocron` composite action
460
+ Runs one task, or a suite, locally — resolving turbo vs the package manager vs the tool
461
+ itself. Ids are flat and verb-first: `typecheck`, `build`, `test:unit`, `lint:code`.
462
+ `holocron run test` runs the `test` suite (every `test:*` leaf the repo lists, in registry order, each reported),
463
+ `holocron run test unit` and `holocron run test:unit` run one leaf. The words are joined with `:` and the longest match
464
+ wins, so `deploy preview` is `deploy-preview`; the rest is the `job` argument. A `job` targets a sub-job of the tasks
465
+ that have them (`holocron run lint commit-msg <file>`). A leaf whose tool isn't installed, or that has no local
466
+ equivalent, is skipped with a note — `--required` makes that a failure. For a task with no sub-jobs the `job` slot is the first passthrough arg
467
+ (`holocron run build src/`). `holocron run format` writes (prettier), `holocron run format --check` is read-only.
468
+ `holocron run --help` lists the tasks grouped under the six intent groups. The old dotted names
469
+ (`verification.unitTests`, `sourceQuality.formatting`, …) still work and print a deprecation notice on stderr. <!-- holocron-allow-old-name -->
470
+
471
+ **CI runs the identical command.** The reusable workflows (`typecheck`, `test-unit`,
472
+ `lint-code`, …) call the `holocron` composite action
273
473
  (`theholocron/.github/.github/actions/holocron@main`), whose body is
274
- `holocron run <task> [job]`. So `holocron run test` locally and the `Test` CI
474
+ `holocron run <task> [job]`. So `holocron run test:unit` locally and the `test:unit` CI
275
475
  job execute the same thing — at a turbo root both are
276
- `turbo run test -- --coverage` (the registry's org-default flags flow through
277
- turbo's `--`).
476
+ `turbo run <task>`. `test:unit` always runs with coverage (`vitest run --coverage`); there is no `--coverage` flag.
477
+
478
+ ### `holocron deploy on-release`
479
+
480
+ ```bash
481
+ holocron deploy on-release [--channel <alpha|main>] [--from <commit>] [--to <commit>] [--dry-run]
482
+ ```
483
+
484
+ Deploys every workspace package whose own `holocron.config` declares
485
+ `{ name: "deploy", with: { on: "release" } }` (holocron#930), when the
486
+ release (`--channel`: its prerelease identifier, or `main`/empty for a stable
487
+ release) is on the task's `channel` (default `main`, the stable release) and changes the package or one of its
488
+ `workspace:*` dependencies between `--from` (the previous release's commit,
489
+ omitted when there is none) and `--to` (default `HEAD`). It waits for npm to
490
+ serve the release's new dependency versions, then runs `pnpm --filter <pkg>
491
+ deploy`. A package that doesn't apply is skipped; a deploy that
492
+ doesn't succeed exits non-zero. The shared `release` workflow runs it as its own
493
+ `deploy` job after the release job (opt in with `deploy-on-release: true`).
278
494
 
279
- ### `holocron ci`
495
+ ### `holocron run ci`
280
496
 
281
497
  ```bash
282
- holocron ci [--all] [--filter <pkg>] [--dry-run]
498
+ holocron run ci [--all] [--filter <pkg>] [--dry-run]
283
499
  ```
284
500
 
285
501
  Runs the merge-gating checks locally, in CI order, exiting non-zero on the
@@ -288,29 +504,53 @@ first failure — the "will my PR be green?" pre-flight, and what the
288
504
  (falling back to every `ci: true` task when nothing is marked required).
289
505
  `--all` forces the full set; `--filter` is a `turbo --filter=` passthrough.
290
506
  Order comes from the `CI_ORDER` constant in `@theholocron/astromech`
291
- (`typecheck → lint → test → build → …`). Tasks with no local equivalent
292
- (`codeql`, `deploy`) are reported as enforced-in-CI, never failures; `audit`
293
- expands into its sub-jobs (each under its own `audit / …` check context)
294
- unless the repo ships its own `"audit"` script.
295
-
296
- ### Lint parity
297
-
298
- The `lint` task takes an optional `linters` array — one list that drives
299
- **both** the CI `lint` job (super-linter's `VALIDATE_*` env, baked into the
300
- generated `.github/workflows/lint.yml` thin caller as a `super-linter-env`
301
- input) **and** `holocron run lint` locally. Omitted → auto-detected from the
302
- config files present plus the org always-on set.
507
+ (`typecheck → lint:code → … → test:unit → build → …`). Tasks with no local equivalent
508
+ (`scan:code`, `deploy`) are reported as enforced-in-CI, never failures. `format` runs in its
509
+ read-only `--check` form.
510
+
511
+ ### `holocron deploy`
512
+
513
+ `holocron deploy <branch>` (a Git-linked deployment) and `holocron deploy
514
+ --files <dir>` (a directory of build output, no linked repo) both trigger a
515
+ deployment through the configured `deployment` capability, then **wait for
516
+ it to finish**. The command reports success, exit code 0, only when the
517
+ deployment reaches `ready`. A deployment that ends `error` or `cancelled`,
518
+ or is still building after 10 minutes, is a failure: the exit code is
519
+ non-zero and the output includes the provider's own reason
520
+ (`DeploymentRecord.errorMessage`, e.g. Vercel's `Command "npm install"
521
+ exited with 1`). Providers build asynchronously, so accepting a deployment
522
+ isn't the same as it going live (holocron#911).
523
+
524
+ Provider plugins support this through the `Deployment` capability's
525
+ existing `getDeployment(id)`. They should set `errorMessage` on the
526
+ returned `DeploymentRecord` whenever the provider exposes one.
527
+
528
+ ### `holocron monitoring check <url>`
529
+
530
+ Runs every configured `monitoring` check against `<url>` instead of each
531
+ check's own configured `url` — no vendor API call, only a plain HTTP
532
+ request evaluated locally against the exact same `assertions` already in
533
+ `holocron.config.ts`. Built for CI: the `check-preview` task runs
534
+ this against a PR's Vercel preview deployment (triggered by
535
+ `deployment_status`, not `push`/`pull_request`), reusing the production
536
+ check definitions to catch a breaking change before merge instead of
537
+ after (holocron#1010).
303
538
 
304
539
  ```ts
305
540
  export default defineConfig({
306
- tasks: [{ name: "lint", linters: ["eslint", "prettier", "yamllint", "actionlint", "gitleaks"] }],
307
- providers: { source: "github" },
541
+ tasks: ["check-preview"],
542
+ providers: {
543
+ monitoring: ["checkly", { accountId: "...", checks: [/* ... */] }],
544
+ },
308
545
  });
309
546
  ```
310
547
 
548
+ Exits non-zero if any check fails — the job's own pass/fail is what
549
+ posts as the GitHub check, no separate reporting step needed.
550
+
311
551
  ### Skills installer
312
552
 
313
- `holocron setup` can install shared skills from `@theholocron/skills` into the local repo:
553
+ `holocron repo setup` can install shared skills from `@theholocron/skills` into the local repo:
314
554
 
315
555
  ```ts
316
556
  export default defineConfig({
@@ -371,18 +611,124 @@ export default acmeConfig;
371
611
 
372
612
  ```
373
613
 
614
+ ## Notifications — server as code
615
+
616
+ A `notifications` provider can declare a whole server in `holocron.config.ts`
617
+ (`layout`: categories, channels, roles, access, server settings, welcome screen,
618
+ pinned messages, onboarding, the bot application) and the CLI applies and checks it. The
619
+ commands and the config are provider-neutral; Discord is the first provider.
620
+
621
+ | Command | What it does |
622
+ | ------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
623
+ | `holocron notifications setup` | Applies the layout; looks everything up by name first, so a re-run changes nothing that matches |
624
+ | `holocron notifications channel add <name>` | Creates one channel (optionally with a webhook whose URL is stored as a secret) |
625
+ | `holocron notifications create-server --name <n> --owner <id>` | Creates the server the layout describes, writes its id into the config, applies the layout, invites the owner, waits for them to join and hands the server over (`--resume` finishes an interrupted hand-over; `--force` allows a second server) |
626
+ | `holocron notifications link` | Links your account on the provider to your GitHub account by signing in with the provider (OAuth, nothing secret kept) |
627
+ | `holocron notifications sync-roles` | Makes who holds a role match a GitHub team, through the links file (`--keep` only adds; `--provider` picks one) |
628
+ | `holocron project reconcile [board]` | Sets item statuses where a rule says so: an open blocker means `blocked`, no blocker left means `planned`, closed means `done` (status only; `--dry-run` previews; an open item with no status is reported). Runs hourly from `platform.projectReconcile.yml`, which `repo setup` writes when `providers.projects` declares boards |
629
+ | `holocron doctor notifications [--built-only] [--deliver] [--post]` | Step by step: does the server match the layout, do the secrets exist, and does a person (not the bot) own it. A full run also leaves one `🧪 test:` post in the layout's `testChannel`; `--built-only` posts nothing; `--deliver` posts to every channel and removes each post; `--post [channel]` posts the result as the bot (the failures, or only the summary when none failed). It also reads the Checkly feed (webhook, alert channel, subscriptions), the Comlink relay (Worker, route, bindings, the tracker's integration and rule) and the Notifications workflow (the file, and the secrets it reads), posting nothing, so `--built-only` keeps those steps. The `bot-permissions` row compares what the bot's roles hold server-wide with the permissions it needs (`ServerInspection` `botPermissions`; the owner and Administrator hold all, channel overwrites are left to the forum row) and, when some are missing, fails naming them in plain words with the invite link that asks for them again (not a credential); it is skipped, with the reason, when the roles could not be read |
630
+ | `holocron doctor variables` | Do the Actions variables `repo.variables` declares exist, with those values (and, for the organization's, that visibility)? One row for the repository and one for the organization, each only when declared. Read-only. A `repo.variables.sync` entry adds one row that checks its names exist (no value is read or shown). An undeclared variable counts only with `prune`; a scope that cannot be read (no provider, no org token) is a skip, not a failure |
631
+ | `holocron doctor ci` | Has the repo moved to the flat task vocabulary? One row each for the `tasks:` entries (a deprecated alias, with the id that replaces it), the generated callers (file, workflow and job name equal the task id), the `ci` aggregate's `expected` list (no id missing or stale) and where the callers point (`theholocron/.github` or holocron `ci-v1`). Read-only, no provider needed; exits 1 on a failure. The ruleset's checks are `doctor repo`'s |
632
+ | `holocron doctor github-app` | Does each GitHub App instance in `githubApps` (`{ name, manifest, environment }`) have what its manifest file (`default_permissions`, `default_events`) declares? Reads the App's id and key from the vault `environment` in memory (never printed), calls `GET /app` and `GET /app/installations` with the App's JWT. Rows per instance: `reachable`, `events` (all subscribed), `permissions` (each granted at least at the manifest's level, `read < write < admin`), `installations` (none lags the App, a pending update fails). Extras on the App are `info` in a row's detail, never a failure. A gap ends with the App's settings page (user- or organization-owned) and, for an installation, its own page. Credentials that cannot be read are a skip; an unreadable manifest or a rejected JWT fails. Read-only |
633
+ | `holocron doctor environments` | Do the deployment environments exist? The ones `repo setup` creates (`staging`, `production`, when an `environments` provider is configured) plus every environment `repo.secrets.environments` names. One row. A missing one fails; an extra one is never a problem; no provider with some named is a skip |
634
+ | `holocron doctor project [--post [channel]]` | Does each board `providers.projects.boards` declares exist and match its template? One row per board, status list, field, view and automation rule; the board's extras and the items `project reconcile` would fix (no status, closed but not done, open blocker but not blocked, blocked without a blocker) are `info` rows, never a failure. A template file that does not read fails, naming the file. Read-only; `--post` shares the failing rows through the `notifications` bot. |
635
+ | `holocron doctor credentials` | Does the credentials environment exist, and does it hold every variable the configured plugins take? One row per variable and provider (`ok`, `fail` when missing, `skip` for an optional one not set), listing names only: it never reads a value, and says which spelling of the environment's name matched. |
636
+ | `holocron doctor repo` | Do the repository settings `repo setup` writes match the config? One row per group the config declares: description and homepage, default branch and merge settings, topics, custom properties, team access, branch protection; labels and security settings are always checked. Read-only. Reads through optional `Source` methods (`getRepoInfo`, `getProperties`, `listLabels`, `listTeamAccess`, `getRulesetDetail`, `getSecurity`); a group that cannot be read is a skip, not a failure |
637
+ | `holocron doctor dns` | Do the DNS records the config makes setup write exist, pointing where they should? The preview domain's apex and wildcard CNAMEs (Cloudflare), the docs host a wiki provider asks for, and the relay Worker's proxied placeholder; one row per zone. A missing record, another target or a missing proxy fails; an extra record never does. A record `ensureCustomDomain()` returns is not checked, because asking for it creates the domain |
638
+ | `holocron doctor monitoring` | Do the checks the `monitoring` provider declares exist in its account, matched by name, with the URL, method and (when stated) frequency, locations, activated, should-fail, body and assertions the config sets? One row. A check in the account that is not declared is never a problem. A provider that cannot list its checks, failed to load or has no token is a skip |
639
+ | `holocron doctor secrets` | Do the secrets the config says must exist (`repo.secrets`, plus the notification layout's webhook and role secrets and the Chromatic token) exist? One row for the repository, the organization and each environment, each only when something is expected there. Names only; an undeclared secret is never a problem. A scope that cannot be read is a skip |
640
+ | `holocron doctor tokens` | For every token a configured plugin lists (every plugin that resolves one: GitHub's nine, Discord's three, Sentry's two, ...), looked up as the plugin does (its env var, the vendor's, the repo's `.env` and the `credentials` providers, then keyring; a `credentials` provider that cannot be read is a failing row) and verified with its `verifyToken`: set and accepted is ok (with how long is left, when the token expires), set and rejected fails, a token with 7 days or fewer left fails with `expires in N days: renew it`, not set is a skip. No token is printed. What a token may do is not checked, because GitHub does not return a fine-grained token's permissions |
641
+ | `holocron doctor --all` | Runs every deep check (`doctor notifications`, `doctor variables`, ...) and prints one row each, then `N ok, N fail, N skipped` counted in checks. Exits 1 when any check failed. A check that cannot run (no provider, no token) is a skip with the reason. Read-only: notifications runs as `--built-only`, so it posts nothing |
642
+ | `holocron provider [<name> [setup\|doctor]] [--with <file or key=value>]... [--dry-run] [--yes]` | Set up or check what only one provider has (a Linear team's labels, project and auto-close/auto-archive), from that provider's own block of `holocron.config`, or, for a provider the config does not name, from `--with` (repeatable: `team=ENG`, or a `.json`/`.ts`/`.js` file holding the provider's options object; it goes over the config entry's options where there is one, never writes the config, and the token still comes from `--token` or the usual chain). A plugin offers it by returning `setup` / `doctor` from `createPlugin`; `doctor` is read-only, `setup` is idempotent, never deletes, supports `--dry-run` and asks first unless `--yes`. Rows are the shared `printCheckRows` ones; nothing secret is printed |
643
+ | `holocron probe [<provider> [<probe>]] [--dry-run] [--yes] [--lcov <path>]` | Poke a live provider on purpose: list the probes, inspect something, or run a proof that cleans up after itself. Plugins offer the probes; nothing secret is printed; one that changes things asks first |
644
+
645
+ Channels fed by another service are declared with `feeds` on the channel: `{ name: "uptime", webhook: true, feeds: { monitoring: true } }` (the object form `{ name?, webhook?, avatar?, alerts? }` overrides the alert channel's name, default `Message to #<channel>`, the dedicated webhook's name, default the provider's, and gives that webhook a picture, a file path or `https://` URL, or `{ url, image }` (the URL is tried first, the file in the repo is the fallback), else the provider's own icon (`Monitoring.icon`, a URL and an embedded `data:` image; Checkly provides one), else the layout's `webhookAvatar`). A channel's own `webhook.avatar` and the layout's `webhookAvatar` take the same forms, a path or `https://` URL, or `{ url, image }` (the URL first, the file in the repo as the fallback); a relative path is read from the repository root. `setup` gives the channel a dedicated webhook and hands its URL straight to the `monitoring` provider through `Monitoring.ensureAlertChannel({ name, format: "discord", url? })`, which finds an alert channel of that name or creates one and subscribes every check; the URL is never stored as a secret. It needs a `monitoring` provider that implements `ensureAlertChannel` (Checkly does). `alerts: { failure?, recovery?, degraded?, sslExpiry?, sslExpiryThreshold? }` (`MonitoringAlertSettings`, passed as `ensureAlertChannel({ ..., alerts })`) sets what the alert channel alerts about: a new channel defaults to failure, recovery and degraded on and `sslExpiry` off, and on an existing channel a setting left out is left alone. A check in the `monitoring` provider's `checks` can route itself with `alertChannels: ["Message to #uptime"]` (`MonitoringCheckInput.alertChannels`, the alert channel names): it is then subscribed to those channels only, and a name no feed declares is refused by `setup`. `inspectAlertChannel` returns the channel's `alerts` and the `misrouted` checks that `doctor notifications` reports. Two feeds may not share an alert channel name, and a feed's webhook may not share the channel's own webhook name (compared without regard to case).
646
+
647
+ Channels fed by the error tracker add `feeds: { errors: true }` and a layout `relay: { host, zone?, script?, code? }`: `setup` makes a dedicated webhook, points `host` at the Comlink Worker (a proxied DNS record via `dns`), and deploys it via `workers.deployScript` with the webhook URL as `COMLINK_<SOURCE>_WEBHOOK`. The URL never leaves the Worker's secret; other secrets on the script are kept. With `relay.errors: { projects, integration?, rule?, frequency? }` it also ensures the tracker's integration and alert rule (`errors.ensureRelay`) and deploys the Worker with the integration's signing secret. Needs `errors`, `workers` and `dns`; only one channel may be fed.
648
+
649
+ `holocron secrets sync <env> --worker <script>` also sets every key of the vault environment on a deployed Worker as a secret, through the optional `Workers.setSecret(script, name, value)` (write-only; the Cloudflare plugin implements it). The Worker must already exist. Sync only adds and updates: it never removes a binding, because a Worker also holds secrets that are not in the vault (for Comlink, the `COMLINK_<SOURCE>_WEBHOOK` that `notifications setup` sets), so a key deleted from the vault must be deleted from the Worker separately. Keys are written one at a time and each is reported; if one fails, the exit code is 1 and re-running finishes the rest, since every write is idempotent.
650
+
651
+ `holocron secrets sync <env>` also writes GitHub Actions **variables** when `repo.variables.sync` has an entry for that environment: each key goes through the `source` provider's `syncVariables` (repo scope) or `syncOrgVariables` (`org=<name>`), one call per key and never pruning, so a name the vault lacks or holds blank is skipped and GitHub keeps its value. Keep these in a vault environment of their own (a vault does not tell a visible value from a secret one, and a variable is readable afterwards). An environment declared only by variable entries pushes nothing to `secrets` or `deployment`. Values are never printed. `doctor variables` checks that the names exist at their scope.
652
+
653
+ `holocron secrets list [--scope repo|environment|organization] [--env <name>] [--json]` prints the names (never the values; GitHub does not return them) of the secrets at each scope, sorted, under a heading per scope. With no `--scope` it lists the repository, plus the organization (`--org`, `HOLOCRON_ORG` or the config's `org`) and the `--env` environment when they are known; `--scope` lists one. A scope that cannot be listed is reported under its heading and the rest are still listed; the exit code is 1 if any failed. `--json` prints `[{ scope, names, error? }]`. It is read-only, so `--dry-run` does not apply. Comparing the names with what the config expects is `doctor secrets`, which also covers the keys of the vault environments declared in `repo.secrets.sync`.
654
+
655
+ A channel with `kind: "forum"` is a forum, with `tags` (each a `name`, an `emoji` and whether it is `moderated`); `setup` adds and corrects the tags but never removes one, and a text channel of the same name must be deleted first, since neither kind can become the other. A forum takes no `webhook` and none of the webhook feeds (`monitoring`, `errors`); its one feed is `feeds: { prs: true }`, the pull-request feed, where Sentinel keeps one post per pull request tagged with its state. The object form `feeds: { prs: { exclude: { repos?: ["owner/name"], authors?: ["login"] } } }` leaves some pull requests out. `feeds.prs` belongs to a forum (a text channel with it is refused) and the forum's `tags` must hold all eight states, `draft`, `conflicts`, `failing`, `changes requested`, `ready`, `waiting`, `merged` and `closed`, or `setup` refuses the layout. `setup` does nothing more for it; `doctor notifications` checks that the forum exists, has those tags and that the bot can start posts and manage threads in it (`ServerInspection` `botThreads`, filled in by the Discord provider after the forum's own permission overwrites). `defaultReaction` is the emoji Discord offers on every post (its Get Started step "Set a default reaction for posts"); Discord allows only one, so `starterReactions` lists more (`["👍", "👎", "❓"]` for approve, disapprove, question) that the bot adds to the first message of each post `specs sync` makes. Existing posts are left alone, a reaction the bot cannot add is skipped, and `doctor notifications` checks `defaultReaction`. Unicode emoji only.
656
+
657
+ `holocron specs sync [--channel <forum>] [--files <paths...>] [--pr <number>] [--refresh] [--refresh-origin] [--refresh-titles] [--withdraw] [--notify]` keeps one forum post per ADR and spec in step with the files: it makes the post (titled with the document's own title, tagged with the file's status), writes its address into the file as `discussion.url`, and turns a gatekeeper's retag into an edit of the file's `status:` line in the working tree for a pull request. `--files` syncs only those files and `--pr` names a pull request on their posts, for a pull request that is still open. `--refresh-origin` is a one-off repair: it rewrites the `PR:` line of existing posts to the one the current rule names (the optional `prLine` of `DiscussionPatch`, and of `DiscussionPost`), and nothing else in them. `--refresh-titles` is the same kind of one-off repair for the title: it renames each existing post whose title differs from the current rule's (the optional `title` of `DiscussionPatch`, and of `DiscussionPost`) and changes nothing else; a rate limit stops it cleanly, reports how many documents are left, and the next run continues. The id that opens a post is a link to the file on the default branch (`repo.defaultBranch`, else `HEAD`) once the file is there, and plain text while `--pr` names an open pull request; the normal sync rewrites a post's id line (and only that line) whenever it is not the one the rule gives and no `--pr` is given, which covers a file that moved and a post made while its pull request was open, so it needs no extra request when every post is right and `--refresh` is no longer needed for it. `--refresh` runs the `PR:`, title and id-link repairs in one pass (the id link is the optional `idLine` of `DiscussionPatch`, and of `DiscussionPost`, which changes only the message's first line); `--refresh-origin` and `--refresh-titles` still do only their own. `--withdraw` (with `--files`, on the base branch) tags the posts of a pull request closed unmerged `withdrawn` and closes them; `withdrawn` is a tag only, never a file `status:`, and the next sync puts the file's own status back. It cannot be combined with `--refresh`, `--refresh-origin` or `--refresh-titles`. `--dry-run` previews it. The posts behind it are also usable for other keys, such as a pull request (`owner/repo#123`, the state a tag name): `DiscussionPatch.body` replaces the first message's text (the provider keeps its footer), and the optional `Notifications` methods `postDiscussionMessage(channel, key, content, { quiet })` (a closed post is unarchived, written to and closed again), `readDiscussionMessages(channel, key, limit)` (text, newest first, 50 by default) and `dedupeDiscussion(channel, key)` (keeps the oldest post of the key, deletes the rest, returns how many) append to, read and repair a post. See the [`specs sync` page](https://docs.theholocron.dev/holocron/commands/specs-sync/).
658
+
659
+ Pinned messages are declared under the provider's `layout` as `pinned: [{ channel, content }]` or `pinned: [{ channel, file }]` (`channel` is a channel in the layout, `content` up to 2000 characters, `file` a text file in the repository, such as `.holocron/pinned/rules.md`, whose contents are the message and win over a hand edit; `doctor notifications` reports the drift); the layout is the whole set of messages the bot has pinned in those channels.
660
+
661
+ Config shape added for role sync and sign-in (all under the provider's `layout`):
662
+
663
+ | Field | Meaning |
664
+ | ----------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
665
+ | `testChannel` | The channel (it must declare a webhook) the doctor posts its one self-removing test message to. |
666
+ | `roles[].team` | A GitHub team slug. The role follows that team: given to linked members, taken from linked people who left. Never touches holders nobody has linked. |
667
+ | `links.hostedUrl` | A hosted sign-in service (Sentinel). `link` asks `<hostedUrl>/api/link` and uses it when it answers `{ "service": "link" }`; otherwise it signs in locally. |
668
+ | `links.localPort` | The port of the local sign-in listener (default `53682`, bound to `127.0.0.1` only). |
669
+ | `links.repo` / `links.path` | Where the links file lives (default `<org>/.github-private` and `notifications/<provider>-links.json`). |
670
+ | `application.intents.members` | Switches the Server Members Intent on or off (needed to list who holds a role). |
671
+ | `application.redirectUris` | Redirect URIs that must be registered. Check-only: the API cannot set them, so `setup` and the doctor say what to add in the developer portal. |
672
+
673
+ The links file is `{ "version": 1, "links": [{ "github": "<id>", "login": "<login>", "member": "<id>" }] }`: two public ids per person.
674
+ It is read and written through the `source` capability with the `github.user-sync` token (fine-grained, **only** the
675
+ private config repo, Contents read and write), so no other token needs Contents on other repositories.
676
+
677
+ Capability surface behind it (all optional, so providers without the concept omit them):
678
+
679
+ - `Source.teamMembers(teamSlug, org?)`, `Source.readFile(path, repo?)` (returns the `version` a later write must present),
680
+ `Source.writeFile(path, content, { message, version?, repo? })` (a changed `version` is refused: a real conflict check).
681
+ - `Notifications.createServer({ name, force? })`, `inviteLink(channel)`, `hasMember(id)`, `transferServer(id)` and `serverOwner()`: create a server, invite its owner, check they joined, hand it over, and say who owns it.
682
+ - `Notifications.roleMembers(role)`, `addRoleMember(role, memberId)`, `removeRoleMember(role, memberId)`.
683
+ - `Notifications.checkDelivery({ channel, name, category?, text })`: posts through a channel's webhook, reads the post back and removes it, and says which step failed.
684
+ - `Notifications.signIn()` returning `{ authorizeUrl, complete }`: the provider-neutral shape of "prove which account you own".
685
+ - `Notifications.ensureApplication` returns `{ updated, todo? }`; each `todo` is a step only a person can do (a portal setting
686
+ the API ignores) and is reported as a failure until done.
687
+
688
+ Provider credentials for Discord: `discord.holonet` is the **bot token** (Bot → Reset Token), `discord.holonet.oauth` is the
689
+ **OAuth client secret** (OAuth2 → Reset Secret). They are different values; the API can create or read neither.
690
+
691
+ See the [command pages](https://theholocron.dev/holocron/commands/notifications-setup/) for each command's options.
692
+
693
+ ## Syncing issues
694
+
695
+ - **`holocron sync issues --from <a> --to <b> [--since <time>] [--yes]`** — one direction of an issue sync between two
696
+ trackers (`github`, `linear:team=ENG`); swap the ends for the other direction. Unlike `holocron migrate` it updates what
697
+ it made: it creates a missing issue, sets title, description, state, priority, labels and assignee when they differ, and
698
+ adds the comments the target lacks (oldest first, never edited or deleted). It writes nothing to the source, and changes
699
+ a target issue only when a migrate or sync made it (the recorded `holocron:migrated-from` marker or attachment). A source
700
+ issue that itself carries such an address is skipped, which stops the two directions echoing each other. The report
701
+ counts created, updated, unchanged, skipped (guard) and left, and ends with the `not carried across` list. A rate limit
702
+ stops the run with a `left` count. A target needs the optional `Issues.sync`; GitHub and Linear have it.
703
+
704
+ ## Cleaning up
705
+
706
+ - **`holocron repo properties apply [--yes]`** — adds the capabilities the organization's `holocron_capabilities`
707
+ custom property does not allow yet (`holocron doctor repo` names them). The allowed values are organization-wide, so it
708
+ asks first (`--yes` skips the question; `--dry-run` only reads). It only adds values: nothing is removed or reordered
709
+ and the rest of the definition is kept. It needs a token with the organization's Custom properties: admin.
710
+ - **`holocron repo cleanup branches [--local | --remote]`** — lists the branches whose pull request is merged,
711
+ grouped, each saying whether it lives locally, on the remote or both; then asks, and offers a checklist with the
712
+ proven-merged ones ticked. Proven merged means an ancestor of `origin/alpha` or `origin/main`, or the head (or an
713
+ ancestor, or one of the commits) of a merged pull request. Branches with more commits than their merged PR, with
714
+ an open PR, checked out (here or in a worktree), or never merged are listed unticked. `--dry-run` only lists;
715
+ `--yes` deletes every proven-merged branch without asking. Pull requests and remote branches come from the
716
+ `source` provider (`Source.listPullRequests`, `listPullRequestCommits`, `listBranches`, `deleteBranch`).
717
+
374
718
  ## Upgrading
375
719
 
376
- - **`holocron upgrade node <to>`** — patches every Node.js version pin
720
+ - **`holocron repo upgrade node <to>`** — patches every Node.js version pin
377
721
  (`engines.node`, `.nvmrc`, `.node-version`, `.github/workflows/*`,
378
- `Dockerfile`, `.tool-versions`) to a new major.
379
- - **`holocron upgrade deps`** — bumps every `@theholocron/*` pin in
722
+ `Dockerfile`, `.tool-versions`) to a new major. The rewrite is
723
+ `runUpgradeNode()` in `@theholocron/astromech/upgrade`; the command only
724
+ binds the logger and reads `upgrade.node.extra`.
725
+ - **`holocron repo upgrade deps`** — bumps every `@theholocron/*` pin in
380
726
  `pnpm-workspace.yaml` (`catalog:` and every named `catalogs.*`) to its latest
381
727
  published version, then migrates `holocron.config.ts` from the 7.x preset
382
728
  shape (`const { repo, workflows } = node()` + `repo.requiredChecks` +
383
729
  `workflows:`) to the current composed-preset API (`const preset = node()` +
384
730
  `...preset` + `tasks:` + `extraRequiredChecks:`). Mechanical transforms only —
385
- review the result, then `pnpm install && holocron ci`. `--pins-only` skips
731
+ review the result, then `pnpm install && holocron run ci`. `--pins-only` skips
386
732
  the config migration; `--dry-run` prints without writing.
387
733
 
388
734
  ## Auth — fine-grained tokens
@@ -397,37 +743,60 @@ holocron auth set github.issues ghp_xxx # issue management
397
743
  holocron auth set github.sync ghp_xxx # sync-github workflow templates
398
744
  holocron auth set github.release ghp_xxx # semantic-release
399
745
  holocron auth set github.admin ghp_xxx # setup, secrets, environments
746
+ holocron auth set github.user-sync ghp_xxx # the files that link GitHub accounts to other services' members
400
747
  ```
401
748
 
402
749
  The resolution chain per capability is:
403
750
 
404
- ```
751
+ ```text
405
752
  --token flag → HOLOCRON_<FEATURE>_TOKEN env var → keyring("github.<feature>")
406
753
  ```
407
754
 
408
755
  See [`docs/tokens.md`](../../docs/tokens.md) for required PAT scopes per feature.
409
756
 
757
+ A plugin lists the tokens it takes in a `tokens: PluginToken[]` export (each resolver exposes its own as `.token`), as
758
+ `{ envName, vendorEnvName?, keyringKey, verifiable?, optional? }`, so `holocron doctor tokens`, `doctor credentials` and
759
+ `auth export` can find them. Set `optional: true` on a token the plugin only falls back to, such as the Discord
760
+ webhook a post uses when it names no channel: the checks that expect every listed token do not expect it, and
761
+ `auth export` marks it `(optional)` when it has no value. `verifiable: false` marks a token `verifyToken` does not
762
+ apply to.
763
+
410
764
  Additional `auth` subcommands:
411
765
 
412
766
  ```sh
413
767
  holocron auth check github.read # re-verify a stored token
414
768
  holocron auth unset github.read # remove a stored token
415
769
  holocron auth list # show all stored providers
770
+ holocron auth export # tokens as NAME=value lines in a private .credentials, to import into a credentials environment
771
+ holocron auth export --only HOLOCRON_SENTRY_TOKEN # one token, to the clipboard
772
+ holocron auth export --blank --vendor-names --file --only VERCEL_TOKEN --only HOLOCRON_RELEASE_TOKEN # the stored ones filled, the rest empty
773
+ holocron github-app create --name "My App - dev" --webhook-url https://<tunnel>/api/webhook --manifest apps/sentinel/github-app.manifest.json # a GitHub App from a manifest (the flow is in @theholocron/dagobah); its id, key and webhook secret to a private .credentials
774
+ holocron tunnel up --name sentinel-dev --hostname sentinel-dev.example.dev --port 3000 --run # a named tunnel with a stable hostname to a local port
775
+ holocron project list # the org's project boards (also: create [template], close <number-or-name>, setup)
776
+ holocron migrate project v5 --from github --to github:repo=acme/new --dry-run # preview putting a board's issues on a board of the same name in the target (GitHub; --template picks the template a missing board is made from, default roadmap); a target without boards (Linear) gets a project of that name instead
777
+ holocron project move v4 v5 --dry-run # preview carrying the open items of one project board to another; without --dry-run it asks first (--yes skips)
778
+ holocron doctor project # does each declared board match its template? (read-only; --post shares the failures)
779
+ holocron dev # the app, its vault variables and the tunnel, in one command (settings: `dev` in holocron.config)
416
780
  ```
417
781
 
418
782
  ## Logging
419
783
 
420
784
  Operational output goes through `@theholocron/observability/logger` — separate
421
- from the user-facing `print` surface. Global flags:
785
+ from the user-facing `print` surface. Local/interactive runs are quiet by
786
+ default: no colorized or raw structured console output, regardless of the
787
+ resolved level (Axiom shipping, when configured, is unaffected). CI is
788
+ untouched — its existing stdout output still appears. Global flags:
422
789
 
423
790
  ```sh
424
- holocron doctor --verbose # log level → debug (full structured output)
425
- holocron doctor --quiet # log level → error (suppress info + warn)
791
+ holocron doctor --log debug # show console output at this level (opt back in)
792
+ holocron doctor --verbose # log level → debug, and shows console output
793
+ holocron doctor --quiet # log level → error, and shows console output
426
794
  holocron doctor --debug # print "Run ID: <uuid>" at command end for Axiom lookup
427
795
  ```
428
796
 
429
- Level resolution (highest priority first): `--verbose` / `--quiet` →
430
- `HOLOCRON_LOG_LEVEL` → `log.level` in `holocron.config` → `"info"`.
797
+ Level resolution (highest priority first): `--log <level>` → `--verbose` /
798
+ `--quiet` → `HOLOCRON_LOG_LEVEL` → `log.level` in `holocron.config` →
799
+ `"info"`.
431
800
 
432
801
  ```ts
433
802
  export default defineConfig({
@@ -438,8 +807,18 @@ export default defineConfig({
438
807
  Axiom shipping activates when a token (`HOLOCRON_AXIOM_TOKEN` / `AXIOM_TOKEN`
439
808
  → OS keyring `axiom.<org>` → `axiom`) **and** a dataset
440
809
  (`HOLOCRON_AXIOM_DATASET` / `AXIOM_DATASET` → `log.axiom.dataset` in
441
- `holocron.config`) both resolve; env vars win. The token is never read from a
442
- config file. See the [logging guide](https://docs.theholocron.dev/holocron/logging/).
810
+ `holocron.config` → the credential chain, where the 1Password-synced
811
+ `AXIOM_DATASET` lives) both resolve; env vars win. The token is never read from a
812
+ config file.
813
+
814
+ Each service has one dataset for all its environments (ADR-0019; the CLI's is `holocron`); each line carries `env`
815
+ (`local`, `development`, `ci`, `test`, `production`), so a query filters by it.
816
+ `log.axiom.dataset` is only a per-repo override. The dataset side of the
817
+ resolution is **execution-context-aware** (ADR-0012): a `global`-context command
818
+ (`clone`, `new`, `auth set`, `upgrade node`, `plugin create`, …) is not "about"
819
+ any repo, so it skips the repo's `log.axiom.dataset` and uses the CLI's dataset
820
+ from the environment or credential chain. `repo-aware` / `workspace` commands
821
+ (`doctor`, `deploy`, `sync`, …) keep the full chain above. See the [logging guide](https://docs.theholocron.dev/holocron/logging/).
443
822
 
444
823
  ### Telemetry
445
824
 
@@ -468,38 +847,80 @@ export default defineConfig({
468
847
 
469
848
  Each SDK sits behind an interface from `@theholocron/observability` —
470
849
  `ErrorSink` (Sentry) / `AnalyticsSink` (PostHog); the adapters, and the only
471
- `@sentry/node` / `posthog-node` imports, live in that package. `telemetry.ts`
472
- holds the orchestration + the Holocron-specific credential resolution
473
- (`telemetry/resolve.ts`). See the
850
+ `@sentry/node` / `posthog-node` imports, live in that package. the orchestration is `createCliTelemetry` in `@theholocron/kyber/observability`;
851
+ `observability/telemetry.ts` binds it to the Holocron-specific credential resolution
852
+ (`observability/resolve-telemetry.ts`). See the
474
853
  [telemetry guide](https://docs.theholocron.dev/holocron/telemetry/).
475
854
 
855
+ ### ESLint bundle options
856
+
857
+ Most repos need nothing here — `@theholocron/eslint-config`'s `library()`
858
+ bundle resolves with zero committed file (Bucket A: `eslint --config
859
+ <shared path>`, no per-repo `eslint.config.ts` needed). A package with a
860
+ genuine, deliberate exception — e.g. Web Crypto globals a Node-targeted
861
+ package still needs, exempted via `library()`'s own `browserPackages`
862
+ option in that package's own `eslint.config.ts` — should also declare the
863
+ same paths in `holocron.config`'s `eslint.browserPackages`:
864
+
865
+ ```ts
866
+ export default defineConfig({
867
+ eslint: {
868
+ browserPackages: ["clients/github-client/src/app-auth"],
869
+ },
870
+ });
871
+ ```
872
+
873
+ This doesn't replace the package's own `eslint.config.ts` (local/CI linting
874
+ still resolves that file directly) — it's the signal Sentinel's centralized
875
+ static-analysis check (holocron#849/#858) reads instead, since that check
876
+ never reads a PR's own committed config files. Without it, Sentinel flags a
877
+ false-positive `eslint-plugin-n` node-builtins finding on exactly the files
878
+ the local override exists to exempt.
879
+
476
880
  ## What's in here
477
881
 
478
- - `src/capabilities/` — the 14 capability interfaces that providers
479
- implement
882
+ - `src/plugin/capabilities.ts` — the 21 capability interfaces that
883
+ providers implement
480
884
  - `src/config.ts` — config schema, `defineConfig`, `resolveConfig`,
481
885
  `CapabilityConfigPackage`
482
- - `src/config/load-config.ts` — `loadConfig` — reads `holocron.config.*`
483
- (file discovery via [`@theholocron/datapad`](../datapad))
484
- - `src/define-config.ts` — `defineConfig` typed pass-through
485
- - `src/logger.ts` — CLI-side `@theholocron/observability/logger` wiring
886
+ - `src/config/config-system.ts` — `defineConfig`, `loadConfig` (reads `holocron.config.*`;
887
+ file discovery via [`@theholocron/datapad`](../datapad)) and `compose`, all built from
888
+ kyber's `createConfigSystem`; Holocron's own rules sit beside it in `config/config`,
889
+ `compose/{merge-*,task-name}` and `load-config/{derive-defaults,read-package-json-name}`
890
+ - `src/observability/logger.ts` — CLI-side `@theholocron/observability/logger` wiring
486
891
  (`buildCliLogger`, `resolveLogLevel`)
487
- - `src/loader.ts` — `PluginLoader` — dynamic-imports plugins, resolves
488
- capability config packages, builds the capability registry
892
+ - `src/plugin/loader/loader.ts` — `PluginLoader`, a binding of kyber's `createPluginSystem`
893
+ (`@theholocron/kyber/plugins`: dynamic-imports plugins, resolves capability config packages,
894
+ builds the capability registry); Holocron's prefix, plugin options and the credentials / `.env`
895
+ step are passed in
489
896
  - `src/cli.ts` — yargs entry, dispatches subcommands. `holocron run` / `ci` are
490
897
  delegated to [`@theholocron/astromech`](../astromech) (`createAstromech`).
491
- `holocron sync` / `holocron setup` write each repo's `.github/workflows/*.yml`
492
- thin callers from `astromech.thinCallers()`; `holocron sync-github` pushes the
898
+ `holocron repo sync` / `holocron repo setup` write each repo's `.github/workflows/*.yml`
899
+ thin callers from `astromech.thinCallers()`; `holocron sync github` pushes the
493
900
  reusable `workflow_call` implementations + composite actions from
494
- `astromech.reusableTemplates()` to `theholocron/.github` (a pure sync target —
901
+ `astromech.reusableTemplates()` (written in `ci/`) to `theholocron/.github` (a pure sync target —
495
902
  never hand-edit its `.github/workflows/*`).
496
- - `src/commands/` — `setup`, `sync`, `doctor`, `deploy`, `secret set`,
497
- `secrets sync`, `publish`, `bump-versions`, `sync-github`, `upgrade node`,
498
- `upgrade deps`, `plugin create`, `auth`
499
- - `src/interactive-menu.ts` — the interactive fallback: `COMMAND_REGISTRY`,
500
- `launchMenu` (Layers 1–2, picks + spawns a fresh `holocron <command>`),
501
- `promptForPositionals` (Layer 3, called inline from each leaf command's
502
- `cli.ts` handler)
903
+ - `src/commands/` — `repo setup`, `repo sync`, `doctor`, `deploy`,
904
+ `monitoring check`, `secrets list`, `secrets set`, `secrets sync`, `package publish`,
905
+ `package bump-versions`, `package move-tag`, `sync github`, `repo upgrade node`,
906
+ `repo upgrade deps`, `new plugin`, `auth`, `notifications setup`,
907
+ `notifications channel add`, `notifications link`,
908
+ `notifications sync-roles`, `doctor ci`, `doctor credentials`, `doctor dns`, `doctor environments`, `doctor monitoring`, `doctor notifications`, `doctor project`, `doctor repo`, `doctor secrets`, `doctor tokens`, `doctor variables`
909
+ `notifications sync-roles`, `project reconcile`, `doctor ci`, `doctor credentials`, `doctor dns`, `doctor environments`, `doctor monitoring`, `doctor notifications`, `doctor repo`, `doctor secrets`, `doctor tokens`, `doctor variables`
910
+ - `src/commands/deep-checks/` — `DEEP_CHECKS`, the registry every `doctor <capability>` is
911
+ listed in; the shared row renderer and `declaredVsActual` helper are in `@theholocron/kyber/ui`
912
+ - `src/interactive-menu/` — the interactive fallback's Holocron data: `command-catalog`
913
+ (`COMMAND_REGISTRY`, built with kyber's `createCommandRegistry`; each entry's `positionals`/`options`), `run-menu-entries`
914
+ and `probe-menu-entries` (menu entries built at run time), and `menu.ts`,
915
+ kyber's `createMenu` bound to the `holocron` bin and `HOLOCRON_` env prefix.
916
+ The engine itself (the picker, argv and secret env, the dry-run preview, the spawn)
917
+ and the prompting pieces — `promptForPositionals`, `promptForOptions` — live in
918
+ `@theholocron/kyber` (`/menu`, `/prompts`), which has no knowledge of commands.
919
+ - `src/plugin/connection-probe/` — the `connectionProbe` a plugin offers, kyber's `connectionProbe` (`@theholocron/kyber/checks`)
920
+ typed as the plugin system's `Probe` with `verify` returning `VerifyTokenResult`; `monitoring check` runs its checks
921
+ with kyber's `runCheckAgainstUrl`.
922
+ - `src/commands/move-tag/` — the `package move-tag` yargs command, the registry entry and `runMoveTag`, which adds the
923
+ command-bound logger to kyber's `moveMajorTag` (`@theholocron/kyber/git`).
503
924
 
504
925
  ## Status
505
926