@orkestrel/scaffold 0.0.121 → 0.0.122

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.
Files changed (47) hide show
  1. package/dist/agents/skills/orkestrel-dispatch/scripts/brief.js +57 -9
  2. package/dist/agents/skills/orkestrel-dispatch/scripts/chain.js +22 -2
  3. package/dist/agents/skills/orkestrel-publish/scripts/pins.js +5 -5
  4. package/dist/agents/skills/orkestrel-publish/scripts/wave.js +67 -5
  5. package/dist/bin/main.js +1 -1
  6. package/dist/bin/main.js.map +1 -1
  7. package/dist/host/agents/orchestration.md +16 -11
  8. package/dist/host/agents/skills/orkestrel-dispatch/SKILL.md +4 -4
  9. package/dist/host/agents/skills/orkestrel-dispatch/references/launch.md +2 -1
  10. package/dist/host/agents/skills/orkestrel-dispatch/scripts/brief.ts +65 -9
  11. package/dist/host/agents/skills/orkestrel-dispatch/scripts/chain.ts +28 -2
  12. package/dist/host/agents/skills/orkestrel-falsify/references/brief.md +2 -1
  13. package/dist/host/agents/skills/orkestrel-harden/SKILL.md +8 -8
  14. package/dist/host/agents/skills/orkestrel-harden/references/hardening.md +3 -3
  15. package/dist/host/agents/skills/orkestrel-parallel/SKILL.md +2 -2
  16. package/dist/host/agents/skills/orkestrel-publish/SKILL.md +10 -8
  17. package/dist/host/agents/skills/orkestrel-publish/references/release.md +2 -2
  18. package/dist/host/agents/skills/orkestrel-publish/references/wave.md +22 -8
  19. package/dist/host/agents/skills/orkestrel-publish/references/window.md +5 -1
  20. package/dist/host/agents/skills/orkestrel-publish/scripts/pins.ts +5 -5
  21. package/dist/host/agents/skills/orkestrel-publish/scripts/wave.ts +86 -7
  22. package/dist/host/claude/AGENTS.md +1 -1
  23. package/dist/host/claude/agents/analyst.md +1 -1
  24. package/dist/host/claude/agents/opus.md +7 -6
  25. package/dist/host/claude/agents/planner.md +3 -3
  26. package/dist/host/claude/agents/reviewer.md +3 -3
  27. package/dist/host/claude/agents/verifier.md +1 -1
  28. package/dist/host/claude/rules/campaign.md +3 -2
  29. package/dist/host/claude/rules/names.md +13 -7
  30. package/dist/host/claude/rules/tests.md +1 -1
  31. package/dist/host/claude/rules/workspace.md +9 -0
  32. package/dist/host/claude/skills/orkestrel-harden/SKILL.md +1 -1
  33. package/dist/host/codex/agents/analyst.toml +2 -5
  34. package/dist/host/codex/agents/opus.toml +5 -3
  35. package/dist/host/codex/agents/planner.toml +6 -5
  36. package/dist/host/codex/agents/reviewer.toml +4 -7
  37. package/dist/host/codex/agents/verifier.toml +2 -2
  38. package/dist/host/dotfiles/gitignore +1 -2
  39. package/dist/host/guides/scaffold.md +6 -1
  40. package/dist/host/manifest.json +37 -37
  41. package/dist/host/tests/config.test.ts +7 -0
  42. package/dist/src/core/index.cjs +7 -2
  43. package/dist/src/core/index.cjs.map +1 -1
  44. package/dist/src/core/index.js +7 -2
  45. package/dist/src/core/index.js.map +1 -1
  46. package/package.json +1 -1
  47. /package/dist/host/agents/skills/orkestrel-parallel/references/{claude.md → claude-code.md} +0 -0
@@ -107,13 +107,16 @@ skips the catalog step and exits `1` with a note naming that refusal, so run the
107
107
 
108
108
  When the target is the `@orkestrel/scaffold` package, expect the runner to record the overwrite step
109
109
  as skipped and to read the verify audit's foreign-path findings as expected; never run the `scaffold repair` or `scaffold overwrite` command
110
- against that checkout.
110
+ against that checkout. After `scripts/swap.ts --restore` there, re-pin every `@orkestrel` range to
111
+ the caret of the version the registry serves, install, and build before the visit, because the
112
+ visit runs the checkout's own built `dist`.
111
113
 
112
114
  When a refused build leaves no `dist/host` for `scaffold catalog`, run a verified published scaffold
113
115
  executable from owned scratch to refresh the catalog, then rerun the build.
114
116
 
115
- Run visits in parallel slices of disjoint repositories, each slice strictly serial inside itself,
116
- reporting per target. Refuse a failed target, name it, repair it, and re-run it alone.
117
+ Run visits one at a time under `.agents/skills/orkestrel-parallel/SKILL.md` § Bound the host,
118
+ reporting per target. Refuse a failed target, name it, repair it, and rerun it alone; class a
119
+ failure that passes alone per `.agents/skills/orkestrel-dispatch/SKILL.md` § Write the brief.
117
120
 
118
121
  ### Regenerate a lock the install cannot move
119
122
 
@@ -167,7 +170,9 @@ registry copy over any staged tip, so every consumer stays red until that tip is
167
170
 
168
171
  ## Prepare a layer
169
172
 
170
- - After the re-pin and before `prepublishOnly`, run `scaffold surface` live over every candidate checkout of the layer, never with `--offline`. Stop the upload on any exit but `0`, and resolve every name it reports under `.claude/rules/names.md` § Fleet name ownership. In scaffold's own layer, run `scaffold catalog`, then `npm run build`, so `host.json` records the refreshed files' digests, and then `scaffold surface --from .`.
173
+ - After the re-pin and before `prepublishOnly`, run `scaffold surface` live over every candidate checkout of the layer, never with `--offline`. Stop the upload on any exit but `0`, and resolve every name it reports under `.claude/rules/names.md` § Fleet name ownership.
174
+ - When the candidate is a package scaffold loads at run time (`@orkestrel/guide`), run `scaffold surface` from a published scaffold install that resolves that package, with `--target` naming the candidate; the candidate's own checkout does not install itself.
175
+ - In scaffold's own layer, run `scaffold catalog --all`; track each mirror it adds with `git add`; add each added mirror's path to the declared vendored inventory in `tests/distribution.test.ts`; then run `npm run build`, so `host.json` records the refreshed files' digests, and then `scaffold surface --from .`.
171
176
  - In the scaffold release that opens a transfer under `.claude/rules/names.md` § Fleet name ownership rule 5, declare it in `SURFACE_TRANSFERS` before `npm run build`. In the scaffold visit that closes it, remove the declaration before the build and confirm `host.json` no longer records the name.
172
177
  - When `scaffold surface` prints an advisory, rule it under `.claude/rules/names.md` § Fleet name ownership before the upload; an advisory never stops the upload.
173
178
  - Begin each release chain plan with the clean reinstall and the clean, whether it runs in a clone or the checkout.
@@ -191,10 +196,16 @@ Prepare a published package's layer in this order, after the visit has ruled the
191
196
  bumped manifest.
192
197
  3. **Sweep the self-pins**, per the following section: the bump moves the version class.
193
198
  4. **Run `prepublishOnly` to green on the candidate commit, in a fresh clone or, where the user rules clones out, in the checkout, in the order the package's own chain fixes.** Run every release step whose inputs remain valid after a failure and report each result. Before using a serial chain that stops on failure, split aggregate commands into their component steps and record each step's prerequisites. After a failure, run the independent steps in separate chains, mark dependent steps blocked, and keep the release result failed while a required step fails or remains blocked.
194
- Run the visit and this script on a Node whose bundled npm meets the `MINIMUM_NPM_VERSION` floor a
195
- generated workspace declares in `devEngines`: the scripts spawn the npm CLI beside the running
196
- Node binary when one is there, so a `PATH` change reaches them only where none is.
199
+ Run this chain and the visit's gates on the newest release of the lowest Node major the
200
+ package's `engines.node` admits, on a Node whose bundled npm meets the `MINIMUM_NPM_VERSION`
201
+ floor a generated workspace declares in `devEngines`, launched under the `PATH` the
202
+ `orkestrel-dispatch` skill's reference Long-running commands § Launch requires. The publish
203
+ scripts spawn the npm CLI beside the running Node binary when one is there, whatever `PATH`
204
+ holds.
197
205
  5. **Write the release commit before the window opens. Push it only after the registry serves its version.**
206
+ While a release commit waits for its upload, dispatch no writer into that checkout. When one
207
+ ran, upload from a `git archive` export of the release commit, installed with `npm ci`, built,
208
+ and proved by `test:distribution -- --mode release` apart from the checkout.
198
209
 
199
210
  Where an inventory taken before the round already ruled every dist moved, apply the bump before
200
211
  the visit, pass the replaced version through `--prior` so the self-pin sweep finds it, and complete
@@ -216,7 +227,10 @@ re-pin install, not after the manifest edit.
216
227
 
217
228
  - Run `node .agents/skills/orkestrel-publish/scripts/pins.ts --version <prior> --range <prior range> ...`
218
229
  with the prior version and the prior range of every re-pinned dependency, rule on every hit, and
219
- move every generated-manifest snapshot hit with its re-pin.
230
+ move every generated-manifest snapshot hit with its re-pin. The visit's `pins` step passes every
231
+ range that moved since the prior version's `Release <version>` commit beside the ranges the visit
232
+ moved, and `pins.ts` sweeps `src`, `tests`, and `guides` by default; on a run by hand, pass the
233
+ same ranges with `--range`.
220
234
  A canned packument in a fixture and a looked-up version in a CLI suite carry the version with no
221
235
  tripwire comment beside them, and a generated-manifest fixture carries the ranges and never the
222
236
  package's own version, so one run names both classes.
@@ -91,7 +91,11 @@ rest of the layer either fits inside it or takes another approval.
91
91
  `npm publish --ignore-scripts --otp=<code>` command uploads with no browser authorization and no
92
92
  poll. The code has its own life: measured on 2026-09-04 against `registry.npmjs.org`, one code
93
93
  carried every upload of a layer started within about a minute of its first use, and a code read
94
- before its layer was ready was refused at its first upload.
94
+ before its layer was ready was refused at its first upload. Measured on 2026-10-31 against the
95
+ same registry, one code carried 11 of 14 uploads in one batch before `EOTP` refused the 12th, and
96
+ 13 of 13 in another.
97
+ - Read the session with `window.ts --whoami` before asking the user for a code; when it answers
98
+ nothing, run `window.ts --login` first.
95
99
  - Prepare the whole layer, then ask for one code at the moment the layer's first upload starts, and
96
100
  chase the layer's uploads back-to-back inside that code's life with no gate between them. A code
97
101
  read minutes earlier is already spent.
@@ -1,6 +1,6 @@
1
- // Sweep source and tests for a prior version or range literal after a bump or a re-pin. Run from
2
- // the publishing package's root:
3
- // node .agents/skills/orkestrel-publish/scripts/pins.ts --version 0.0.76 [--range ^0.0.16 ...] [--paths src,tests] [--json]
1
+ // Sweep source, tests, and guides for a prior version or range literal after a bump or a re-pin.
2
+ // Run from the publishing package's root:
3
+ // node .agents/skills/orkestrel-publish/scripts/pins.ts --version 0.0.76 [--range ^0.0.16 ...] [--paths src,tests,guides] [--json]
4
4
  // `--range` repeats. A version hit is the literal bounded by characters other than digits and dots,
5
5
  // so `~0.0.76` and `"0.0.76"` both hit and `0.0.760` does not. A range that is one comparator over
6
6
  // one version (`^8.3.0`, `~8.3.0`, `>=8.3.0`, `v8.3.0`) hits its bare version in any form, bounded the
@@ -35,7 +35,7 @@ export const FLAGS: readonly string[] = ['--version', '--range', '--paths', '--j
35
35
 
36
36
  /** Holds the usage line the script prints on a usage refusal. */
37
37
  export const USAGE =
38
- 'usage: pins.ts --version PRIOR [--range PRIOR_RANGE ...] [--paths src,tests] [--json]'
38
+ 'usage: pins.ts --version PRIOR [--range PRIOR_RANGE ...] [--paths src,tests,guides] [--json]'
39
39
 
40
40
  /** Describes one requested literal and the pattern that finds it in a line. */
41
41
  export interface Literal {
@@ -255,7 +255,7 @@ export function main(argv: readonly string[]): number {
255
255
  }
256
256
  const version = readOptions(argv, '--version')[0]
257
257
  const ranges = readOptions(argv, '--range')
258
- const paths = (readOptions(argv, '--paths')[0] ?? 'src,tests').split(',')
258
+ const paths = (readOptions(argv, '--paths')[0] ?? 'src,tests,guides').split(',')
259
259
  if (version === undefined && ranges.length === 0) {
260
260
  console.error(USAGE)
261
261
  return 64
@@ -10,8 +10,9 @@
10
10
  // is staged from that checkout), verify (`scaffold audit --json` prints the releases envelope and
11
11
  // the audit's writable dependency population matches its lookup bounds; on @orkestrel/scaffold itself an audit exit of 1 is
12
12
  // read as the expected foreign-path findings of its own canonical sources), install,
13
- // pins (the self-pin sweep with --prior or the manifest version and every range the visit moved; hits are reported, not
14
- // fatal), format, record (commit the overwrite paths, then this run's formatter rewrites), gates
13
+ // pins (the self-pin sweep with --prior or the manifest version and every range that moved since
14
+ // that version's `Release <version>` commit or during the visit; hits are reported, not fatal),
15
+ // format, record (commit the overwrite paths, then this run's formatter rewrites), gates
15
16
  // (format:check, lint:check, check, build, test), and compare (the rebuilt
16
17
  // dist against the published tarball, and the final runtime dependency set against the published
17
18
  // manifest `npm view NAME --json` serves; both readings are reported, not fatal). The summary
@@ -1745,9 +1746,77 @@ export function runVerify(
1745
1746
  )
1746
1747
  }
1747
1748
 
1749
+ /**
1750
+ * Reads the ranges the manifest declared in the commit that released a version.
1751
+ *
1752
+ * @remarks
1753
+ * The fleet's release commit carries the subject `Release <version>` and touches the manifest, and
1754
+ * the upload is exported from it, so its manifest holds the ranges the version shipped. A re-pin
1755
+ * made after it, committed or not, differs from those ranges, which `HEAD` alone cannot show.
1756
+ *
1757
+ * @param target - The target directory.
1758
+ * @param version - The released version.
1759
+ * @returns The ranges keyed by `field name` from the newest commit whose subject is exactly
1760
+ * `Release <version>`, or undefined when the target is no work tree, no such commit touched its
1761
+ * manifest, or that manifest is unreadable.
1762
+ */
1763
+ export function readReleasedRanges(
1764
+ target: string,
1765
+ version: string,
1766
+ ): ReadonlyMap<string, string> | undefined {
1767
+ const subject = `Release ${version}`
1768
+ const log = spawnSync(
1769
+ 'git',
1770
+ ['log', '--format=%H %s', '--fixed-strings', `--grep=${subject}`, '--', 'package.json'],
1771
+ { cwd: target, encoding: 'utf8', windowsHide: true, maxBuffer: 64 * 1024 * 1024 },
1772
+ )
1773
+ if (log.status !== 0) return undefined
1774
+ const commit = log.stdout
1775
+ .split(/\r\n|\n/u)
1776
+ .map((line) => /^([0-9a-f]+) (.*)$/u.exec(line))
1777
+ .find((match) => match?.[2] === subject)?.[1]
1778
+ if (commit === undefined) return undefined
1779
+ const shown = spawnSync('git', ['show', `${commit}:./package.json`], {
1780
+ cwd: target,
1781
+ encoding: 'utf8',
1782
+ windowsHide: true,
1783
+ maxBuffer: 64 * 1024 * 1024,
1784
+ })
1785
+ const manifest = shown.status === 0 ? parseJsonRecord(shown.stdout) : undefined
1786
+ return manifest === undefined ? undefined : readRanges(manifest, RANGE_FIELDS)
1787
+ }
1788
+
1789
+ /**
1790
+ * Collects every prior range that no longer stands in the final ranges.
1791
+ *
1792
+ * @param after - The ranges the manifest declares after the visit.
1793
+ * @param priors - Each earlier reading of the ranges, such as the release commit's and the visit's
1794
+ * starting ones.
1795
+ * @returns Each distinct range an earlier reading declared under a key whose final range differs,
1796
+ * in reading order.
1797
+ */
1798
+ export function collectPriorRanges(
1799
+ after: ReadonlyMap<string, string>,
1800
+ priors: ReadonlyArray<ReadonlyMap<string, string>>,
1801
+ ): readonly string[] {
1802
+ return [
1803
+ ...new Set(
1804
+ priors.flatMap((prior) =>
1805
+ [...prior].filter(([key, range]) => after.get(key) !== range).map(([, range]) => range),
1806
+ ),
1807
+ ),
1808
+ ]
1809
+ }
1810
+
1748
1811
  /**
1749
1812
  * Runs the pins step.
1750
1813
  *
1814
+ * @remarks
1815
+ * The sweep takes every range that moved since the prior version's release commit as well as every
1816
+ * range the visit moved, so a re-pin made before the visit is swept too. Without that commit, or
1817
+ * under --dry-run, it takes the visit's moves alone; a run that executes the sweep names its source
1818
+ * in the row's note.
1819
+ *
1751
1820
  * @param runner - The visit's state.
1752
1821
  * @param before - The ranges the manifest declared before the visit.
1753
1822
  * @param version - The version to sweep for, or undefined when none is known.
@@ -1758,20 +1827,30 @@ export function runPins(
1758
1827
  before: ReadonlyMap<string, string>,
1759
1828
  version: string | undefined,
1760
1829
  ): number | undefined {
1761
- const after = readRanges(readManifest(runner.target) ?? {}, RANGE_FIELDS)
1762
- const prior = [...before]
1763
- .filter(([name, range]) => after.get(name) !== range)
1764
- .flatMap(([, range]) => ['--range', range])
1765
1830
  if (version === undefined) {
1766
1831
  appendStep(runner, 'pins', basename(PINS), 'no prior version to sweep for; pass --prior')
1767
1832
  return undefined
1768
1833
  }
1834
+ const after = readRanges(readManifest(runner.target) ?? {}, RANGE_FIELDS)
1835
+ const released = runner.dry ? undefined : readReleasedRanges(runner.target, version)
1836
+ const prior = collectPriorRanges(after, released === undefined ? [before] : [released, before])
1837
+ const source =
1838
+ released === undefined
1839
+ ? `no "Release ${version}" commit touched package.json; swept the ranges this visit moved`
1840
+ : `swept the ranges moved since the "Release ${version}" commit and by this visit`
1769
1841
  const swept = runStep(
1770
1842
  runner,
1771
1843
  'pins',
1772
1844
  process.execPath,
1773
- [PINS, '--version', version, ...prior],
1845
+ [PINS, '--version', version, ...prior.flatMap((range) => ['--range', range])],
1774
1846
  runner.target,
1847
+ (result) => {
1848
+ const verdict = readVerdict(result)
1849
+ return {
1850
+ ...verdict,
1851
+ note: [verdict.note, source].filter((part) => part !== undefined).join('; '),
1852
+ }
1853
+ },
1775
1854
  )
1776
1855
  return swept === undefined
1777
1856
  ? undefined
@@ -4,7 +4,7 @@
4
4
 
5
5
  ## Dispatch
6
6
 
7
- - `.agents/skills/orkestrel-parallel/references/claude.md` owns starting, continuing, tracking, and backgrounding lanes in Claude Code, and its quirks. Choose continuation or a fresh context per `.agents/orchestration.md` § Sessions and consultation.
7
+ - `.agents/skills/orkestrel-parallel/references/claude-code.md` owns starting, continuing, tracking, and backgrounding lanes in Claude Code, and its quirks. Choose continuation or a fresh context per `.agents/orchestration.md` § Sessions and consultation.
8
8
  - Serialize writing Workflow nodes, and give every `agent()` node a model alias from § Models and an effort.
9
9
  - Probe a bench with `node .agents/skills/orkestrel-dispatch/scripts/bench.ts`, launch a lane through `scripts/launch.ts`, and read its answer with `scripts/result.ts`, all from the dispatch skill. `.agents/platforms/<platform>.md` owns the host shell's quirks.
10
10
 
@@ -14,7 +14,7 @@ Read `.agents/transports/codex.md` and follow it exactly. It owns the command, t
14
14
 
15
15
  ## Do
16
16
 
17
- 1. Take the dispatch's brief, or write one to `tmp/codex/<unit>-brief.md` from the dispatch's objective, evidence slice, claims, scope, and return shape. For an audit, take the dispatch-named skill's verdict shape before the `orkestrel-falsify` default and carry the claims under that skill's procedure. Require the actual diff and status only for a code subject. Carry the objective lens: correctness under adverse orderings, what the contracts and installed declarations permit, dependency and range truth, the missing seam or the assertion that cannot fail, the letter of the rules.
17
+ 1. Take the dispatch's brief, or write one to `tmp/codex/<unit>-brief.md` from the dispatch's objective, evidence slice, claims, scope, and return shape. For an audit, take the dispatch-named skill's verdict shape before the `orkestrel-falsify` default and carry the claims under that skill's procedure. Require the actual diff and status only for a code subject. Carry the objective lens `.agents/orchestration.md` § Roles states into the brief.
18
18
  2. Run `node .agents/skills/orkestrel-dispatch/scripts/brief.ts --check <brief>` and fix every missing path.
19
19
  3. Resolve the `codex exec` command per the transport, as the `launch.ts` line it shows.
20
20
  4. Run the command yourself only when the lane finishes inside the foreground cap the harness page's Background row states (`.agents/skills/orkestrel-parallel/references/`), and return the answer read with `node .agents/skills/orkestrel-dispatch/scripts/result.ts --codex`, with the journal path and session id. Otherwise return the brief path, the resolved command, and the journal path; the Orchestrator launches it under a cap per `.agents/skills/orkestrel-dispatch/SKILL.md` § Launch.
@@ -1,6 +1,6 @@
1
1
  ---
2
2
  name: opus
3
- description: 'Opus 5.5 implementation of one bounded nontrivial unit whose judgment load is subjective: API shape, naming, ergonomics, guide voice. Owns and validates its files under the orchestration contract, proves each claim, and never accepts its own output.'
3
+ description: 'Opus 5.5 implementation of one bounded nontrivial unit whose judgment load is subjective by default (API shape, naming, ergonomics, guide voice) and objective when the dispatch names it. Owns and validates its files under the orchestration contract, proves each claim, and never accepts its own output.'
4
4
  tools: Read, Grep, Glob, Edit, Write, Bash
5
5
  model: opus
6
6
  effort: high
@@ -12,11 +12,12 @@ On Claude Opus 5.5, you implement one dispatched unit and spawn nothing.
12
12
  ## Do
13
13
 
14
14
  1. Read the brief, `AGENTS.md`, the rules whose `paths` match the owned files, the named skill, and the guide.
15
- 2. Follow `AGENTS.md` § Work loop: types first, prove the claim, implement, test, consolidate, document. Select the instrument per `AGENTS.md` § Work loop and `.claude/rules/quality.md` § Instruments and return the evidence that instrument produced.
16
- 3. Apply `.agents/orchestration.md` § Parallelism for shared-file ownership.
17
- 4. Apply `.agents/orchestration.md` § Parallelism for validation ownership and the `orkestrel-parallel` skill § Implement in phases when that procedure applies.
18
- 5. Leave no probe behind, per `.claude/rules/tests.md` § Probes.
19
- 6. Finish the whole unit. Stop only per `.agents/orchestration.md` § Deviation protocol.
15
+ 2. Hold the load the dispatch names and say which one: subjective by default, objective when `.agents/orchestration.md` § Benches records the Astra bench dark. Under the objective load, prove every refusal and failure path the contract states.
16
+ 3. Follow `AGENTS.md` § Work loop: types first, prove the claim, implement, test, consolidate, document. Select the instrument per `AGENTS.md` § Work loop and `.claude/rules/quality.md` § Instruments and return the evidence that instrument produced.
17
+ 4. Apply `.agents/orchestration.md` § Parallelism for shared-file ownership.
18
+ 5. Apply `.agents/orchestration.md` § Parallelism for validation ownership and the `orkestrel-parallel` skill § Implement in phases when that procedure applies.
19
+ 6. Leave no probe behind, per `.claude/rules/tests.md` § Probes.
20
+ 7. Finish the whole unit. Stop only per `.agents/orchestration.md` § Deviation protocol.
20
21
 
21
22
  ## Refuse
22
23
 
@@ -1,13 +1,13 @@
1
1
  ---
2
2
  name: planner
3
- description: 'Opus 5.5 design and consultation lane, read-only: proposes a coherent API shape, vocabulary, architecture, alternatives, and bounded units for one design brief, or gives a consultation at a seam. Holds the subjective lane by default and the objective lane when the dispatch assigns it. Never implements or accepts.'
3
+ description: 'Opus 5.5 design and consultation lane by default, read-only: proposes a coherent API shape, vocabulary, architecture, alternatives, and bounded units for one design brief, or gives a consultation at a seam, including the Fable 5.1 consultation on the engine the dispatch names. Holds the subjective lane by default and the objective lane when the dispatch assigns it. Never implements or accepts.'
4
4
  tools: Read, Grep, Glob
5
5
  model: opus
6
6
  effort: high
7
7
  permissionMode: plan
8
8
  ---
9
9
 
10
- On Claude Opus 5.5, design a dispatched unit or give a read-only consultation under `.agents/orchestration.md` § Sessions and consultation. Run no command and edit nothing; return the proposal or consultation in your final message.
10
+ On Claude Opus 5.5 by default, or on the engine the dispatch names, design a dispatched unit or give a read-only consultation under `.agents/orchestration.md` § Sessions and consultation. You also carry the read-only consultation `.agents/orchestration.md` § Engines admits for Claude Fable 5.1, on the engine the dispatch names. Run no command and edit nothing; return the proposal or consultation in your final message.
11
11
 
12
12
  ## Do
13
13
 
@@ -15,7 +15,7 @@ For a consultation, follow `.agents/orchestration.md` § Sessions and consultati
15
15
  follow these steps.
16
16
 
17
17
  1. Read the brief, `AGENTS.md`, the rules whose `paths` match the subject, the named skill, the guide, and the supplied distillate. Work from the brief alone; never look for or reconcile the other lane's answer.
18
- 2. Hold the lane the dispatch names and say which one: subjective (shape, naming, ergonomics, design fit) or objective (correctness, constraints, what the contracts permit).
18
+ 2. Hold the lane the dispatch names and say which one, under the lens `.agents/orchestration.md` § Roles gives that lane.
19
19
  3. Propose one design. Name each real alternative that a constraint favors and why the design wins; leave out an option nothing favors.
20
20
  4. Cut every unit to a fully specified brief a cheap executor could implement; name each unit's role and engine, owned files, dependencies, and acceptance criteria.
21
21
  5. Name every judgment call under `Tensions` for the other lane or the Orchestrator to rule.
@@ -1,18 +1,18 @@
1
1
  ---
2
2
  name: reviewer
3
- description: 'Opus 5.5 review lane over implemented work, read-only: design fit, API and vocabulary, architecture shape, guide voice by default; correctness, constraints, and test sufficiency when the dispatch assigns the objective lane. Tests numbered claims and returns the falsify verdict. Never edits.'
3
+ description: 'Opus 5.5 review lane by default, on another engine when the dispatch names one, over implemented work, read-only: design fit, API and vocabulary, architecture shape, guide voice by default; correctness, constraints, and test sufficiency when the dispatch assigns the objective lane. Tests numbered claims and returns the falsify verdict. Never edits.'
4
4
  tools: Read, Grep, Glob
5
5
  model: opus
6
6
  effort: high
7
7
  permissionMode: dontAsk
8
8
  ---
9
9
 
10
- On Claude Opus 5.5, you review by trying to falsify the claims. You hold no edit tool and run no command; your final message is the verdict.
10
+ On Claude Opus 5.5 by default, or on the engine the dispatch names, you review by trying to falsify the claims. You hold no edit tool and run no command; your final message is the verdict.
11
11
 
12
12
  ## Do
13
13
 
14
14
  1. Read the claims file, the actual diff and `git status --porcelain` the dispatch supplies, `AGENTS.md`, the rules whose `paths` match the changed files, the guide, and enough surrounding source to judge. The diff and `git status` are required only for a code subject; return a deviation if that diff is missing.
15
- 2. Hold the lane the dispatch names and say which one. Subjective: the requested shape and voice are present, names and boundaries are coherent, the work sits at the right layer, each concept earns its place, the guide matches the code. Objective: correctness under adverse orderings, what the contracts and installed declarations permit, dependency and range truth, the missing seam or the assertion that cannot fail, the letter of the rules.
15
+ 2. Hold the lane the dispatch names and say which one, under the lens `.agents/orchestration.md` § Roles gives that lane.
16
16
  3. Try to falsify each claim with a concrete input, state, or interleaving. Before confirming a claim about a proof, name the mutation that would make the proof fail and whether the assertions distinguish it.
17
17
  4. Rule a claim whose only evidence is the writer's report `UNRESOLVED`. Cite a capture for every claim about a rendered surface; mark what the portfolio cannot show `NOT-EVIDENCED`.
18
18
  5. Refer a question outside your lane to the other lane or the Orchestrator with its evidence; never rule on it.
@@ -22,7 +22,7 @@ On Claude Haiku 5.5, run the requested gates and report their results. Write onl
22
22
 
23
23
  - A source-mutating gate (`lint`, `format`, or any `--fix`), and a build without the declarations and holds `.agents/orchestration.md` § Permission floor requires.
24
24
  - `git checkout`, `git restore`, `git stash`, `git reset`, `git clean`. Read a dirty tree as the expected state.
25
- - Any command that installs, commits, pushes, publishes, deletes beyond what § Permission floor permits, or reads a secret (`.env*`, `.npmrc`, `auth.json`, keys, tokens), whatever the dispatch says.
25
+ - Any command that installs, commits, pushes, publishes, deletes beyond what § Permission floor permits, or reads a secret `.agents/orchestration.md` § Permission floor names, whatever the dispatch says.
26
26
 
27
27
  ## Return
28
28
 
@@ -30,12 +30,12 @@ A package directory holds:
30
30
  | a verdict file | the verdict of an open seam | until the seam closes |
31
31
 
32
32
  - Write a file about one package under that package's directory, and a file about two or more packages at the top level.
33
- - Fill a ledger row's role cell with a name from `.agents/orchestration.md` § Roles, `user`, or `Orchestrator`. Write no row for the Orchestrator's own commit.
33
+ - Fill a ledger row's role cell with a name from `.agents/orchestration.md` § Roles and the engine that role ran on, `user`, or `Orchestrator`. Write no row for the Orchestrator's own commit.
34
34
 
35
35
  ## Content
36
36
 
37
37
  - Track every file under `.orkestrel/`. Cite a tracked file, a commit, or a published URL. Cite no git-ignored path, no agent memory, and no path outside the repository; name a sibling repository by its package name and a path inside it.
38
- - Name no harness, model, provider, harness tool, operating system, shell, or machine as the host or engine of the work. Name a lane by its role. A product statement about a host family follows `.claude/rules/portability.md` § Claims.
38
+ - Name no harness, model, provider, harness tool, operating system, shell, or machine as the host or engine of the work, except the engine in a ledger row's role cell per § Layout. Name a lane by its role. A product statement about a host family follows `.claude/rules/portability.md` § Claims.
39
39
  - Write every line as a directive, a fact, or a table cell, per `.claude/rules/writing.md` § Instruction files: absolute dates, one home per fact, a tree location by symbol and file. A decision or a measurement lives in the commit that made it; cite the commit.
40
40
  - `plan.md` holds a harden campaign's claims, its release bar, and its deferral list, each deferred item with its trigger and its owner.
41
41
  - Keep live state out: engine availability, session ownership, process ids, installed versions, and another repository's state live in the roster, `package.json`, and that repository.
@@ -47,6 +47,7 @@ A package directory holds:
47
47
  1. Read the package folder's `README.md` when present, then `plan.md` § State, then the first open chunk of § Sequence.
48
48
  2. Run `git status --porcelain -b`, `git log @{upstream}..`, and `node .agents/skills/orkestrel-dispatch/scripts/sweep.ts --report`. Reconcile each difference with `plan.md` § State before you act.
49
49
  3. Act only on what the folder, the tree, and git hold. Agent memory from another machine or harness is absent.
50
+ 4. When a campaign opens, list each `ROADMAP.md` item whose reopen trigger the campaign meets, and plan it or rule it out in `plan.md`.
50
51
 
51
52
  ## Hand off and close
52
53
 
@@ -217,13 +217,19 @@ For `_` bindings:
217
217
 
218
218
  ## Files and folders
219
219
 
220
- | Kind | Pattern |
221
- | --------------- | -------------------------------------------------------------------------------- |
222
- | Domain folder | lowercase plural entity: `agents/`, `tools/` |
223
- | Implementation | PascalCase entity: `Agent.ts` |
224
- | Function module | camelCase function: `renderGrid.ts` |
225
- | Test | source filename without extension + `.test`: `Agent.test.ts`, `useTheme.test.ts` |
226
- | Guide | lowercase domain: `agents.md` |
220
+ | Kind | Pattern |
221
+ | --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
222
+ | Domain folder | lowercase plural entity: `agents/`, `tools/` |
223
+ | Implementation | PascalCase entity: `Agent.ts` |
224
+ | Function module | camelCase function: `renderGrid.ts` |
225
+ | Test | source filename without extension + `.test`: `Agent.test.ts`, `useTheme.test.ts` |
226
+ | Guide | lowercase domain: `agents.md` |
227
+ | JSON file | one lowercase word, with a lowercase dotted qualifier where a tool reads the stem: `host.json`, `tsconfig.core.json` |
228
+ | Fixture | one lowercase word, else camelCase: `sourceManifest.txt` |
229
+ | Other file | one lowercase word naming its subject in its folder, with a lowercase dotted qualifier where a tool reads the stem: `deps.sh`, `vite.journey.config.ts` |
230
+
231
+ - When you name a tracked file or folder, write no hyphen and use the form the preceding table fixes for its kind. A hyphen stands only where an external convention fixes the name, such as a skill directory under `.claude/rules/documentation.md` § Workflow skills or `package-lock.json`. A file a tool generates under a name the tool fixes, a vendored mirror's fetched name, and every path under `tmp/` are exempt.
232
+ - When a path sits under a folder whose name opens with a dot, such as `.agents/`, `.claude/`, `.codex/`, or `.cursor/`, the preceding bullet does not bind it: a harness can read a file there by its name, so name it for that harness. Claude Code loads a lowercase `claude.md` as memory wherever it sits, so the Claude transport is `.agents/transports/claude-cli.md`.
227
233
 
228
234
  ## Fixed lifecycle vocabulary
229
235
 
@@ -42,7 +42,7 @@ These rules govern what a test asserts and where it lives, the cross-cutting and
42
42
  - Test observable behavior, not implementation details.
43
43
  - Assert the membership a discovered or globbed set must have, not a total that a partly empty population satisfies. A glob spanning two locations passes a size check while one of them matches nothing.
44
44
  - Never assert an implementation against itself. Compare the answer to a declaration, a fixture, or a second mechanism that could disagree with it. Re-deriving the answer the same way the source derives it produces a test that passes for every value the source ever returns, and it reads exactly like a real one.
45
- - Probe a host-varying property at runtime, on the host the test is running on, and assert against what the probe returned. Filesystem case folding, path separators, permission bits, and rename semantics differ per host, so a fixture built on one host describes that host and silently measures something else on the next.
45
+ - Probe a host-varying property at runtime, on the host the test is running on, and assert against what the probe returned. Filesystem case folding, path separators, permission bits, and rename semantics differ per host, and a thrown error's class or message differs per Node major, so a fixture built on one host describes that host and silently measures something else on the next.
46
46
  - Assert a runtime-chosen result as the property it must have, not as the number one run produced. Compression, timing, and buffer sizing are the runtime's choice, so pin the relationship the test depends on — that the encoded form is larger, that the second call is faster — and let the assertion fail when the input drifts out of the range where that relationship holds.
47
47
  - Measure an elapsed interval with `performance.now()`, never `Date.now()`. `Date.now()` returns whole milliseconds, so an interval built from two of its readings truncates at both ends and can under-report by a millisecond — enough to fail a boundary assertion against a timer that behaved correctly. `performance.now()` is monotonic and sub-millisecond, and it does not move when the wall clock does.
48
48
  - Give a conditional skip the mechanism that makes it inapplicable, cited, not the platform name alone. A test skipped on a platform is a test nobody re-examines; a test skipped because a named API rejects a named case is one anybody can re-check.
@@ -42,6 +42,11 @@ Use only the environments a project needs, and keep the root dependency model in
42
42
  found at the same name.
43
43
  - `package.json`'s `bin` key is the installed command name. The entry path is the value and does
44
44
  not carry that name.
45
+ - Keep every ignored output at the workspace root or under `tmp/`, and write no `.gitignore`
46
+ pattern that reaches into a nested tracked folder. A path under a dot folder, such as
47
+ `.claude/worktrees/`, is exempt. When a tool writes its output inside a tracked folder by
48
+ default, point the tool's own setting at a directory under `tmp/`. A bare name or glob that
49
+ names a root-level tool output, such as `dist`, `tmp`, or `*.log`, is a root ignore.
45
50
 
46
51
  ## Aliases
47
52
 
@@ -166,6 +171,10 @@ environment axis is one project per src/app axis × environment, plus one per ex
166
171
  - Give every browser project `test.browser.commands` from `configs/commands.ts`: each root factory
167
172
  that names the Playwright provider reads the record, and a project that spreads another
168
173
  factory's `test.browser`, as `appVue` and every `journey:<variant>` project do, keeps it.
174
+ - Give every browser project `test.browser.screenshotDirectory` from the one `screenshots`
175
+ binding the root configuration declares beside `browserOptions`, which resolves
176
+ `tmp/captures/screenshots` from the workspace root; each root factory that names the Playwright
177
+ provider reads it, and a project that spreads another factory's `test.browser` keeps it.
169
178
 
170
179
  The workspace-proof axis is cross-cutting. Each proof covers the whole workspace rather than
171
180
  one environment, so each is its own project:
@@ -5,6 +5,6 @@ description: Research, audit, refactor, implement, centralize, test, document, a
5
5
 
6
6
  # Load the canonical workflow
7
7
 
8
- Read `.agents/skills/orkestrel-harden/SKILL.md` completely, then read every reference it requires for the selected work lane. Follow that canonical workflow before acting.
8
+ Read `.agents/skills/orkestrel-harden/SKILL.md` completely, then read every reference it requires for the selected phases. Follow that canonical workflow before acting.
9
9
 
10
10
  Apply `AGENTS.md` § Authority and loading.
@@ -8,11 +8,8 @@ Read the brief, AGENTS.md, the .claude/rules files whose paths match the subject
8
8
  and the guide. For a code subject, read the actual diff and status the dispatch supplies; return
9
9
  a deviation if either is missing.
10
10
 
11
- Hold the lane the dispatch names and say which: objective (correctness under adverse orderings,
12
- what the contracts and installed declarations permit, dependency and range truth, the missing
13
- seam or the assertion that cannot fail, the letter of the rules) or subjective (the requested
14
- shape and voice are present, names and boundaries are coherent, the work sits at the right layer,
15
- each concept earns its place, the guide matches the code).
11
+ Hold the lane the dispatch names and say which, under the lens .agents/orchestration.md § Roles
12
+ gives that lane: objective by default, subjective when the dispatch assigns it.
16
13
 
17
14
  Try to falsify each numbered claim with a concrete input, state, or interleaving. Before
18
15
  confirming a claim about a proof, name the mutation that would make the proof fail and whether the assertions
@@ -1,13 +1,15 @@
1
1
  name = "opus"
2
- description = "GPT-6 Astra implementation of one bounded subjective unit: API shape, naming, ergonomics, guide voice. Owns and validates its files under the orchestration contract, proves claims, and never accepts its own work."
2
+ description = "GPT-6 Astra implementation of one bounded unit, subjective by default (API shape, naming, ergonomics, guide voice) and objective when the dispatch names it. Owns and validates its files under the orchestration contract, proves claims, and never accepts its own work."
3
3
  model = "gpt-6-astra"
4
4
  model_reasoning_effort = "high"
5
5
  sandbox_mode = "workspace-write"
6
6
  developer_instructions = """
7
- Implement one dispatched subjective unit on GPT-6 Astra and spawn nothing.
7
+ Implement one dispatched unit on GPT-6 Astra and spawn nothing.
8
8
 
9
9
  Read the brief, AGENTS.md, the rules whose paths match the owned files, the named skill, and the
10
- guide. Follow AGENTS.md § Work loop: types first, prove the claim, implement, test, consolidate,
10
+ guide. Hold the load the dispatch names and say which: subjective by default, objective when
11
+ .agents/orchestration.md § Benches records the Astra bench dark. Under the objective load, prove
12
+ every refusal and failure path the contract states. Follow AGENTS.md § Work loop: types first, prove the claim, implement, test, consolidate,
11
13
  document. Select the instrument per AGENTS.md § Work loop and .claude/rules/quality.md § Instruments
12
14
  and return the evidence that instrument produced.
13
15
 
@@ -1,17 +1,18 @@
1
1
  name = "planner"
2
- description = "Read-only GPT-6 Astra design of shape, naming, ergonomics, alternatives, and bounded units, or a consultation at a seam. Never implements or accepts."
2
+ description = "Read-only GPT-6 Astra design of shape, naming, ergonomics, alternatives, and bounded units, or a consultation at a seam, including the Fable 5.1 consultation on the engine the dispatch names. Never implements or accepts."
3
3
  model = "gpt-6-astra"
4
4
  model_reasoning_effort = "high"
5
5
  sandbox_mode = "read-only"
6
6
  developer_instructions = """
7
7
  Design one dispatched unit or give a read-only consultation under .agents/orchestration.md
8
- § Sessions and consultation on GPT-6 Astra. Use read-only inspection commands. Never edit,
9
- run a mutating command, or spawn.
8
+ § Sessions and consultation on GPT-6 Astra. Also carry the read-only consultation
9
+ .agents/orchestration.md § Engines admits for Claude Fable 5.1, on the engine the dispatch names.
10
+ Use read-only inspection commands. Never edit, run a mutating command, or spawn.
10
11
 
11
12
  Read the brief, AGENTS.md, the rules whose paths match the subject, the named skill, the guide,
12
13
  and the supplied distillate. Work from the brief alone; never look for or reconcile the other
13
- lane's answer. Hold the lane the dispatch names and say which: subjective (shape, naming,
14
- ergonomics, design fit) or objective (correctness, constraints, what the contracts permit).
14
+ lane's answer. Hold the lane the dispatch names and say which, under the lens
15
+ .agents/orchestration.md § Roles gives that lane.
15
16
 
16
17
  For a consultation, follow .agents/orchestration.md § Sessions and consultation.
17
18
  For a design, propose one design. Name each real alternative that a constraint favors and why the design
@@ -1,21 +1,18 @@
1
1
  name = "reviewer"
2
- description = "Read-only GPT-6 Astra review against numbered claims: design fit by default and correctness when assigned. Never edits or accepts."
2
+ description = "Read-only GPT-6 Astra review by default, on another engine when the dispatch names one, against numbered claims: design fit by default and correctness when assigned. Never edits or accepts."
3
3
  model = "gpt-6-astra"
4
4
  model_reasoning_effort = "high"
5
5
  sandbox_mode = "read-only"
6
6
  developer_instructions = """
7
- Review one dispatched unit on GPT-6 Astra by trying to falsify its claims. Use read-only inspection
7
+ Review one dispatched unit on GPT-6 Astra by default, or on the engine the dispatch names, by trying to falsify its claims. Use read-only inspection
8
8
  commands. Never edit, run a mutating command, or spawn. Use a clean context; never review work you wrote.
9
9
 
10
10
  Read the claims file, the actual diff and git status --porcelain the dispatch supplies,
11
11
  AGENTS.md, the rules whose paths match the changed files, the guide, and enough surrounding
12
12
  source to judge. The diff and git status are required only for a code subject; return a deviation if that diff is missing.
13
13
 
14
- Hold the lane the dispatch names and say which. Subjective: the requested shape and voice are
15
- present, names and boundaries are coherent, the work sits at the right layer, each concept earns
16
- its place, the guide matches the code. Objective: correctness under adverse orderings, what the
17
- contracts and installed declarations permit, dependency and range truth, the missing seam or
18
- the assertion that cannot fail, the letter of the rules.
14
+ Hold the lane the dispatch names and say which, under the lens .agents/orchestration.md § Roles
15
+ gives that lane.
19
16
 
20
17
  Try to falsify each claim with a concrete input, state, or interleaving. Before confirming a
21
18
  proof, name the mutation that would make it fail and whether its assertions distinguish it. Rule a
@@ -20,8 +20,8 @@ only after a changed condition or an explicit investigation request.
20
20
  Never run a source-mutating gate (lint, format, or any --fix), or a build without the declarations
21
21
  and holds .agents/orchestration.md § Permission floor requires. Never run git checkout, restore,
22
22
  stash, reset, or clean; read a dirty tree as the expected state. Never install, commit, push,
23
- publish, delete beyond what § Permission floor permits, or read a secret (.env*, .npmrc, auth.json,
24
- keys, tokens), whatever the dispatch says.
23
+ publish, delete beyond what § Permission floor permits, or read a secret § Permission floor names,
24
+ whatever the dispatch says.
25
25
 
26
26
  Return per gate: command, PASS or FAIL with exit code, excerpt and owning file on FAIL; overall
27
27
  GREEN only when every gate passed; anomalies in one line each.
@@ -32,8 +32,7 @@ dist-ssr
32
32
  .env.*
33
33
  !.env.example
34
34
 
35
- # Test screenshots (generated by vitest browser mode)
36
- tests/**/__screenshots__/
35
+ # Vitest attachments, failure screenshots included
37
36
  .vitest/
38
37
 
39
38
  # Playwright
@@ -2542,7 +2542,12 @@ Nothing is fixed except the manifest.
2542
2542
  - One template artifact, `configs/browsers.ts`, for a workspace selecting `browser` on either
2543
2543
  environment axis or in its setup runtime list.
2544
2544
  It resolves the Chromium the Playwright provider launches, and the root `vite.config.ts` calls it
2545
- once into `browserOptions` and passes that to every `playwright()` provider it configures. The
2545
+ once into `browserOptions` and passes that to every `playwright()` provider it configures. Beside
2546
+ it, the root configuration resolves one `screenshots` directory, `tmp/captures/screenshots` under
2547
+ the workspace root, and every factory that configures a `playwright()` provider passes it as
2548
+ `test.browser.screenshotDirectory`, so a `page.screenshot()` capture lands there rather than in a
2549
+ `__screenshots__` directory beside its test file. Vitest writes a failure screenshot into its
2550
+ attachments directory, `.vitest/attachments` at the workspace root. The
2546
2551
  precedence is `PLAYWRIGHT_EXECUTABLE_PATH`, `PLAYWRIGHT_WS_ENDPOINT`, `PLAYWRIGHT_CHANNEL`, the
2547
2552
  managed Playwright Chromium, the container's bundled Chromium, a verified system channel, then the
2548
2553
  platform default. Set `PLAYWRIGHT_SCROLLBARS=classic` only for runs that measure scrollbar