@orkestrel/scaffold 0.0.122 → 0.0.123
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/dist/agents/skills/orkestrel-harden/scripts/interleave.js +539 -0
- package/dist/agents/skills/orkestrel-parallel/scripts/lane.js +1163 -0
- package/dist/bin/main.js +8 -6
- package/dist/bin/main.js.map +1 -1
- package/dist/host/agents/orchestration.md +13 -7
- package/dist/host/agents/skills/orkestrel-dispatch/SKILL.md +4 -3
- package/dist/host/agents/skills/orkestrel-dispatch/references/launch.md +1 -1
- package/dist/host/agents/skills/orkestrel-falsify/SKILL.md +4 -2
- package/dist/host/agents/skills/orkestrel-falsify/references/brief.md +2 -2
- package/dist/host/agents/skills/orkestrel-harden/SKILL.md +16 -3
- package/dist/host/agents/skills/orkestrel-harden/references/bindings.md +18 -12
- package/dist/host/agents/skills/orkestrel-harden/references/document.md +1 -0
- package/dist/host/agents/skills/orkestrel-harden/references/equivalence.md +2 -1
- package/dist/host/agents/skills/orkestrel-harden/references/hardening.md +1 -1
- package/dist/host/agents/skills/orkestrel-harden/references/methods.md +642 -0
- package/dist/host/agents/skills/orkestrel-harden/scripts/interleave.ts +665 -0
- package/dist/host/agents/skills/orkestrel-parallel/SKILL.md +14 -13
- package/dist/host/agents/skills/orkestrel-parallel/references/claude-code.md +13 -13
- package/dist/host/agents/skills/orkestrel-parallel/references/t3.md +1 -0
- package/dist/host/agents/skills/orkestrel-parallel/scripts/lane.ts +1433 -0
- package/dist/host/agents/skills/orkestrel-publish/SKILL.md +2 -1
- package/dist/host/agents/skills/orkestrel-publish/references/wave.md +18 -1
- package/dist/host/claude/AGENTS.md +1 -1
- package/dist/host/claude/agents/analyst.md +1 -1
- package/dist/host/claude/agents/builder.md +1 -1
- package/dist/host/claude/agents/distiller.md +2 -1
- package/dist/host/claude/agents/grok.md +2 -2
- package/dist/host/claude/agents/opus.md +2 -2
- package/dist/host/claude/agents/orkestrel.md +7 -7
- package/dist/host/claude/agents/reviewer.md +2 -2
- package/dist/host/claude/agents/scout.md +2 -1
- package/dist/host/claude/agents/verifier.md +1 -1
- package/dist/host/claude/rules/campaign.md +12 -11
- package/dist/host/claude/rules/documentation.md +1 -0
- package/dist/host/claude/rules/patterns.md +14 -0
- package/dist/host/claude/rules/quality.md +4 -1
- package/dist/host/claude/rules/tests.md +4 -0
- package/dist/host/claude/rules/writing.md +1 -1
- package/dist/host/claude/skills/orkestrel-parallel/SKILL.md +1 -1
- package/dist/host/codex/agents/analyst.toml +4 -1
- package/dist/host/codex/agents/astra.toml +1 -1
- package/dist/host/codex/agents/builder.toml +1 -1
- package/dist/host/codex/agents/distiller.toml +4 -2
- package/dist/host/codex/agents/grok.toml +5 -2
- package/dist/host/codex/agents/opus.toml +3 -3
- package/dist/host/codex/agents/reviewer.toml +6 -2
- package/dist/host/codex/agents/scout.toml +5 -1
- package/dist/host/codex/agents/verifier.toml +6 -1
- package/dist/host/guides/agent.md +1119 -457
- package/dist/host/guides/ndjson.md +81 -21
- package/dist/host/guides/ollama.md +22 -18
- package/dist/host/guides/probe.md +29 -27
- package/dist/host/guides/scaffold.md +36 -14
- package/dist/host/guides/study.md +538 -450
- package/dist/host/manifest.json +66 -48
- package/dist/src/core/index.cjs +8 -7
- package/dist/src/core/index.cjs.map +1 -1
- package/dist/src/core/index.d.cts +1 -4
- package/dist/src/core/index.d.ts +1 -4
- package/dist/src/core/index.js +8 -7
- package/dist/src/core/index.js.map +1 -1
- package/dist/src/server/index.cjs +87 -8
- package/dist/src/server/index.cjs.map +1 -1
- package/dist/src/server/index.d.cts +50 -6
- package/dist/src/server/index.d.ts +50 -6
- package/dist/src/server/index.js +87 -10
- package/dist/src/server/index.js.map +1 -1
- package/package.json +2 -2
|
@@ -56,7 +56,8 @@ The user's current instruction wins. `references/release.md` binds every release
|
|
|
56
56
|
§ Prepare a layer.
|
|
57
57
|
|
|
58
58
|
Run that sequence for every layer, from the registry reading to the registry close;
|
|
59
|
-
[wave.md](references/wave.md) § Refresh the registry between layers owns the refresh.
|
|
59
|
+
[wave.md](references/wave.md) § Refresh the registry between layers owns the refresh. After the
|
|
60
|
+
last layer of a fleet wave serves, close the wave per [wave.md](references/wave.md) § Close the wave.
|
|
60
61
|
|
|
61
62
|
## Accept the release
|
|
62
63
|
|
|
@@ -171,7 +171,7 @@ registry copy over any staged tip, so every consumer stays red until that tip is
|
|
|
171
171
|
## Prepare a layer
|
|
172
172
|
|
|
173
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
|
|
174
|
+
- When the candidate is a package scaffold loads at run time (`@orkestrel/guide`) and the scaffold its checkout installs exports `resolveSurfaceLoader` from `@orkestrel/scaffold/server`, run `npm run build` in that checkout, then run `scaffold surface` from it; the run loads the package through the checkout's own exports and refuses with `TARGET` while its build output is absent. When the installed scaffold lacks that export, run `scaffold surface` from a published scaffold install that resolves the package, with `--target` naming the candidate.
|
|
175
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 .`.
|
|
176
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.
|
|
177
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.
|
|
@@ -258,3 +258,20 @@ takes the self-pin sweep too, because the snapshot class moves with no bump.
|
|
|
258
258
|
an install in that gap fails `E404`.
|
|
259
259
|
- Run every visit and install that follows a publish with `npm_config_prefer_online=true`. A
|
|
260
260
|
cached packument answers `notarget` for a version the registry already serves.
|
|
261
|
+
|
|
262
|
+
## Close the wave
|
|
263
|
+
|
|
264
|
+
After the last layer serves, re-pin the development ranges the visits left behind. A visit re-pins
|
|
265
|
+
only what the registry serves when it runs, so a `devDependencies` range naming a package that a
|
|
266
|
+
later layer published still names the earlier release, and the install nests older copies of that
|
|
267
|
+
package's own dependencies.
|
|
268
|
+
|
|
269
|
+
1. Take a registry sweep after the last window confirms. In every catalog package, list each
|
|
270
|
+
`devDependencies` range on an `@orkestrel` package that trails the release the sweep reads.
|
|
271
|
+
2. In each listed package, re-pin those ranges to the caret of the served release, install, and
|
|
272
|
+
sweep the self-pins per § Sweep the self-pins, because the snapshot class moves with no bump.
|
|
273
|
+
3. Run the package's gates, then rule its bump under § Rule on the bump. Commit the re-pin of a
|
|
274
|
+
package the ruling leaves unbumped as `Re-pin <names> development ranges`; prepare and publish
|
|
275
|
+
each bumped package per § Prepare a layer, in layer order.
|
|
276
|
+
4. Take the sweep again after those publishes, because each release they add can leave another
|
|
277
|
+
package's development range trailing. Close the wave when a sweep lists no trailing range.
|
|
@@ -25,7 +25,7 @@
|
|
|
25
25
|
|
|
26
26
|
## Wiring
|
|
27
27
|
|
|
28
|
-
- `.mcp.json` registers `probe` from `node_modules/@orkestrel/probe/dist/bin/main.js
|
|
28
|
+
- In the `@orkestrel/scaffold` checkout, `.mcp.json` registers `probe` from `node_modules/@orkestrel/probe/dist/bin/main.js`; run `npm ci --ignore-scripts` before the first `prove` call in a fresh checkout. A target holds no `.mcp.json`, so register `probe` there per `.claude/rules/quality.md` § Instruments.
|
|
29
29
|
- `.claude/settings.json` hooks: `SessionStart` runs the Cloud setup hooks under `scripts/` and the dispatch skill's `sweep.ts --report`; `Stop` runs `git diff --check`. `scripts/` holds Cloud setup hooks and nothing else.
|
|
30
30
|
- `claude mcp serve` exposes this harness to a Codex- or Cursor-primary session.
|
|
31
31
|
- `.claude/skills/<name>/SKILL.md` loads `.agents/skills/<name>/SKILL.md` and adds nothing.
|
|
@@ -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 `.agents/orchestration.md` § Roles states into the brief.
|
|
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, and the receipts the dispatch supplies for each probe a claim needs, because this route runs none.
|
|
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.
|
|
@@ -22,7 +22,7 @@ On Claude Sonnet 5.5, you implement a fully specified unit with a closed shape.
|
|
|
22
22
|
|
|
23
23
|
- A unit that leaves naming, API shape, or architecture open, or that spans packages: report it as belonging to `astra` or `opus`.
|
|
24
24
|
- A unit whose mechanism is open: a protocol, a lock, a retry class, or a termination path; report it as belonging to `astra` or `opus`.
|
|
25
|
-
- Installing, committing, pushing, reading a secret, or a destructive command.
|
|
25
|
+
- Installing, committing outside a branch, clone, copy, or worktree the brief names, pushing, reading a secret, or a destructive command.
|
|
26
26
|
|
|
27
27
|
## Return
|
|
28
28
|
|
|
@@ -16,7 +16,8 @@ On Claude Haiku 5.5, you read and distill the supplied evidence. You decide noth
|
|
|
16
16
|
2. Read every named input in full. Record each fact with `file:line`.
|
|
17
17
|
3. Separate what the inputs state from what you infer; label inference.
|
|
18
18
|
4. Name every input row the distillate did not reach.
|
|
19
|
+
5. When the inputs outlast one turn, stop at a clean boundary and never sample. A partial read never counts as a whole read.
|
|
19
20
|
|
|
20
21
|
## Return
|
|
21
22
|
|
|
22
|
-
`Question`, `Evidence` (cited facts), `Contradictions` (each conflicting statement cited), `Distillate` (the smallest context the next engine needs), `Unknowns
|
|
23
|
+
`Question`, `Evidence` (cited facts), `Contradictions` (each conflicting statement cited), `Distillate` (the smallest context the next engine needs), `Unknowns`, `Coverage` (per input: its identity, the last line read without a gap, and the next slice). Nothing else, and no recommendation.
|
|
@@ -15,11 +15,11 @@ Read `.agents/transports/cursor.md` and follow it exactly. It owns the model pin
|
|
|
15
15
|
## Do
|
|
16
16
|
|
|
17
17
|
1. Require a bounded question and an exact file scope from the dispatch. Refuse an unbounded one.
|
|
18
|
-
2. Write the brief to `tmp/cursor/<unit>-brief.md`: read-only, the evidence sought, `file:line` pointers required, no raw dumps, no decisions.
|
|
18
|
+
2. Write the brief to `tmp/cursor/<unit>-brief.md`: read-only, the evidence sought, `file:line` pointers required, no raw dumps, no decisions, and a stop at a clean boundary with a coverage table when the inputs outlast one answer; a partial read never counts as a whole read.
|
|
19
19
|
3. Resolve the command per the transport: `node .agents/skills/orkestrel-dispatch/scripts/bench.ts --cursor --resolve` prints the entry, and the transport shows the `launch.ts` line.
|
|
20
20
|
4. Run the command yourself only as `.agents/transports/cursor.md` § Command allows, and return the answer read with `scripts/result.ts --cursor`. Otherwise return the brief path, the resolved command, and the journal path; the Orchestrator launches it under a cap.
|
|
21
21
|
5. Launch with `--status`, so `launch.ts` records `git status --porcelain` before and after. Any change is a deviation.
|
|
22
22
|
|
|
23
23
|
## Return
|
|
24
24
|
|
|
25
|
-
`Question` (one line), `Evidence` (cited facts), `Distillate` (the smallest context the next engine needs), `Unknowns`, `Journal` (path and session id from the `init` event), `Deviation` (dark bench, command failure, dirty containment). Nothing else.
|
|
25
|
+
`Question` (one line), `Evidence` (cited facts), `Distillate` (the smallest context the next engine needs), `Unknowns`, `Coverage` (per input: its identity, the last line read without a gap, and the next slice), `Journal` (path and session id from the `init` event), `Deviation` (dark bench, command failure, dirty containment). Nothing else.
|
|
@@ -12,7 +12,7 @@ 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. Hold the load the dispatch names and say which one: subjective by default, objective when
|
|
15
|
+
2. Hold the load the dispatch names and say which one: subjective by default, objective when the dispatch assigns it. Under the objective load, prove every refusal and failure path the contract states.
|
|
16
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
17
|
4. Apply `.agents/orchestration.md` § Parallelism for shared-file ownership.
|
|
18
18
|
5. Apply `.agents/orchestration.md` § Parallelism for validation ownership and the `orkestrel-parallel` skill § Implement in phases when that procedure applies.
|
|
@@ -21,7 +21,7 @@ On Claude Opus 5.5, you implement one dispatched unit and spawn nothing.
|
|
|
21
21
|
|
|
22
22
|
## Refuse
|
|
23
23
|
|
|
24
|
-
Adding a dependency, suppressing a diagnostic, mocking project-owned behavior, leaving a TODO in scope, committing, pushing, installing, reading a secret, a destructive command.
|
|
24
|
+
Adding a dependency, suppressing a diagnostic, mocking project-owned behavior, leaving a TODO in scope, committing outside a branch, clone, copy, or worktree the brief names, pushing, installing, reading a secret, a destructive command.
|
|
25
25
|
|
|
26
26
|
## Return
|
|
27
27
|
|
|
@@ -44,7 +44,7 @@ Treat generated package identifiers, versions, layers, and dependency ranges as
|
|
|
44
44
|
| Package | Version | Layer | Runtime dependencies | Peer dependencies |
|
|
45
45
|
| ----------------------- | --------- | ----- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------- |
|
|
46
46
|
| `@orkestrel/abort` | `0.0.15` | L1 | `@orkestrel/contract` `^0.0.21` | |
|
|
47
|
-
| `@orkestrel/agent` | `0.0.
|
|
47
|
+
| `@orkestrel/agent` | `0.0.35` | L5 | `@orkestrel/tool` `^0.0.21`, `@orkestrel/abort` `^0.0.15`, `@orkestrel/queue` `^0.0.19`, `@orkestrel/budget` `^0.0.16`, `@orkestrel/ndjson` `^0.0.15`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20`, `@orkestrel/workflow` `^0.0.25`, `@orkestrel/workspace` `^0.0.14` | |
|
|
48
48
|
| `@orkestrel/brief` | `0.0.14` | L4 | `@orkestrel/reason` `^0.0.16`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/interpret` `^0.0.19` | |
|
|
49
49
|
| `@orkestrel/browser` | `0.0.31` | L5 | `@orkestrel/mcp` `^0.0.41`, `@orkestrel/html` `^0.0.14`, `@orkestrel/pool` `^0.0.19`, `@orkestrel/tool` `^0.0.21`, `@orkestrel/codec` `^0.0.8`, `@orkestrel/router` `^0.0.19`, `@orkestrel/server` `^0.0.25`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/process` `^0.0.17`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/markdown` `^0.0.19`, `@orkestrel/websocket` `^0.0.17` | |
|
|
50
50
|
| `@orkestrel/budget` | `0.0.16` | L1 | `@orkestrel/contract` `^0.0.21` | |
|
|
@@ -66,10 +66,10 @@ Treat generated package identifiers, versions, layers, and dependency ranges as
|
|
|
66
66
|
| `@orkestrel/mcp` | `0.0.41` | L4 | `@orkestrel/sse` `^0.0.12`, `@orkestrel/tool` `^0.0.21`, `@orkestrel/codec` `^0.0.8`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/jsonrpc` `^0.0.1`, `@orkestrel/process` `^0.0.17`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/websocket` `^0.0.17` | `@orkestrel/router` `^0.0.19`, `@orkestrel/server` `^0.0.25` |
|
|
67
67
|
| `@orkestrel/middleware` | `0.0.27` | L4 | `@orkestrel/abort` `^0.0.15`, `@orkestrel/budget` `^0.0.16`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21` | `@orkestrel/server` `^0.0.25`, `@orkestrel/database` `^0.0.20` |
|
|
68
68
|
| `@orkestrel/msg` | `0.0.15` | L2 | `@orkestrel/codec` `^0.0.8`, `@orkestrel/contract` `^0.0.21` | |
|
|
69
|
-
| `@orkestrel/ndjson` | `0.0.
|
|
70
|
-
| `@orkestrel/ollama` | `0.0.
|
|
69
|
+
| `@orkestrel/ndjson` | `0.0.15` | L1 | `@orkestrel/contract` `^0.0.21` | |
|
|
70
|
+
| `@orkestrel/ollama` | `0.0.29` | L6 | `@orkestrel/tool` `^0.0.21`, `@orkestrel/agent` `^0.0.35`, `@orkestrel/budget` `^0.0.16`, `@orkestrel/ndjson` `^0.0.15`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/process` `^0.0.17`, `@orkestrel/contract` `^0.0.21` | |
|
|
71
71
|
| `@orkestrel/pool` | `0.0.19` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
72
|
-
| `@orkestrel/probe` | `0.0.
|
|
72
|
+
| `@orkestrel/probe` | `0.0.28` | L5 | `@orkestrel/lsp` `^0.0.13`, `@orkestrel/mcp` `^0.0.41`, `@orkestrel/pool` `^0.0.19`, `@orkestrel/tool` `^0.0.21`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/process` `^0.0.17`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21` | `vitest` `^4.1.12 \|\| ^5.0.5`, `typescript` `^6.0.3 \|\| ^7.0.2` |
|
|
73
73
|
| `@orkestrel/process` | `0.0.17` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
74
74
|
| `@orkestrel/program` | `0.0.19` | L4 | `@orkestrel/rater` `^0.0.20`, `@orkestrel/reason` `^0.0.16`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/qualifier` `^0.0.20` | |
|
|
75
75
|
| `@orkestrel/qualifier` | `0.0.20` | L3 | `@orkestrel/reason` `^0.0.16`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
@@ -78,19 +78,19 @@ Treat generated package identifiers, versions, layers, and dependency ranges as
|
|
|
78
78
|
| `@orkestrel/reason` | `0.0.16` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
79
79
|
| `@orkestrel/relation` | `0.0.18` | L3 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20` | |
|
|
80
80
|
| `@orkestrel/router` | `0.0.19` | L2 | `@orkestrel/abort` `^0.0.15`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
81
|
-
| `@orkestrel/scaffold` | `0.0.
|
|
81
|
+
| `@orkestrel/scaffold` | `0.0.122` | L3 | `@orkestrel/console` `^0.0.18`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/process` `^0.0.17`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/markdown` `^0.0.19`, `@orkestrel/template` `^0.0.12` | |
|
|
82
82
|
| `@orkestrel/sea` | `0.0.21` | L3 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/process` `^0.0.17`, `@orkestrel/contract` `^0.0.21` | |
|
|
83
83
|
| `@orkestrel/server` | `0.0.25` | L3 | `@orkestrel/abort` `^0.0.15`, `@orkestrel/codec` `^0.0.8`, `@orkestrel/router` `^0.0.19`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21` | |
|
|
84
84
|
| `@orkestrel/sqlite` | `0.0.16` | L1 | `@orkestrel/contract` `^0.0.21` | |
|
|
85
85
|
| `@orkestrel/sse` | `0.0.12` | L2 | `@orkestrel/codec` `^0.0.8`, `@orkestrel/contract` `^0.0.21` | |
|
|
86
|
-
| `@orkestrel/study` | `0.0.
|
|
86
|
+
| `@orkestrel/study` | `0.0.10` | L7 | `@orkestrel/tool` `^0.0.21`, `@orkestrel/abort` `^0.0.15`, `@orkestrel/agent` `^0.0.35`, `@orkestrel/budget` `^0.0.16`, `@orkestrel/ollama` `^0.0.29`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21` | |
|
|
87
87
|
| `@orkestrel/table` | `0.0.14` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
88
88
|
| `@orkestrel/template` | `0.0.12` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
89
89
|
| `@orkestrel/terminal` | `0.0.22` | L3 | `@orkestrel/sse` `^0.0.12`, `@orkestrel/form` `^0.0.15`, `@orkestrel/console` `^0.0.17`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20` | |
|
|
90
90
|
| `@orkestrel/test` | `0.0.32` | L1 | `@orkestrel/contract` `^0.0.21` | `vitest` `^4.1.12 \|\| ^5.0.5`, `playwright` `^1.64.1` |
|
|
91
91
|
| `@orkestrel/timeout` | `0.0.15` | L1 | `@orkestrel/contract` `^0.0.21` | |
|
|
92
92
|
| `@orkestrel/tool` | `0.0.21` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
93
|
-
| `@orkestrel/toolbox` | `0.0.
|
|
93
|
+
| `@orkestrel/toolbox` | `0.0.24` | L6 | `@orkestrel/form` `^0.0.15`, `@orkestrel/tool` `^0.0.21`, `@orkestrel/agent` `^0.0.35`, `@orkestrel/server` `^0.0.25`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20`, `@orkestrel/relation` `^0.0.18`, `@orkestrel/terminal` `^0.0.22`, `@orkestrel/workflow` `^0.0.25`, `@orkestrel/workspace` `^0.0.14` | |
|
|
94
94
|
| `@orkestrel/websocket` | `0.0.17` | L2 | `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21` | |
|
|
95
95
|
| `@orkestrel/worker` | `0.0.19` | L4 | `@orkestrel/pool` `^0.0.19`, `@orkestrel/queue` `^0.0.19`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20` | |
|
|
96
96
|
| `@orkestrel/workflow` | `0.0.25` | L4 | `@orkestrel/abort` `^0.0.15`, `@orkestrel/queue` `^0.0.19`, `@orkestrel/budget` `^0.0.16`, `@orkestrel/emitter` `^0.0.14`, `@orkestrel/timeout` `^0.0.15`, `@orkestrel/contract` `^0.0.21`, `@orkestrel/database` `^0.0.20` | |
|
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: reviewer
|
|
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.'
|
|
3
|
+
description: 'Opus 5.5 review lane by default, on another engine when the dispatch names one, over implemented work, read-only on source: 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 source.'
|
|
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 by default, or on the engine the dispatch names, you review by trying to falsify the claims. You
|
|
10
|
+
On Claude Opus 5.5 by default, or on the engine the dispatch names, you review by trying to falsify the claims. You edit no source; your final message is the verdict. Where the dispatch names a probe folder and your tools include a shell, write throwaway probes only in that folder, run them under `.agents/skills/orkestrel-parallel/SKILL.md` § Bound the host, and delete the folder before returning. Otherwise take the receipts the dispatch supplies, and run no mutating command.
|
|
11
11
|
|
|
12
12
|
## Do
|
|
13
13
|
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: scout
|
|
3
|
-
description: 'Haiku 5.5 read-only exact bounded lookup of files, symbols, and seams. Returns file:line pointers and a shape summary. Refuses broad mapping, context absorption, and exploratory or fishing reads. Never edits or judges quality.'
|
|
3
|
+
description: 'Haiku 5.5 read-only exact bounded lookup of files, symbols, and seams, and row-by-row marking of a list a search made against explicit criteria. Returns file:line pointers and a shape summary. Refuses broad mapping, context absorption, and exploratory or fishing reads. Never edits or judges quality.'
|
|
4
4
|
tools: Read, Grep, Glob
|
|
5
5
|
model: haiku
|
|
6
6
|
effort: low
|
|
@@ -17,6 +17,7 @@ distillation, and exploratory or fishing reads; refer the dispatch to `.agents/o
|
|
|
17
17
|
1. Take one bounded question: what to find and where to stop.
|
|
18
18
|
2. Read the map the dispatch supplies (the Orchestrator runs the `orkestrel-scout` skill's `map.ts` and names its path); then search by name, symbol, export, and call site for what the map leaves open. Open a file only far enough to confirm a match. Refuse a dispatch that names no map and asks for one.
|
|
19
19
|
3. Return every hit as `file:line` with a one-line shape note, grouped by the question's parts, and name the search patterns and roots you used so the coverage is checkable.
|
|
20
|
+
4. When the dispatch supplies a listed population and explicit criteria, mark every listed row against those criteria and return each row with its mark, and leave a row the criteria do not decide unmarked with its reason. Add no row by hand.
|
|
20
21
|
|
|
21
22
|
## Return
|
|
22
23
|
|
|
@@ -13,7 +13,7 @@ On Claude Haiku 5.5, run the requested gates and report their results. Write onl
|
|
|
13
13
|
## Do
|
|
14
14
|
|
|
15
15
|
1. Run exactly the commands the dispatch names, in order. Default scoped sweep: the format check and lint over the changed files, then the `check:` script and `test:` script of each project the dispatch names. Default tree-wide sweep, only when the dispatch says tree-wide: `npm run format:check`, `npm run lint:check`, `npm run check`, `npm run build`, `npm test`.
|
|
16
|
-
2. Read each gate bare. Never pipe it through `tail` or `grep`.
|
|
16
|
+
2. Read each gate bare. Never pipe it through `tail` or `grep`. When the shell yields before a command exits, keep its output in its file and resume per `.agents/skills/orkestrel-parallel/SKILL.md` § Run evidence commands until its terminal record arrives, and send no final message before every named gate has returned. When a dispatch hands you a log the Orchestrator ran, read it whole and confirm its command, its source identity, its output, and its exit; that read is the independent receipt, and you report it as a read.
|
|
17
17
|
3. Record each gate's outcome by exit code. A gate that "mostly passes" failed.
|
|
18
18
|
4. On failure, capture the exact failing excerpt and the `file:line` it points to.
|
|
19
19
|
5. Report a failure that passes alone per `orkestrel-dispatch` § Write the brief, and rerun nothing to reclassify it; run again only after a changed condition or an explicit investigation request.
|
|
@@ -19,15 +19,15 @@ paths:
|
|
|
19
19
|
|
|
20
20
|
A package directory holds:
|
|
21
21
|
|
|
22
|
-
| File | Holds
|
|
23
|
-
| ----------------------- |
|
|
24
|
-
| `plan.md` | a concise goal and exit criterion that serve as a goal prompt, each repository the campaign changes by name and URL, § State, open decisions, deferred triggers, § Sequence | always |
|
|
25
|
-
| `ledger.md` | one row per dispatched unit and per user ruling on routing or scope: date, unit, role, carrier
|
|
26
|
-
| `README.md` | the file map, the read order, and the id law
|
|
27
|
-
| `tenets.md` | the campaign's standing directives
|
|
28
|
-
| `ideas.md`, `gaps.md` | ideas set aside and library gaps, each numbered for life with its reason and its reopen trigger
|
|
29
|
-
| `designs/`, `research/` | one design or research record per subject
|
|
30
|
-
| a verdict file | the verdict of an open seam
|
|
22
|
+
| File | Holds | When |
|
|
23
|
+
| ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- |
|
|
24
|
+
| `plan.md` | a concise goal and exit criterion that serve as a goal prompt, each repository the campaign changes by name and URL, the governing skills, § State, open decisions, deferred triggers, § Sequence | always |
|
|
25
|
+
| `ledger.md` | one row per dispatched unit and per user ruling on routing or scope: date, unit, role, carrier | always |
|
|
26
|
+
| `README.md` | the file map, the read order, and the id law | when the folder holds more than `plan.md` and `ledger.md` |
|
|
27
|
+
| `tenets.md` | the campaign's standing directives | when the campaign has any |
|
|
28
|
+
| `ideas.md`, `gaps.md` | ideas set aside and library gaps, each numbered for life with its reason and its reopen trigger | when the campaign records any |
|
|
29
|
+
| `designs/`, `research/` | one design or research record per subject | when a chunk needs one |
|
|
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
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.
|
|
@@ -37,14 +37,15 @@ A package directory holds:
|
|
|
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
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
|
-
- `plan.md` holds
|
|
40
|
+
- `plan.md` holds the campaign's claims, its release bar, each seam's stop with what reopens it, and its deferral list, each deferred item with its trigger and its owner.
|
|
41
|
+
- Name in `plan.md` the skill that governs each kind of unit the campaign dispatches, whatever the campaign's own name: `orkestrel-harden` for hardening a package.
|
|
41
42
|
- 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.
|
|
42
43
|
- Keep unit briefs, reports, journals, and drafts under the checkout's `tmp/` (`.agents/orchestration.md` § tmp layout); that file's § Cleanup owns what a unit's files carry before they go.
|
|
43
44
|
- Record a prepared, unapplied change as text in `plan.md` or as a commit, never as a pointer to scratch.
|
|
44
45
|
|
|
45
46
|
## Start a session
|
|
46
47
|
|
|
47
|
-
1. Read the package folder's `README.md` when present, then `plan.md` § State, then the first open chunk of § Sequence.
|
|
48
|
+
1. Read the package folder's `README.md` when present, then `plan.md` § State, then each skill the plan names as governing the first open chunk of § Sequence, then that chunk.
|
|
48
49
|
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
50
|
3. Act only on what the folder, the tree, and git hold. Agent memory from another machine or harness is absent.
|
|
50
51
|
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`.
|
|
@@ -83,6 +83,7 @@ A skill, a tool description, and a refusal message each teach an agent mid-task.
|
|
|
83
83
|
|
|
84
84
|
- Put the critical path in the tool description, most important first: a worked example, the key nesting, and the reply shape. A harness abbreviates a nested schema and delivers the description intact; never leave the teaching to the schema.
|
|
85
85
|
- Make every refusal orient, name, or correct: state the subject, the exact key path, the expected shape, the valid candidates, and every missing and every unexpected item, in one refusal.
|
|
86
|
+
- Keep the provenance a refusal computed, its token, span, reason, and known field, and never invent the value or field the caller intended.
|
|
86
87
|
- Refuse with the correction in the message. Never coerce a malformed input and never silently drop it. Encode an untrusted identifier in a diagnostic, JSON-encoded and length-capped, so malformed input cannot fake or balloon the message.
|
|
87
88
|
- Acknowledge or state, never both: a mutation returns an acknowledgement and a read returns state. Say the split in both tools' descriptions so neither is polled for the other's job.
|
|
88
89
|
|
|
@@ -129,6 +129,20 @@ Use these orthogonal contract mechanisms:
|
|
|
129
129
|
- every parsed result satisfies its guard.
|
|
130
130
|
- Derive parser and guard from one source or test the round trip.
|
|
131
131
|
|
|
132
|
+
### Admission and refusal
|
|
133
|
+
|
|
134
|
+
- Choose a hand guard or a shape-derived guard by how often the check runs and the representation it reads: a hand guard with a directional parity test on a per-element path, a derived guard at a once-per-entity door, and a diagnostic computed only after a refusal.
|
|
135
|
+
- Admit an owned record from its raw form before any projection or fallback, so that corrupt-present stays apart from absent.
|
|
136
|
+
- State shape soundness, lossless semantic admission, and authoring validity as separate claims, and prove each at its own door.
|
|
137
|
+
- Treat persistence canonicalization as part of the declared identity round trip: when a store reorders or rewrites a value, compare identity by the value the declaration states.
|
|
138
|
+
- Make a refusal bound cover the ownership and capture work paid before validation: read the limit during capture, with one counter per door call.
|
|
139
|
+
- Report unread or unavailable evidence as unread, never as a false predicate or an empty success.
|
|
140
|
+
- Bind admission and approval to the exact operation and instance, and recheck that binding across every suspension and callback before dispatch.
|
|
141
|
+
- Name the actual commit boundary in a restore or continuation claim, partial commits included.
|
|
142
|
+
- Declare a parser's whole admitted language and its adjacent refusals, and prove them with a table of the class; never tune the parser example by example.
|
|
143
|
+
- Derive a decision from validated inputs; an editable prior result never substitutes for them.
|
|
144
|
+
- Project a value outward by its authorship and trust origin, and accept no authority channel from a client.
|
|
145
|
+
|
|
132
146
|
### Foreign contracts
|
|
133
147
|
|
|
134
148
|
A value returned by an interface another package publishes is FOREIGN data. Own it, validate what you dereference, and narrow nothing.
|
|
@@ -33,7 +33,7 @@ These rules govern how a claim is established: the evidence read before a change
|
|
|
33
33
|
|
|
34
34
|
## Instruments
|
|
35
35
|
|
|
36
|
-
- An instrument counts as evidence only after it has failed: pair every probe, comparison, or matrix with a negative control that must fail under the same conditions, drawn from outside the population the instrument covers.
|
|
36
|
+
- An instrument counts as evidence only after it has failed: pair every probe, comparison, or matrix with a negative control that must fail under the same conditions, drawn from outside the population the instrument covers and inside the domain the claim promises. A control that leaves that domain proves nothing about the instrument; prove that behavior under its own claim.
|
|
37
37
|
- When a TypeScript question can name a workspace project, a case (files plus a test), and a control (files, test, the stage it must fail at, and why), call the `prove` tool the `probe` MCP server registers. Quote its closing line (`receipt probe:<digest>:…` or `no receipt`) where the claim is reported. A `no receipt` line leaves the claim unproved; report the stage that refused.
|
|
38
38
|
- When the question supplies no project, no case, or no control, write a probe per `.claude/rules/tests.md` § Probes and report the probe's control and coverage.
|
|
39
39
|
- When no `probe` server is registered, register one outside the repository in the harness's user or local MCP scope, pointing at `node_modules/@orkestrel/probe/dist/bin/main.js`, and start it in the repository whose projects the question names. A scaffold target holds no `.mcp.json`.
|
|
@@ -57,6 +57,9 @@ These rules govern how a claim is established: the evidence read before a change
|
|
|
57
57
|
- Before the run, size the fixtures, the sample count, and the threshold to the case, and state the reasoning and the result that makes the change worth keeping.
|
|
58
58
|
- Read the difference between methods against each side's reported uncertainty, and record no difference smaller than the result stated before the run.
|
|
59
59
|
- Report every measurement with the host's other load during the run.
|
|
60
|
+
- Compare two builds interleaved in one run under one host load, with `.agents/skills/orkestrel-harden/scripts/interleave.ts`. Read a published side from the published artifact, calibrate the run with identical sides, and bar the relative change, never an absolute reading from another moment.
|
|
61
|
+
- Price a cost over the caller's whole lifecycle and workload: construction, restoration, transfer, and decision beside the warm saving.
|
|
62
|
+
- Derive a cost stop from what the consumer spends around the cost, and state responsiveness and throughput as separate claims.
|
|
60
63
|
- Rule a cost finding advisory unless it shows its effect under a realistic load, and never carry an advisory into a fix brief as a required change; `.agents/skills/orkestrel-falsify/SKILL.md` § Verdict shape places it in a review lane's return.
|
|
61
64
|
|
|
62
65
|
## Ecosystem reuse
|
|
@@ -51,6 +51,8 @@ These rules govern what a test asserts and where it lives, the cross-cutting and
|
|
|
51
51
|
- A browser case restores every runner global it changes before it ends.
|
|
52
52
|
- A proof that runs before `build` in the gate chain resolves no `dist/` path; a proof that deliberately consumes the built entry loads it at run time through a resolved path and refuses a stale build.
|
|
53
53
|
- When data a generator renders gains a new shape, the generator's test gains that shape before the data ships.
|
|
54
|
+
- Give a negative control on a synchronous guard a bounded fixture that finishes with the guard removed and whose count differs, because a test timer on the same thread cannot interrupt it.
|
|
55
|
+
- Prove a consumer migration past what compiles: cover the defaulted branches that still compile and the entries outside the check graph.
|
|
54
56
|
- The revert that proves a repair reddens exactly the test that names the defect. Keep the import and collection graph valid while reverting, and confirm the named test was collected. A revert that reddens anything beyond that test broke the harness, and its count is not evidence.
|
|
55
57
|
- Use `it.todo()` only for explicitly out-of-scope roadmap work, never to complete the current request. Every conditional skip has a narrow verifiable applicability reason. Never commit `.only` or `.skip`; `vitest/no-focused-tests` and `vitest/no-disabled-tests` refuse both.
|
|
56
58
|
- Do not create test files solely for `constants.ts`, barrels, error definitions, or `types.ts`.
|
|
@@ -174,6 +176,8 @@ A test that spawns a process, packs, installs, or drives a real build is a proof
|
|
|
174
176
|
path, run its `distribution` project before accepting the change.
|
|
175
177
|
- Put a study's proofs under `tests/studies/` in the `studies` project, which
|
|
176
178
|
`.claude/rules/workspace.md` § Test project matrix places.
|
|
179
|
+
- Keep a study's developmental selection and its untouched confirmation as separate evidence
|
|
180
|
+
populations, and withhold expected outputs and population membership from the run.
|
|
177
181
|
- Put in `studies` only the proofs the `@orkestrel/study` engine drives. Place a slow end-to-end proof of
|
|
178
182
|
the package's own features that no study engine drives, such as its built executable run against
|
|
179
183
|
a fixture workspace, as an integration test under § Cross-cutting proofs. Place it by the scope it
|
|
@@ -56,7 +56,7 @@
|
|
|
56
56
|
- Cut every clause that persuades, reassures, or explains to a person.
|
|
57
57
|
- Give a rule one home. Point at it from elsewhere; never restate it.
|
|
58
58
|
- When a rule requires a check against a declared type or shape, state the standard at which the check stops.
|
|
59
|
-
- Keep an example only when it disambiguates the rule.
|
|
59
|
+
- Keep an example only when it disambiguates the rule. `.agents/skills/orkestrel-harden/references/methods.md` keeps one reading and one example per method.
|
|
60
60
|
- When an instruction file or a brief governs a quantity, state the property the quantity must serve and leave the figure to the case: never fix a fixture size, a sample count, a percentage, or a threshold, and describe a fixture by its kind and the load it represents. Keep a figure only where it is a contract: a limit a protocol, format, standard, or external tool fixes, an exit code, a measurement with its run, a version, a date, or a cardinality a design law fixes (one word per name, one writer per checkout).
|
|
61
61
|
- Keep `AGENTS.md` to the laws every task needs and the always-loaded set small: move a procedure into a skill and scope a rule with `paths`.
|
|
62
62
|
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: orkestrel-parallel
|
|
3
3
|
description: >-
|
|
4
|
-
Run parallel work: lanes inside one session (subagents, child tasks, background commands) and peer sessions that share a checkout, a machine, or an execution lane. Use when units can progress in parallel, when another session works on the same machine or checkout, or before the first lane on a harness. Bound the host, keep lanes visible, coordinate writers through messages with one runner for checks, and
|
|
4
|
+
Run parallel work: lanes inside one session (subagents, child tasks, background commands) and peer sessions that share a checkout, a machine, or an execution lane. Use when units can progress in parallel, when another session works on the same machine or checkout, or before the first lane on a harness. Bound the host, keep lanes visible, coordinate writers through messages with one runner for checks, and serialize a shared execution lane through one queue per host. Each harness page records its mechanics and quirks; a harness with no page carries no writer or runner lane until its bring-up questions are answered on one.
|
|
5
5
|
---
|
|
6
6
|
|
|
7
7
|
# Load the canonical workflow
|
|
@@ -24,6 +24,9 @@ return Design, Alternatives, Constraints (each with file:line), Refusals (option
|
|
|
24
24
|
forecloses, rule quoted), Measurements, Units, Tensions, Risks; fill the sections your lane owns
|
|
25
25
|
and leave the others empty.
|
|
26
26
|
|
|
27
|
-
Use a clean context; never review work you wrote.
|
|
27
|
+
Use a clean context; never review work you wrote. Edit no source. Where the dispatch names a probe
|
|
28
|
+
folder and your sandbox permits writing there, write throwaway probes only in that folder, run them
|
|
29
|
+
under .agents/skills/orkestrel-parallel/SKILL.md § Bound the host, and delete the folder before
|
|
30
|
+
returning; otherwise take the receipts the dispatch supplies. Never implement, reconcile, accept, or
|
|
28
31
|
spawn. Your final message is the verdict or the proposal.
|
|
29
32
|
"""
|
|
@@ -16,7 +16,7 @@ orkestrel-parallel skill § Implement in phases when that procedure applies. Lea
|
|
|
16
16
|
per .claude/rules/tests.md § Probes.
|
|
17
17
|
|
|
18
18
|
Never add a dependency, suppress a diagnostic, mock project-owned behavior, leave a TODO in scope,
|
|
19
|
-
commit, push, install, read a secret, or run a destructive command. Stop only per
|
|
19
|
+
commit outside a branch, clone, copy, or worktree the brief names, push, install, read a secret, or run a destructive command. Stop only per
|
|
20
20
|
.agents/orchestration.md § Deviation protocol.
|
|
21
21
|
|
|
22
22
|
Refuse a unit that authors a proof class .agents/orchestration.md § Routing sends to opus: report
|
|
@@ -13,7 +13,7 @@ Follow AGENTS.md § Work loop at the size the brief names. Apply .agents/orchest
|
|
|
13
13
|
.claude/rules/tests.md § Probes.
|
|
14
14
|
|
|
15
15
|
A unit that leaves naming, API shape, or architecture open, or that spans packages, belongs to
|
|
16
|
-
astra or opus: stop and say so. A unit whose mechanism is open: a protocol, a lock, a retry class, or a termination path; report it as belonging to astra or opus. Never install, commit, push, read a secret, or run a destructive command. Stop and report
|
|
16
|
+
astra or opus: stop and say so. A unit whose mechanism is open: a protocol, a lock, a retry class, or a termination path; report it as belonging to astra or opus. Never install, commit outside a branch, clone, copy, or worktree the brief names, push, read a secret, or run a destructive command. Stop and report
|
|
17
17
|
per .agents/orchestration.md § Deviation protocol on any conflict with the brief.
|
|
18
18
|
|
|
19
19
|
Return changed files with one line each, the scoped commands run and their results, shared-file
|
|
@@ -8,8 +8,10 @@ Read so the Orchestrator does not have to. Decide nothing.
|
|
|
8
8
|
|
|
9
9
|
Take one bounded question and an exact list of files or a diff. Read every named input in full
|
|
10
10
|
and record each fact with file:line. Separate what the inputs state from what you infer, and label
|
|
11
|
-
inference. Name every input row the distillate did not reach.
|
|
11
|
+
inference. Name every input row the distillate did not reach. When the inputs outlast one turn,
|
|
12
|
+
stop at a clean boundary and never sample. A partial read never counts as a whole read.
|
|
12
13
|
|
|
13
14
|
Return only: Question, Evidence, Contradictions (both sides cited), Distillate (the smallest
|
|
14
|
-
context the next engine needs), Unknowns
|
|
15
|
+
context the next engine needs), Unknowns, Coverage (per input: its identity, the last line read
|
|
16
|
+
without a gap, and the next slice). No recommendation. Never spawn another agent.
|
|
15
17
|
"""
|
|
@@ -10,10 +10,13 @@ Read .agents/transports/cursor.md and follow it exactly: model pin, CLI resoluti
|
|
|
10
10
|
form, journal, recovery. The route is grok, mode --mode=ask, read-only.
|
|
11
11
|
|
|
12
12
|
Require a bounded question and an exact file scope; refuse an unbounded one. Draft the brief
|
|
13
|
-
(read-only, evidence sought, file:line pointers, no raw dumps, no decisions
|
|
13
|
+
(read-only, evidence sought, file:line pointers, no raw dumps, no decisions, and a stop at a clean
|
|
14
|
+
boundary with a coverage table when the inputs outlast one answer; a partial read never counts as a
|
|
15
|
+
whole read). Your sandbox is
|
|
14
16
|
read-only, so return the brief text, its intended path tmp/cursor/<unit>-brief.md, the resolved
|
|
15
17
|
command, and the journal path; the Orchestrator writes and launches them.
|
|
16
18
|
|
|
17
|
-
Return only: Question, Evidence, Distillate, Unknowns,
|
|
19
|
+
Return only: Question, Evidence, Distillate, Unknowns, Coverage (per input: its identity, the last
|
|
20
|
+
line read without a gap, and the next slice), Journal (path and session id), Deviation.
|
|
18
21
|
Never spawn another agent.
|
|
19
22
|
"""
|
|
@@ -7,8 +7,8 @@ developer_instructions = """
|
|
|
7
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. Hold the load the dispatch names and say which: subjective by default, objective when
|
|
11
|
-
|
|
10
|
+
guide. Hold the load the dispatch names and say which: subjective by default, objective when the
|
|
11
|
+
dispatch assigns it. Under the objective load, prove
|
|
12
12
|
every refusal and failure path the contract states. Follow AGENTS.md § Work loop: types first, prove the claim, implement, test, consolidate,
|
|
13
13
|
document. Select the instrument per AGENTS.md § Work loop and .claude/rules/quality.md § Instruments
|
|
14
14
|
and return the evidence that instrument produced.
|
|
@@ -19,7 +19,7 @@ orkestrel-parallel skill § Implement in phases when that procedure applies. Lea
|
|
|
19
19
|
.agents/orchestration.md § Deviation protocol.
|
|
20
20
|
|
|
21
21
|
Never add a dependency, suppress a diagnostic, mock project-owned behavior, leave a TODO in
|
|
22
|
-
scope, commit, push, install, read a secret, or run a destructive command.
|
|
22
|
+
scope, commit outside a branch, clone, copy, or worktree the brief names, push, install, read a secret, or run a destructive command.
|
|
23
23
|
|
|
24
24
|
Return touched files with one line each, diffstat, scoped validation commands and results, the
|
|
25
25
|
instrument's closing evidence per claim, failing-first test names, shared-file patches, deviation state.
|
|
@@ -1,11 +1,15 @@
|
|
|
1
1
|
name = "reviewer"
|
|
2
|
-
description = "
|
|
2
|
+
description = "GPT-6 Astra review by default, on another engine when the dispatch names one, read-only on source, against numbered claims in the lane the dispatch names: design fit or correctness. Never edits source 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
|
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
|
-
commands.
|
|
8
|
+
commands. Edit no source and spawn nothing. Where the dispatch names a probe folder and your sandbox
|
|
9
|
+
permits writing there, write throwaway probes only in that folder, run them under
|
|
10
|
+
.agents/skills/orkestrel-parallel/SKILL.md § Bound the host, and delete the folder before returning.
|
|
11
|
+
Otherwise take the receipts the dispatch supplies, and run no mutating command. Use a clean context;
|
|
12
|
+
never review work you wrote.
|
|
9
13
|
|
|
10
14
|
Read the claims file, the actual diff and git status --porcelain the dispatch supplies,
|
|
11
15
|
AGENTS.md, the rules whose paths match the changed files, the guide, and enough surrounding
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
name = "scout"
|
|
2
|
-
description = "Read-only exact bounded lookup of files, symbols, and seams; returns file:line pointers and a shape summary; refuses broad mapping, context absorption, and exploratory or fishing reads; never edits or judges."
|
|
2
|
+
description = "Read-only exact bounded lookup of files, symbols, and seams, and row-by-row marking of a list a search made against explicit criteria; returns file:line pointers and a shape summary; refuses broad mapping, context absorption, and exploratory or fishing reads; never edits or judges."
|
|
3
3
|
model = "gpt-6-luna"
|
|
4
4
|
model_reasoning_effort = "low"
|
|
5
5
|
sandbox_mode = "read-only"
|
|
@@ -15,6 +15,10 @@ a match. Refuse a dispatch that names no map and asks for one. Return every hit
|
|
|
15
15
|
one-line shape note, grouped by the question's parts, and name the search patterns and roots you
|
|
16
16
|
used.
|
|
17
17
|
|
|
18
|
+
When the dispatch supplies a listed population and explicit criteria, mark every listed row against
|
|
19
|
+
those criteria and return each row with its mark, and leave a row the criteria do not decide
|
|
20
|
+
unmarked with its reason. Add no row by hand.
|
|
21
|
+
|
|
18
22
|
Return only: Question, Hits, Shape (the shortest summary that places every part of the question), Not found (patterns that returned nothing).
|
|
19
23
|
Never spawn another agent.
|
|
20
24
|
"""
|
|
@@ -12,7 +12,12 @@ and continuation. Never spawn another agent.
|
|
|
12
12
|
Run exactly the commands the dispatch names, in order. Default scoped sweep: the format check and
|
|
13
13
|
lint over the changed files, then the check: and test: scripts of each named project. Default tree-wide sweep, only when the dispatch says tree-wide:
|
|
14
14
|
npm run format:check; npm run lint:check; npm run check; npm run build; npm test. Read each gate
|
|
15
|
-
bare, never through tail or grep.
|
|
15
|
+
bare, never through tail or grep. When the shell yields before a command exits, keep its output in
|
|
16
|
+
its file and resume per .agents/skills/orkestrel-parallel/SKILL.md § Run evidence commands until its
|
|
17
|
+
terminal record arrives, and send no final message before every named gate has returned. When a
|
|
18
|
+
dispatch hands you a log the Orchestrator ran, read it whole and confirm its command, its source
|
|
19
|
+
identity, its output, and its exit; that read is the independent receipt, and you report it as a
|
|
20
|
+
read. Record each outcome by exit code; a gate that mostly passes
|
|
16
21
|
failed. On failure capture the exact excerpt and the file:line it points to. Report a failure that
|
|
17
22
|
passes alone per orkestrel-dispatch § Write the brief, and rerun nothing to reclassify it; run again
|
|
18
23
|
only after a changed condition or an explicit investigation request.
|