owenloop 0.5.29 → 0.5.31
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/CHANGELOG.md +21 -0
- package/dist/packages/work/src/agent/brief.js +73 -4
- package/dist/packages/work/src/agent/loop.d.ts +5 -3
- package/dist/packages/work/src/agent/loop.js +24 -3
- package/dist/packages/work/src/agent/workdir.d.ts +17 -0
- package/dist/packages/work/src/agent/workdir.js +24 -0
- package/dist/packages/work/src/exec/loop.d.ts +8 -1
- package/dist/packages/work/src/exec/loop.js +24 -1
- package/dist/packages/work/src/hub/client.d.ts +10 -3
- package/dist/packages/work/src/hub/client.js +5 -4
- package/dist/packages/work/src/hub/deadline.d.ts +20 -0
- package/dist/packages/work/src/hub/deadline.js +49 -0
- package/dist/packages/work/src/hub/types.d.ts +17 -0
- package/dist/packages/work/src/roles/agent-run.js +1 -0
- package/dist/packages/work/src/roles/exec.js +1 -0
- package/dist/packages/work/src/settings/hub-roster-cache.js +5 -23
- package/dist/packages/work/src/shift/loop.d.ts +59 -0
- package/dist/packages/work/src/shift/loop.js +177 -19
- package/dist/packages/work/src/shift/protocol.d.ts +53 -1
- package/dist/packages/work/src/shift/runtime.js +10 -1
- package/dist/packages/work/src/shift/server.js +2 -0
- package/dist/src/bundle/manifest.d.ts +6 -1
- package/dist/src/bundle/manifest.js +38 -8
- package/dist/src/bundle/runtime.d.ts +1 -1
- package/dist/src/bundle/runtime.js +1 -0
- package/dist/src/bundle/types.d.ts +3 -2
- package/dist/src/cli.js +17 -3
- package/dist/src/crypto/records.d.ts +1 -0
- package/dist/src/crypto/records.js +11 -0
- package/dist/src/defs.js +17 -4
- package/dist/src/engine.d.ts +3 -0
- package/dist/src/engine.js +55 -10
- package/dist/src/factory.d.ts +2 -0
- package/dist/src/factory.js +2 -0
- package/dist/src/index.d.ts +1 -1
- package/dist/src/index.js +1 -1
- package/dist/src/schemas/index.d.ts +3 -0
- package/dist/src/schemas/order.v1.schema.json +3 -0
- package/dist/src/store/def-source.d.ts +1 -1
- package/dist/src/store/def-source.js +25 -0
- package/dist/src/store/gc.js +9 -7
- package/dist/src/store/index.d.ts +1 -0
- package/dist/src/store/index.js +1 -0
- package/dist/src/store/install.js +9 -3
- package/dist/src/store/instruction-source.js +9 -5
- package/dist/src/store/verified-lock.d.ts +4 -0
- package/dist/src/store/verified-lock.js +38 -0
- package/dist/src/store.js +4 -2
- package/dist/src/types.d.ts +40 -0
- package/docs/authoring.md +12 -0
- package/docs/bundles.md +3 -2
- package/docs/cli.md +39 -11
- package/docs/shift-logs.md +53 -4
- package/package.json +1 -1
- package/plugins/claude-code/plugin/.claude-plugin/plugin.json +1 -1
- package/plugins/claude-code/plugin/.mcp.json +1 -1
- package/plugins/codex/plugins/owenloop/.codex-plugin/plugin.json +1 -1
- package/plugins/codex/plugins/owenloop/.mcp.json +1 -1
package/docs/bundles.md
CHANGED
|
@@ -75,13 +75,13 @@ The following rules apply:
|
|
|
75
75
|
- `runtime` is optional. When present, `runtime` is a closed mapping that contains `minVersion`, `features`, or both; `runtime: {}` is invalid. Runtime requirements are described in [Runtime compatibility](#runtime-compatibility).
|
|
76
76
|
- `workflows` is a non-empty map from workflow names to archive-relative YAML paths. Each workflow name matches `/^[a-z][a-z0-9-]*$/`. Each path must pass the canonical bundle path safety policy, and two workflow names may not point to the same path.
|
|
77
77
|
- A workflow file's `name:` must equal the workflow map key. For example, `workflows.delivery` must point to a file containing `name: delivery`.
|
|
78
|
-
- An installed workflow is addressable from `calls:` by the qualified name `<package>/<workflow>`. A bare `calls: <workflow>` inside a CAS-installed bundle is resolved to a sibling from that same bundle; it does not search unrelated bundles for a same-named workflow.
|
|
78
|
+
- An installed workflow is addressable from `calls:` by the qualified name `<package>/<workflow>`. A bare `calls: <workflow>` inside a CAS-installed bundle is resolved to a sibling from that same bundle; it does not search unrelated bundles for a same-named workflow. An exact `calls: namespace/package@version` selects the verified bundle's default (or sole) workflow. An exact `calls: namespace/package@version#workflow` selects that named workflow from the same verified bundle, whether or not it is the default. The `#workflow` suffix must match a registered `workflows` key; `#` is forbidden in store coordinates and the selector cannot be empty. Both forms use the base `namespace/package@version` as their `lock` key. The coordinator resolves through the verified store index and compares the selected child's bundle digest to that lock. A worker follows the verified parent's lock digest to a locally verified CAS object and selects the same workflow; it never uses coordinator-supplied child definition bytes.
|
|
79
79
|
- `default` is optional. When present, `default` must be one of the workflow map keys. When absent, a single-workflow package has that one workflow as its implicit default. A package with two or more workflows has no implicit default: the versioned coordinate is not registered as a callable target, while explicit `<package>/<workflow>` targets remain available.
|
|
80
80
|
- `platforms` is a duplicate-free list of portable selector strings. The list is advisory; platform policy is outside the package parser.
|
|
81
81
|
- `integrity.algorithm` is exactly `sha256`.
|
|
82
82
|
- `integrity.files` contains one lowercase 64-hex SHA-256 digest for every regular archive file except `bundle.yaml`. The packer regenerates this map from the source bytes and does not edit the source manifest. Excluding `bundle.yaml` avoids a recursive self-hash; the def digest still covers the complete canonical tar, including `bundle.yaml`.
|
|
83
83
|
- `capabilities` contains requested capability classes and values. The manifest does not grant capabilities.
|
|
84
|
-
- `lock` maps exact
|
|
84
|
+
- `lock` maps base exact coordinates `namespace/package@version` to lowercase 64-hex **bundle digests**. A named exact call `namespace/package@version#workflow` requires a lock at `namespace/package@version`; a selector is never a lock key. Every exact `calls:` target in every workflow must have a matching base-coordinate lock. A bare same-bundle call has no lock entry: its pin is the containing bundle's own digest. The runtime compares bundle digests at child spawn; it does not compare this lock value with a per-workflow instruction digest. Authors using named exact calls should declare `runtime.features: [exact-workflow-selector.v1]` so older engines refuse the bundle before executing it. Published Owenloop 0.5.30 and 0.5.25 do not support this feature. Use the first published release containing it as `runtime.minVersion` once released.
|
|
85
85
|
|
|
86
86
|
Canonical manifest serialization uses the fixed top-level order `formatVersion`, `package`, optional `runtime`, `workflows`, optional `default`, `platforms`, `integrity`, `capabilities`, `lock`. Runtime fields use the order `minVersion`, then `features`; feature identifiers are sorted by ascending UTF-8 bytes. Other mapping keys and list values use their defined canonical order. Strings use JSON-style double quotes, indentation is two spaces, and the document has exactly one final newline. Inspection rejects a manifest that parses correctly but is not in that canonical byte form.
|
|
87
87
|
|
|
@@ -105,6 +105,7 @@ The mapping is strict and closed. Unknown keys are rejected. At least one of `mi
|
|
|
105
105
|
|
|
106
106
|
This Owenloop release advertises exactly these feature identifiers:
|
|
107
107
|
|
|
108
|
+
- `exact-workflow-selector.v1`
|
|
108
109
|
- `harness-policy-enforcement.v1`
|
|
109
110
|
- `native-judge-policy-inheritance.v1`
|
|
110
111
|
- `neutral-approval-modes.v1`
|
package/docs/cli.md
CHANGED
|
@@ -290,7 +290,8 @@ that field a **crew**: `serve_crews` contains the selected crew names. Passing
|
|
|
290
290
|
the Scoped Identity. Do not treat `attended_at` as a liveness signal: every
|
|
291
291
|
accepted `shift next` records attendance and makes the next presence ping due,
|
|
292
292
|
but attendance is advisory and observability-only. Attendance never changes
|
|
293
|
-
routing, dispatch, or lease behavior.
|
|
293
|
+
routing, dispatch, or lease behavior. The liveness signal is `last_poll_at` on
|
|
294
|
+
`shift status` — see [`shift status`](#shift-status---state-dir-path).
|
|
294
295
|
|
|
295
296
|
<a id="config-dir"></a>
|
|
296
297
|
By default, the execution settings file is `$HOME/.owenloop/settings.json`.
|
|
@@ -384,13 +385,31 @@ Only a workdir the ORDER names is checked. A step that declares neither
|
|
|
384
385
|
controls — the shift's own launch directory for a command step, or
|
|
385
386
|
`<workRoot>/<workflow>/<run>/` for an agent step — and is never denied.
|
|
386
387
|
|
|
388
|
+
The same two workers make one more check on an order-named workdir, right after
|
|
389
|
+
the roots check: the directory must still **exist** on this machine and be a
|
|
390
|
+
directory. A `workdirFrom:` value can outlive what it names — a cleanup step
|
|
391
|
+
reclaims a worktree, then a rejection re-arms an earlier step whose workdir was
|
|
392
|
+
that worktree — and spawning there used to fail inside the command or the
|
|
393
|
+
harness with a message that never mentioned the cwd. Now `owenloop work exec`
|
|
394
|
+
and `owenloop work agent-run` refuse before anything is spawned or any session
|
|
395
|
+
opens. The shape is the same as the roots refusal: a **release**, not a
|
|
396
|
+
failure. The worker releases its claim with the reason
|
|
397
|
+
`step workdir no longer exists: <absolute path>`, exits non-zero (outcome
|
|
398
|
+
`workdir-missing`, exit 1), and writes one line to `<log-dir>/<run>.log` naming
|
|
399
|
+
the path. The step stays eligible and the order returns to the pickup window;
|
|
400
|
+
the hub records the reason as a routing alert (`owenloop routing alerts
|
|
401
|
+
--workflow <wf>`), keeping only the first observation per workflow and step. The
|
|
402
|
+
roots check runs first, so a workdir that is both outside the roots and missing
|
|
403
|
+
is reported as denied, not missing. A step that names no workdir is never
|
|
404
|
+
existence-checked — its fallback directory is one owenloop itself controls.
|
|
405
|
+
|
|
387
406
|
`--work-root` is not `--work-root`'s neighbour `workRoot`. `workRoot` (settings,
|
|
388
407
|
`OWENLOOP_WORK_ROOT`) is the ONE directory owenloop **creates** per-run
|
|
389
408
|
directories under. `--work-root` / `allowedWorkdirRoots` is the SET of
|
|
390
409
|
directories an order is **allowed to name**. Neither derives from the other.
|
|
391
410
|
|
|
392
411
|
```bash
|
|
393
|
-
owenloop shift start
|
|
412
|
+
owenloop shift start openai --work-root ~/code --work-root /srv/work
|
|
394
413
|
```
|
|
395
414
|
|
|
396
415
|
**Crew roster selection.** A selected roster candidate decides
|
|
@@ -590,24 +609,28 @@ either `exec` or `agent-run`:
|
|
|
590
609
|
- `order-dropped`: `{ "type": "order-dropped", "workflow": "...", "run": "...", "step": "...", "reason": "unsupported-worker", "message": "..." }` — the shift refused one order. Match on `reason` (`malformed-digest`, `malformed-worker`, `unsupported-worker`, `verification-failed`, `metadata-unavailable`, `agent-lane-closed`, `dispatch-cap-full`, `agent-cap-full`, `claim-expired`); display `message`. The capacity, expiry, and `agent-lane-closed` reasons return the claim to the hub; malformed and unsupported reasons leave it for the pickup window.
|
|
591
610
|
- `ended`: `{ "type": "ended" }`, delivered to a parked `next` when `shift end` explicitly shuts down the daemon
|
|
592
611
|
- `wedged`: `{ "type": "wedged", "faults": [{ "kind": "missing", "role": "temp-dir", "path": "...", "message": "..." }], "streak": 3 }` — the shift stopped because the MACHINE it runs on is broken, and the process exits non-zero straight after. `faults` is never empty; display each `message`, which names the path and the operator fix. `streak` is how many consecutive hub calls failed before the check ran, and is `0` for the check a shift makes at start. A shift that cannot reach the hub does NOT produce this — a hub outage or a rate-limit ban recovers on its own, and the shift keeps polling through it. This record means a human must change something on the host.
|
|
612
|
+
- `stalled`: `{ "type": "stalled", "lastPollAt": 1738000000000, "sinceMs": 45000 }` — the poll loop has made no progress (no hub call settled, no cycle completed) for longer than the stall threshold (three poll intervals plus the per-call hub deadline, 45 s at the defaults). It is written by a watchdog timer that does not depend on the loop, so it is the one record a hung loop cannot suppress. `lastPollAt` is the last completed cycle (epoch ms, `null` if none completed since start) and `sinceMs` is how long the loop has made no progress. A hub that stops answering does not produce this: every poll-loop hub call settles within its deadline and is reported as `hub-error` records and the `wedged` streak; `stalled` is the backstop for a hang that is not under a deadline (local dispatch, the filesystem, a spawn, or any other non-hub await inside the iteration). **Edge-triggered**: one record per stall episode, re-armed when a cycle completes. A hub-instructed `Retry-After` pause is not a stall and produces none. Compare `last_poll_at` on `shift status` to confirm.
|
|
593
613
|
|
|
594
614
|
Every event above is also appended to `<log-dir>/shift.log` as JSON Lines, which
|
|
595
|
-
is how they survive the daemon. **
|
|
615
|
+
is how they survive the daemon. **Six further record types exist in
|
|
596
616
|
`shift.log` and are never delivered over the socket** — `parked`, `capacity`,
|
|
597
|
-
`hub-error`, `event-queue-overflow`,
|
|
617
|
+
`hub-error`, `event-queue-overflow`, `low-disk`, and `heartbeat`.
|
|
598
618
|
|
|
599
619
|
An idle `shift next` must block until there is work to report, and each of those
|
|
600
|
-
|
|
620
|
+
six would satisfy it with news that no work moved: a startup record, a report
|
|
601
621
|
that the shift is full, a failed call to the hub, a record about the socket
|
|
602
|
-
queue itself,
|
|
622
|
+
queue itself, a report that the disk is too full to start anything, or the
|
|
623
|
+
periodic heartbeat (every five minutes) that says only that the daemon is still
|
|
624
|
+
alive and how many poll cycles it has completed. `parked` and `capacity` are also redundant on the wire — the
|
|
603
625
|
response above already carries live `cap`, `free`, and `running`, which is
|
|
604
626
|
exactly what a `capacity` record restates. `hub-error` is level-triggered at the
|
|
605
627
|
poll interval, so an unreachable hub would otherwise fill the 1000-slot socket
|
|
606
628
|
FIFO and evict the `dispatched`, `failed` and `reaped` records a parked client
|
|
607
629
|
is actually waiting for. In `shift.log`, which is append-only and unbounded, all
|
|
608
|
-
|
|
609
|
-
saturated one,
|
|
610
|
-
|
|
630
|
+
of them are load-bearing: they are the only way to tell an idle shift from a
|
|
631
|
+
saturated one, the only record of every failed hub call, and — through
|
|
632
|
+
`heartbeat` — the only way to tell a daemon that is alive but no longer polling
|
|
633
|
+
from one that is merely quiet. See [`docs/shift-logs.md`](shift-logs.md).
|
|
611
634
|
|
|
612
635
|
`gate` is a reserved protocol shape for a future local representation of a
|
|
613
636
|
pending hub gate. Production code does not construct live hub gate events yet;
|
|
@@ -633,13 +656,18 @@ owenloop shift status [--state-dir <p>]
|
|
|
633
656
|
With a daemon, status exits `0` and prints:
|
|
634
657
|
|
|
635
658
|
```json
|
|
636
|
-
{ "name": "host/project#abc123", "serve_crews": ["alpha"], "cap": 3, "free": 3, "running": 0, "agent_ceiling": 2, "attended_at": null, "started_at": 1738000000000 }
|
|
659
|
+
{ "name": "host/project#abc123", "serve_crews": ["alpha"], "cap": 3, "free": 3, "running": 0, "agent_ceiling": 2, "attended_at": null, "started_at": 1738000000000, "last_poll_at": 1738000005000, "cycles_completed": 1 }
|
|
637
660
|
```
|
|
638
661
|
|
|
639
662
|
`agent_ceiling` is the effective maximum number of `agent-run` children after
|
|
640
663
|
applying `--max-agents`, `--cap`, and `--exec-reserve`; command and exec work
|
|
641
664
|
may still use any free slot inside `cap`. `attended_at` remains `null` until the
|
|
642
|
-
first accepted `next` request.
|
|
665
|
+
first accepted `next` request. `last_poll_at` (epoch ms) and `cycles_completed`
|
|
666
|
+
are the liveness facts: they move only when the poll loop completes a full
|
|
667
|
+
cycle, so a daemon that still answers `status` but whose `last_poll_at` is
|
|
668
|
+
minutes old and whose `cycles_completed` no longer climbs has a loop that is
|
|
669
|
+
hung or wedged, not merely idle. `last_poll_at` is `null` until the first cycle
|
|
670
|
+
completes. Without a daemon, status is still a successful question: it exits `0` and prints
|
|
643
671
|
`{ "status": "no daemon", "socket": "<path>" }`. Invalid arguments exit `2`;
|
|
644
672
|
other client/runtime errors exit `1`.
|
|
645
673
|
|
package/docs/shift-logs.md
CHANGED
|
@@ -219,12 +219,14 @@ Two properties an uploader can rely on:
|
|
|
219
219
|
| `low-disk` | `path`, `freeBytes`, `floorBytes` | the volume holding the work root is below the floor, so the shift declined to start new work. It keeps polling and resumes by itself when space returns; children already running are left alone. **Edge-triggered** (one record per low-disk episode) and **file-only** — see below |
|
|
220
220
|
| `ended` | — | an operator ran `owenloop shift end` |
|
|
221
221
|
| `wedged` | `faults[]` (`kind`, `role`, `path`, `message`), `streak` | the shift stopped because the **machine** it runs on is broken, and the process exited non-zero — see below |
|
|
222
|
+
| `stalled` | `lastPollAt`, `sinceMs` | the poll loop has made no progress — no hub call settled, no cycle completed — for longer than the stall threshold; written by a watchdog timer the loop cannot block. **Edge-triggered** (one record per stall episode, re-armed on the next completed cycle) — see below |
|
|
223
|
+
| `heartbeat` | `cyclesCompleted`, `lastPollAt`, `hubFailureStreak`, `stalled` | periodic proof of life (every five minutes by default) carrying the loop's counters, so a silent file can be told apart from a silent daemon. **File-only** — see below |
|
|
222
224
|
| `gate` | optional `workflow`, `run`, `name`, `question` | reserved; not emitted today |
|
|
223
225
|
|
|
224
226
|
`kind` is `'exec' | 'agent-run'`.
|
|
225
227
|
|
|
226
|
-
**
|
|
227
|
-
`event-queue-overflow`,
|
|
228
|
+
**Six of these are file-only.** `parked`, `capacity`, `hub-error`,
|
|
229
|
+
`event-queue-overflow`, `low-disk`, and `heartbeat` are written to `shift.log` and are never
|
|
228
230
|
delivered over the daemon's Unix socket to `owenloop shift next`. Every other type goes to
|
|
229
231
|
both. The routing rule is `FILE_ONLY_EVENTS` in
|
|
230
232
|
`packages/work/src/shift/runtime.ts`, applied at the one point where an event
|
|
@@ -236,7 +238,9 @@ The split is not importance. It is what each consumer is:
|
|
|
236
238
|
`failed`, `order-dropped`, `bundle-miss`, `ended` — tells a socket client
|
|
237
239
|
something it cannot otherwise learn. Both sinks get it. `wedged` joins them
|
|
238
240
|
for the same reason `ended` does: a client parked on a shift that is about to
|
|
239
|
-
exit is waiting for work that will never arrive.
|
|
241
|
+
exit is waiting for work that will never arrive. `stalled` joins them for the
|
|
242
|
+
same reason again: a client parked on a shift whose loop has hung is waiting
|
|
243
|
+
for work that will not arrive until someone restarts it.
|
|
240
244
|
- A record about the **shift's own condition** is redundant or harmful on the
|
|
241
245
|
socket, and load-bearing in the file.
|
|
242
246
|
|
|
@@ -303,7 +307,7 @@ as a refusal.
|
|
|
303
307
|
|
|
304
308
|
Nothing is lost, because the file is the consumer with no envelope and no
|
|
305
309
|
context. `shift.log` has no per-response `cap`/`free`/`running`, so without
|
|
306
|
-
these
|
|
310
|
+
these six records a reader cannot tell an **idle** shift (no orders offered)
|
|
307
311
|
from a **saturated** one (orders offered, no slots) from a **stranded** one (hub
|
|
308
312
|
unreachable, so nothing was ever offered) from one **refusing work on a full
|
|
309
313
|
disk**, nor tell a quiet log from a lossy one.
|
|
@@ -415,6 +419,51 @@ check ran, and is `0` for the check every shift makes at start, before its first
|
|
|
415
419
|
hub call. It is evidence of what prompted the check, not the reason the shift
|
|
416
420
|
stopped; the `faults` are that.
|
|
417
421
|
|
|
422
|
+
### `stalled` and `heartbeat` — when the loop itself is the fault
|
|
423
|
+
|
|
424
|
+
`wedged` and `hub-error` both need a hub call to **fail**. A hub call that never
|
|
425
|
+
settles fails nothing: the loop awaits it forever, writes no line, makes no
|
|
426
|
+
further hub traffic, and the daemon still answers `shift status` over its
|
|
427
|
+
socket because that is served by a separate task. One shift was observed in
|
|
428
|
+
this state for hours, indistinguishable in the file from an idle one.
|
|
429
|
+
|
|
430
|
+
Two things close that gap. First, every hub call the loop makes — `wake`,
|
|
431
|
+
`whats_next`, `presence_ping` — now runs under a per-call deadline (30 s by
|
|
432
|
+
default), so a call that hangs is cut short and fails exactly like a call that
|
|
433
|
+
failed outright: it is logged, and recorded as a `hub-error` where that call
|
|
434
|
+
already reports one. Only a `wake` failure advances the `wedged` streak, as
|
|
435
|
+
before. Second, a **watchdog timer** that does not depend on the loop returning
|
|
436
|
+
checks when the loop last made progress — the later of the last hub call that
|
|
437
|
+
settled (success or failure) and the last completed cycle. If that is older
|
|
438
|
+
than the stall threshold (three poll intervals plus the per-call deadline, 45 s
|
|
439
|
+
at the defaults) it writes a `stalled` record: `lastPollAt` is the last
|
|
440
|
+
completed cycle (epoch ms, `null` if none has completed since start) and
|
|
441
|
+
`sinceMs` is how long the loop has made no progress. Progress is per call, not
|
|
442
|
+
per cycle, on purpose: one cycle serialises roster sync, presence, wake, the
|
|
443
|
+
inbox and one targeted `whats_next` per workflow, so a slow-but-alive hub with
|
|
444
|
+
a dozen workflows can legitimately take longer than the threshold per cycle
|
|
445
|
+
without being hung. A hung hub is therefore not what `stalled` reports: every
|
|
446
|
+
poll-loop hub call settles within its deadline, so a hub that stops answering
|
|
447
|
+
shows up as `hub-error` records and the `wedged` streak. `stalled` is the
|
|
448
|
+
backstop for a hang that is NOT under a deadline — local dispatch, the
|
|
449
|
+
filesystem, a spawn, or any other non-hub await inside the iteration. It is
|
|
450
|
+
**edge-triggered** like `capacity` and `low-disk` — one
|
|
451
|
+
record per stall episode, re-armed when a cycle next completes — and a
|
|
452
|
+
hub-instructed `Retry-After` pause is deliberately not a stall, since the shift
|
|
453
|
+
is idle on instruction rather than hung. It reaches both sinks, for the reason
|
|
454
|
+
`wedged` does.
|
|
455
|
+
|
|
456
|
+
`heartbeat` is the complement: proof of life on a fixed cadence (five minutes by
|
|
457
|
+
default) regardless of whether anything happened. It carries `cyclesCompleted`,
|
|
458
|
+
`lastPollAt`, `hubFailureStreak`, and `stalled`, so a reader of the file can
|
|
459
|
+
tell a daemon that is alive and idle (heartbeats arrive, `cyclesCompleted`
|
|
460
|
+
climbs) from one that is alive and stuck (heartbeats arrive, `cyclesCompleted`
|
|
461
|
+
frozen, `stalled: true`) from one that is dead (no heartbeats). It is
|
|
462
|
+
**file-only** for the `parked`/`capacity` reason: a parked `shift next` must
|
|
463
|
+
not be woken every five minutes by news that nothing happened. Both timers are
|
|
464
|
+
started only by a long-running shift — `--once` starts neither — and are
|
|
465
|
+
cleared when the shift stops.
|
|
466
|
+
|
|
418
467
|
### Line size
|
|
419
468
|
|
|
420
469
|
Each line is bounded at **512 KiB** (`MAX_RESPONSE_LINE_BYTES`) — the same
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "owenloop",
|
|
3
|
-
"version": "0.5.
|
|
3
|
+
"version": "0.5.31",
|
|
4
4
|
"description": "Deterministic rails for agentic workflows. Declare steps and dependencies; the engine guarantees order, redoes what changes invalidate, and stops what keeps failing — instead of hoping the agent follows through.",
|
|
5
5
|
"license": "Apache-2.0",
|
|
6
6
|
"author": "typicalday",
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "owenloop",
|
|
3
|
-
"version": "0.5.
|
|
3
|
+
"version": "0.5.31",
|
|
4
4
|
"description": "Claude Code driver for owenloop — cross-harness shift and conduct skills driven by the owenloop CLI, plus one hub control-plane MCP server backing an author skill and a fallback worker, for conducting and authoring workflows against your hosted owenloop server.",
|
|
5
5
|
"author": {
|
|
6
6
|
"name": "Typical Day"
|