Commit Graph
1090 Commits
Author SHA1 Message Date
aiongg 325d7b6529 User-local lease lifecycle: outage-aware expiry, time limit from the run, idempotent claims, tombstones, failed replies requeue, cool-off per login
The dashboard stamps its liveness every 5 s; job processes expire leases only while
it serves, and for missing heartbeats only after it served the whole idle window.
After a gap every lease's heartbeat, claim and run start move forward by the gap.
A token's drops count at most once in 10 minutes.

The time limit counts from the first running heartbeat; preparing may last 15 min.
A claim repeated with its request id within 5 minutes gets the same reply (kept in
the dashboard's memory; after a restart the orphaned lease goes back unused).
Ended and requeued leases keep their secret hash, so every lease route answers 410.
ok:false requeues with the next attempt and skips that login; a repeated accepted
submit is accepted again. Cool-off is per login, counts drops and 5 failures in a
row within 30 min, and is cleared by login.
2026-11-19 23:16:13 +00:00
aiongg aaa9fae512 Protocol 3: claim request ids and the fast lane, heartbeat phases, helper slots, busy count and last error, the per-user call limit 2026-11-19 23:02:45 +00:00
aiongg eeb45653f8 A take-fix Opus review whose fixes sit only on its code-wrong claims is kept with those fixes instead of rejected 2026-11-19 22:35:56 +00:00
aiongg 2f940fd344 A fix turn gets 3 turns, so a tool call the model makes despite the instruction is denied and answered instead of ending the turn 2026-11-19 22:31:52 +00:00
aiongg b54afbebbd In the user-local mode the reference lint's Haiku check waits up to 10 minutes for a helper instead of 30 seconds 2026-11-19 22:29:11 +00:00
aiongg 9eca9f0547 Merge branch 'ul-myrepos': each user picks the repos their helpers take calls from 2026-11-19 22:20:32 +00:00
aiongg c125437627 Reference search reads the built index and never builds one in the dashboard; builds of one index run one at a time; the query embedder uses one thread 2026-11-19 22:20:16 +00:00
aiongg 238d460eb9 My repos: each user picks the repos their user-local helpers take calls from, within the repos with the pipeline on 2026-11-19 22:18:52 +00:00
aiongg f0fa4f4c63 review.opus and cifix.patch allow 40 turns; a run that uses them all fails as a turn limit, by name 2026-11-19 22:17:14 +00:00
aiongg 750e0ea88c Claims accept a write permission confirmed in the last 10 minutes while GitHub cannot be asked, and say why a call was passed over 2026-11-19 22:12:56 +00:00
aiongg c3fb5fcfca autocraft remote token: a running agent token printed only into a pipe, for autocraft-task login on the same host 2026-11-19 21:16:49 +00:00
aiongg 59514ab617 Merge branch 'ul-cooloff' into ul-int; a cool-off is a helper readiness 2026-11-19 20:59:37 +00:00
aiongg 938d71fe68 User-local helpers keep a per-user plan-usage reserve: readers for Claude and Codex usage, pause and refusal holds, readiness reported to the server and counted by the planner's room 2026-11-19 20:56:00 +00:00
aiongg d78b32c860 User-local mode: six-hour cool-off for a token whose leases are lost three times in two hours, and a minimum effort per tier that a claim must declare 2026-11-19 20:46:11 +00:00
aiongg c43d0447f5 Merge branch 'ul-macos' into ul-int
# Conflicts:
#	src/remote/calls.ts
#	test/autocraft-task.test.ts
#	worker/autocraft-task.mjs
2026-11-19 19:07:51 +00:00
aiongg 8ade9b2a41 Reviews, pre-merge and post-merge reviews, duplicate comparisons and prepared decisions name the model that ran; full ids show as Claude or Codex Astra names 2026-11-19 19:02:01 +00:00
aiongg 15eaa6992b Remote runs record their tool uses: the helper reports them from the Claude stream or the Codex rollout, the server stores them on the run 2026-11-19 18:58:52 +00:00
aiongg 19f90ab5ae Merge branch 'ul-fix2' into ul-int 2026-11-19 18:55:18 +00:00
aiongg dd2a913c0c A helper's claim may name a path root: the server places the sandbox paths of the bundle's text, claude arguments and explorer prompt below it, and reads replies back without it 2026-11-19 18:46:27 +00:00
aiongg 9b0ad2ef77 The agent run keeps the model the harness reported, not the alias the helper claimed with 2026-11-19 18:44:14 +00:00
aiongg c33914a3ec Check requests of an earlier run are not served 2026-11-19 18:38:22 +00:00
aiongg 4da4f9f74f next_task counts only calls whose tier accepts the model the helper runs 2026-11-19 18:38:13 +00:00
aiongg 4d4229f31d The tarball response ends only when git exits cleanly; no helper online is a warning 2026-11-19 18:37:51 +00:00
aiongg 0d3a6ee7f8 Marker characters are removed from data and follow-up messages in remote prompts 2026-11-19 18:37:45 +00:00
aiongg 720393bb70 The cost report's per-user breakdown is for staff and follows the hidden-repo filter 2026-11-19 18:37:19 +00:00
aiongg d796cbfbbe Helper logins in agent runs, the now feed and the item trace go to staff only 2026-11-19 18:36:52 +00:00
aiongg 420acaec92 Reference passages are reserved before the search, so parallel calls stay within the allowance 2026-11-19 18:36:30 +00:00
aiongg a1f1e02468 Merge branch 'ul-codex' into ul-int 2026-11-19 18:27:07 +00:00
aiongg be01c2ba19 Codex adapter for the user-local helper
codex exec runs in the helper's bubblewrap sandbox with its own CODEX_HOME login (harness codex), its shell and
the other tools that reach past the task turned off, and reads through read-only file tools that the helper
serves to the run over localhost MCP, confined to the mounts. One Codex call at a time per helper, so refreshes
of its login never race. Codex has no tier defaults: the config names its models.
2026-11-19 18:26:29 +00:00
aiongg ccc44ae06b Merge branch 'ul-tools' into ul-int
# Conflicts:
#	src/dash/mcp/index.ts
#	src/remote/calls.ts
2026-11-19 18:09:33 +00:00
aiongg 96c47fddf3 Authors and co-authors give way on judging calls only while another helper that could take the call is asking for work
An author's claim skips a judging call when another login's helper made an empty claim within the idle window that covered
the call's tier and repos, fits the role rules, has cached write permission and authored nothing in the call. Otherwise the
author's helper gets it. The same rule applies to the next_task count. Empty claims record their tiers and repos.
2026-11-19 18:08:27 +00:00
aiongg 7fa065f41e Per-lease tools of remote calls: item, reference and check tools over HTTP and MCP, the tarball and helper routes, and the helper's tool command 2026-11-19 18:07:11 +00:00
aiongg 57c86871f5 Merge branch 'ul-ui' into ul-int 2026-11-19 18:05:45 +00:00
aiongg c931f23d83 Remote calls: capped union queue time, no raw replies kept, leases bound to a live running token, co-author exclusion
The runner sums the time in which any of a job's calls waited for a helper and leaves at most 12 hours of it out of the
deadline and no-progress stop. A done reply is cleared once the caller reads it. A lease records its token and ends
(returned to the queue, no attempt used) when the token may no longer run tasks or its login is no longer a maintainer
or operator. The sync stores each commit's authors as GitHub lists them, Co-authored-by trailers included, and judging
calls skip them; classify.haiku, security.verify and security.fixcheck judge too. Claims refuse helpers older than the
required version, skip calls the login released in the last 10 minutes and check write permission with the 1-minute
cache. Lease expiry is compare-and-set, and maxWaitMs counts queue time only.
2026-11-19 18:05:25 +00:00
aiongg 9b2945dbb3 User-local prompts carry harness wording that the claim fills; reference docs named as tools
A remote call's prompt marks the phrases that name Claude's tools; the bundle and the run's prompt fill them for
the claiming harness, with exactly the server's text for claude. Calls with the reference server get the
reference input flag and a sentence that names reference_search and reference_around instead of /reference.
2026-11-19 18:03:43 +00:00
aiongg 043863c4b7 Dashboard for the user-local inference mode: connect page setup, inference settings, run attribution, cost by user 2026-11-19 18:01:14 +00:00
aiongg ceb2bd4443 Session MCP tools next_task, task_status and my_work for tokens that can run tasks; remote-calls health check 2026-11-19 17:58:16 +00:00
aiongg 482f498563 Helper follows the merged wire format; the harness's cost reaches the run
The helper keeps a session when the bundle's keepSession says so, submits
usage per model id (explorer included) and the CLI's total cost, and gives a
lease back through the release route when the fault is its own: too old,
wrong platform, missing harness or token, file setup failure, or stopped by
the user. The work loop stops when it can take no call and pauses after a
release. Submit carries costUsd; the server stores it with the call's usage
and records it as the run's cost.
2026-11-19 17:54:45 +00:00
aiongg d219d5c3fe Merge branch 'ul-refsearch' into ul-int 2026-11-19 17:52:35 +00:00
aiongg 19bdcdcbae User-local inference mode, server side: remote calls that maintainers' helpers claim, run and submit
The inference setting (server | user-local) is set with `autocraft inference` or the operator-only
GET|PUT /api/settings/inference, which also holds the accepted model families per tier. In user-local mode
askClaude queues each Claude turn in remote_calls with a task bundle (prompt parts, claude arguments, checkout
and workspace mounts, repo docs inline, per-lease MCP servers) and polls for the result; follow-up turns go to
the helper that holds the session. Helpers claim with an agent token that may run tasks, in the model queue's
order filtered by tier, repo, write permission and author exclusion, and use a per-call lease secret for the
bundle, heartbeat, release and submit routes. Lost leases return to the queue up to the prompt's attempts.
Queue time does not count toward a job's deadline or no-progress stop, and accounts are not consulted.
2026-11-19 17:51:55 +00:00
aiongg ab5ee9fa04 Reference search compares a name/size/mtime fingerprint instead of hashing contents on each call 2026-11-19 17:34:41 +00:00
aiongg 4225f60fe8 Hybrid search over a repo's reference text pages, indexed outside the reference folder 2026-11-19 17:34:13 +00:00
aiongg f1377e1433 Name the per-server task MCP routes 2026-11-19 17:32:40 +00:00
aiongg 919f2fb8b6 Wire format for the user-local inference mode 2026-11-19 17:32:07 +00:00
aiongg d8e45b5d48 The runner reconciles one repo a day off its tick, giving way to the minute sync 2026-11-17 17:03:32 +00:00
aiongg 333b30e785 Account labels on the Live board and in the Now feed are for operators only 2026-11-17 15:39:10 +00:00
aiongg e66daa6f4d Live board: lanes keep their place, open integration batches as lanes, the latest feed events under it; viewers see no account labels; merge queue wait text stays visible
- Lanes keep the order they first arrived in; a lane the server stops sending stays as ended until the reader returns to the tab. Quiet steps are left out before the queued LIMIT.
- Open integration batches are lanes, with the build of their combined tree.
- The latest five finished Now feed events sit under the board.
- Account labels on model calls are for staff only, and a lone call's progress line is blanked when it names a hidden repo, on the board and in the Now feed.
- The merge queue row shows its wait label, conflict note and operator approval as text.
- The live query and the item status list follow their topics; the board does not poll.
- Workable tables switch to a wider layout only with room to spare, so a page scrollbar cannot flip them back and forth.
- Security findings get their own icon; item icons name their kind and state without explanations; the board's reading tip is gone.
- The dashboard spec describes the Live board, the feed list under it and the item peek.
2026-11-17 15:37:56 +00:00
David Mann c41505954f Home opens with a Live board: running and queued work on one time axis, with what each job ran and how long it usually takes
GET /api/live gives running, queued and lately ended jobs (and calls or build tasks of their own) with their model calls and build steps, and the median and 90th percentile of earlier runs of the same work. The board draws them on a shared axis: solid done, pulsing running, dashed the expected rest, a red line where 90% of earlier runs were done. It replaces the Live feed section on Home.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 00:16:40 +01:00
David Mann fc017055df Items rows carry the author, the issues a PR closes and its merge checks; a peek endpoint gives one item in brief
GET /api/items/status rows add author, closes and checks (review, CI, Verify, mergeability, size, our build) for PRs the plan holds a queue view for, veiled as in the merge queue. GET /api/items/:repo/:n/peek returns an item's title, author, labels, status and closes for hover cards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 00:01:15 +01:00
aiongg 0bcc1f8d68 Unsent comments: a final CI fix comment waits on a person, the item bar and repost read the right rows, stored patches stay until their retries are used
- A take PR whose CI fix ended with a failed comment, or with its last unsent one, waits on a person.
- The item bar reads the newest row past unsent comments that are still retried, and counts a fix-up's unsent comments per source and carry-over. Reposting an ask reads the review's own fix-up rows.
- The CI fix keeps a stored patch while its unsent comments are not final and leaves the timing to the plan.
- A later round's patch that needs no lint task is stored when its comment is not sent.
2026-11-16 15:22:17 +00:00