nsaspy/zara
1
0
Fork 0
Code Issues 95 Pull requests 18 Projects Releases Packages Wiki Activity Actions

feat(tasks): long-horizon agent tasks (persistent, resumable, bounded) #268

Closed
nsaspy wants to merge 0 commits from feature/long-horizon-tasks into master
pull from: feature/long-horizon-tasks
merge into: nsaspy:master
Conversation 0 Commits - Files changed -
nsaspy commented 2026-09-04 22:44:03 +00:00
Owner
Copy link

RAGE rebase recovery — current state

This PR was explicitly rebased/recovered onto Zara master after the original branch fell 138 commits behind.

  • pre-rebase head preserved at backup/feature-long-horizon-tasks-pre-rebase-20260902 (e406f105dcfdacf1f33b38bcc57b48f44e32e602)
  • recovery lineage starts from master 9f3b779b876df360001b12d0100698f334beb8b8
  • task engine, persistence, tools, tests, docs, database locking fix, and additive task runtime events are transplanted
  • zara/agent/__init__.py is ported onto the current #260 AgentLoopAdviceRegistry path: task-owned history/context stays external and does not overwrite normal conversation history
  • zara/runtime/backend.py is ported onto the current daemon router/memory/stream path: task-owned turns bypass ordinary top-level intent routing and normal durable chat-memory persistence while retaining the canonical agent/tool/advice loop
  • zara/config.py is ported as a narrow three-way delta: [tasks] remains default-off with strict bounded validation and get_tasks_config()
  • zara/runtime/host.py owns the single task runner on the canonical runtime thread; task steps use coordinator-issued turn ids and the existing backend/tool approval path; restart/shutdown stop the task runner before the backend; the backend publisher is restored when task observation stops
  • current hook, voice, latency, API-service, and plugin runtime ownership from master was preserved

Exact-head recovery evidence

CI #812 on fdff7ef91a13a57199c1f6637b77006a8fc73907 produced a deterministic RED: four t/test_task_host_wiring.py failures all showed RuntimeHost.start() leaving task_runner=None. The cause was not the task config or runner: RuntimeHost correctly requested the canonical backend's principal identity for principal-scoped task persistence, but the modern RuntimeBackend contract exposed no principal identity.

Commit b7cc0a85a6d84352c668d8e892e01b2b5f5e0e79 fixes that narrowly by adding a typed RuntimeBackend.principal_id contract, exposing the bound AgentManager.principal.principal_id from LangGraphRuntimeBackend, and delegating it through AgentRuntimeBackend. No hard-coded local principal and no RuntimeHost reach-through into private manager state.

Fresh exact-head CI #816 is the only merge authority for this new head. Shared-mic Arch/Ubuntu and Android skeleton gates are already green; the main scripts/test-all.sh / nix flake check / package gate must finish green before merge.

Do not merge while any exact-head gate is pending, failing, skipped, cancelled, or inconclusive.

Feature intent

Implements persistent, resumable, bounded long-horizon agent tasks without a second agent loop. Task steps remain ordinary conversation turns through the canonical runtime; approvals use the shared approval lifecycle; cancellation uses the existing coordinator/backend chain; [tasks].enabled remains default-off.

Core task behavior includes principal-scoped persistent task records, bounded step logs, startup recovery to interrupted, explicit resume, step/wall-clock budgets, bounded concurrency, additive task events, and task tools routed through the existing tool registry.

## RAGE rebase recovery — current state This PR was explicitly rebased/recovered onto Zara `master` after the original branch fell 138 commits behind. - pre-rebase head preserved at `backup/feature-long-horizon-tasks-pre-rebase-20260902` (`e406f105dcfdacf1f33b38bcc57b48f44e32e602`) - recovery lineage starts from master `9f3b779b876df360001b12d0100698f334beb8b8` - task engine, persistence, tools, tests, docs, database locking fix, and additive task runtime events are transplanted - `zara/agent/__init__.py` is ported onto the current #260 `AgentLoopAdviceRegistry` path: task-owned history/context stays external and does not overwrite normal conversation history - `zara/runtime/backend.py` is ported onto the current daemon router/memory/stream path: task-owned turns bypass ordinary top-level intent routing and normal durable chat-memory persistence while retaining the canonical agent/tool/advice loop - `zara/config.py` is ported as a narrow three-way delta: `[tasks]` remains default-off with strict bounded validation and `get_tasks_config()` - `zara/runtime/host.py` owns the single task runner on the canonical runtime thread; task steps use coordinator-issued turn ids and the existing backend/tool approval path; restart/shutdown stop the task runner before the backend; the backend publisher is restored when task observation stops - current hook, voice, latency, API-service, and plugin runtime ownership from master was preserved ### Exact-head recovery evidence CI #812 on `fdff7ef91a13a57199c1f6637b77006a8fc73907` produced a deterministic RED: four `t/test_task_host_wiring.py` failures all showed `RuntimeHost.start()` leaving `task_runner=None`. The cause was not the task config or runner: `RuntimeHost` correctly requested the canonical backend's principal identity for principal-scoped task persistence, but the modern `RuntimeBackend` contract exposed no principal identity. Commit `b7cc0a85a6d84352c668d8e892e01b2b5f5e0e79` fixes that narrowly by adding a typed `RuntimeBackend.principal_id` contract, exposing the bound `AgentManager.principal.principal_id` from `LangGraphRuntimeBackend`, and delegating it through `AgentRuntimeBackend`. No hard-coded local principal and no RuntimeHost reach-through into private manager state. Fresh exact-head CI #816 is the only merge authority for this new head. Shared-mic Arch/Ubuntu and Android skeleton gates are already green; the main `scripts/test-all.sh` / `nix flake check` / package gate must finish green before merge. **Do not merge while any exact-head gate is pending, failing, skipped, cancelled, or inconclusive.** ## Feature intent Implements persistent, resumable, bounded long-horizon agent tasks without a second agent loop. Task steps remain ordinary conversation turns through the canonical runtime; approvals use the shared approval lifecycle; cancellation uses the existing coordinator/backend chain; `[tasks].enabled` remains default-off. Core task behavior includes principal-scoped persistent task records, bounded step logs, startup recovery to `interrupted`, explicit resume, step/wall-clock budgets, bounded concurrency, additive task events, and task tools routed through the existing tool registry.
nsaspy closed this pull request 2026-09-04 23:09:33 +00:00
Some checks failed
CI / test (pull_request) Failing after 4s
Details
CI / android skeleton gate (pull_request) Failing after 5s
Details
CI / shared mic / Ubuntu 24.04 (pull_request) Failing after 9m20s
Details
CI / shared mic / Arch Linux (pull_request) Failing after 10m21s
Details

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
bug
Something isn't working

  
documentation
Improvements or additions to documentation

  
duplicate
This issue or pull request already exists

  
enhancement
New feature or request

  
good first issue
Good for newcomers

  
help wanted
Extra attention is needed

  
invalid
This doesn't seem right

  
question
Further information is requested

  
wontfix
This will not be worked on

No labels
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
nsaspy/zara!268
No description provided.