Fifth extraction in the resolved order -- the heaviest and most
store-coupled phase, deliberately placed after the facade (Gateway) and
the data seam (State Store) are already proven.
Scope, per the PRD's target-services table:
Execution Environment: internal/domain/workflow's dispatch paths
become the execution-dispatch contract, consuming State Store's
WorkflowStore contract (Phase 2).
Runner: internal/container, internal/agentexec, internal/worktree
become the isolated worktree/container execution service, consuming
the existing container contract.
Scheduler: internal/domain/scheduling + internal/infrastructure/
cronstore become the cron/ticker contract service.
Depends on Phase 4 (Messaging extraction) per the resolved extraction
order.
Note: internal/storerpc's NATS path to the archie-agent container is
explicitly flagged in question #4's research as an Execution-track
question ("Migration of the storerpc NATS path... under the PRD the
Execution/Runner processes are peers too, so the agent's store access
should eventually use the same contract") -- this phase owns deciding
when that moves, not the State Store phase.
Full reasoning: docs/inspiration/service-decomposition-open-questions-research-2026-09-05.md#1
(Phase 5 section).
Acceptance criteria
Execution Environment, Runner, and Scheduler stood up as separable
services (or a documented decision to combine any of them into one
binary, if scoping during implementation finds no benefit to a full
three-way split -- but the contracts must be independently
addressable).
Execution Environment consumes State Store's WorkflowStore contract
over gRPC rather than importing internal/store directly.
The archie-agent container's storerpc NATS access path is either
migrated onto the same contract-based access or an explicit documented
decision is recorded for why it stays on NATS for this phase.
internal/channels/telegram and other invariants unaffected by this
phase (container pull policy, NATS JetStream work-queue semantics from
AGENTS.md remain intact).
task check passes; a full task lifecycle (intake -> dispatch -> worktree
execution -> completion) is exercised end-to-end against the extracted
services.
Close reason
Not planned: scheduling and container execution are the daemon's dispatch responsibility, one service under 'services own contracts'. Agent containers already use the State Store gRPC contract. Reopen only for a concrete scaling, HA or separate-host need.
Imported from beads archie-core-8cda.7 · GitHub #744 · type task · priority P3 · created 2026-09-05 · closed 2026-10-03
Fifth extraction in the resolved order -- the heaviest and most
store-coupled phase, deliberately placed after the facade (Gateway) and
the data seam (State Store) are already proven.
Scope, per the PRD's target-services table:
- Execution Environment: internal/domain/workflow's dispatch paths
become the execution-dispatch contract, consuming State Store's
WorkflowStore contract (Phase 2).
- Runner: internal/container, internal/agentexec, internal/worktree
become the isolated worktree/container execution service, consuming
the existing container contract.
- Scheduler: internal/domain/scheduling + internal/infrastructure/
cronstore become the cron/ticker contract service.
Depends on Phase 4 (Messaging extraction) per the resolved extraction
order.
Note: internal/storerpc's NATS path to the archie-agent container is
explicitly flagged in question #4's research as an Execution-track
question ("Migration of the storerpc NATS path... under the PRD the
Execution/Runner processes are peers too, so the agent's store access
should eventually use the same contract") -- this phase owns deciding
when that moves, not the State Store phase.
Full reasoning: docs/inspiration/service-decomposition-open-questions-research-2026-09-05.md#1
(Phase 5 section).
### Acceptance criteria
- Execution Environment, Runner, and Scheduler stood up as separable
services (or a documented decision to combine any of them into one
binary, if scoping during implementation finds no benefit to a full
three-way split -- but the contracts must be independently
addressable).
- Execution Environment consumes State Store's WorkflowStore contract
over gRPC rather than importing internal/store directly.
- The archie-agent container's storerpc NATS access path is either
migrated onto the same contract-based access or an explicit documented
decision is recorded for why it stays on NATS for this phase.
- internal/channels/telegram and other invariants unaffected by this
phase (container pull policy, NATS JetStream work-queue semantics from
AGENTS.md remain intact).
- task check passes; a full task lifecycle (intake -> dispatch -> worktree
execution -> completion) is exercised end-to-end against the extracted
services.
### Close reason
Not planned: scheduling and container execution are the daemon's dispatch responsibility, one service under 'services own contracts'. Agent containers already use the State Store gRPC contract. Reopen only for a concrete scaling, HA or separate-host need.
---
_Imported from beads `archie-core-8cda.7` · GitHub #744 · type task · priority P3 · created 2026-09-05 · closed 2026-10-03_
sam
added the lowtask labels 2026-10-27 13:35:01 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Fifth extraction in the resolved order -- the heaviest and most
store-coupled phase, deliberately placed after the facade (Gateway) and
the data seam (State Store) are already proven.
Scope, per the PRD's target-services table:
become the execution-dispatch contract, consuming State Store's
WorkflowStore contract (Phase 2).
become the isolated worktree/container execution service, consuming
the existing container contract.
cronstore become the cron/ticker contract service.
Depends on Phase 4 (Messaging extraction) per the resolved extraction
order.
Note: internal/storerpc's NATS path to the archie-agent container is
explicitly flagged in question #4's research as an Execution-track
question ("Migration of the storerpc NATS path... under the PRD the
Execution/Runner processes are peers too, so the agent's store access
should eventually use the same contract") -- this phase owns deciding
when that moves, not the State Store phase.
Full reasoning: docs/inspiration/service-decomposition-open-questions-research-2026-09-05.md#1
(Phase 5 section).
Acceptance criteria
services (or a documented decision to combine any of them into one
binary, if scoping during implementation finds no benefit to a full
three-way split -- but the contracts must be independently
addressable).
over gRPC rather than importing internal/store directly.
migrated onto the same contract-based access or an explicit documented
decision is recorded for why it stays on NATS for this phase.
phase (container pull policy, NATS JetStream work-queue semantics from
AGENTS.md remain intact).
execution -> completion) is exercised end-to-end against the extracted
services.
Close reason
Not planned: scheduling and container execution are the daemon's dispatch responsibility, one service under 'services own contracts'. Agent containers already use the State Store gRPC contract. Reopen only for a concrete scaling, HA or separate-host need.
Imported from beads
archie-core-8cda.7· GitHub #744 · type task · priority P3 · created 2026-09-05 · closed 2026-10-03