tiny-harness

Appearance

Tasks: Remote sub-tasks hear remote state changes between turns ​

A DAG derived from design.md and testing-plan.md. Each task's _Test:_ names a row of the plan's matrix. Single repository; everything lands on PR #28.

There are nine tasks. Every task is test-first: write the named test, watch it fail, then make it pass.

Task list ​

  • [x] 1. Configuration: updates and follow_interval on RemoteAgentSpec
    • Add the fields and the >= 1 s validator, and update the settings schema snapshot.
    • Depends on: none
    • Requirements: R4.1, R4.5
    • Test: T1/T10 tests/unit/test_config_agents.py; T3 snapshot
  • [x] 2. RemoteAgent: settings plus push, get and subscribe wrappers
    • Constructor and connect take updates and follow_interval, and build_runtime passes them.
    • Add push_supported, set_push_config, delete_push_config, get_task and subscribe.
    • Depends on: 1
    • Requirements: R1.1, R3.1, R5.1
    • Test: T2 (exercised end-to-end); T3 API snapshot
  • [x] 3. Push primitives and PushReceiver
    • In service/a2a/push.py: push_url, token_digest, PushIn (in durable models), PushReceiver.receive with the refusal ordering.
    • Depends on: none
    • Requirements: R2.1, R2.2; abuse cases 1–5, 9
    • Test: T1 tests/unit/service/test_push_receiver.py
  • [x] 4. The push route in create_app, wired in running_harness
    • POST /a2a/push/{task_id} before the SDK routes, reading Content-Length first.
    • running_harness builds PushReceiver(client) and passes push_base_url to Activities.
    • Depends on: 3
    • Requirements: R2.1, abuse case 5
    • Test: T8 tests/security/test_remote_push.py (route-level cases)
  • [x] 5. remote_out and the new activities
    • Factor remote_out out of _remote_turn.
    • Add register_remote_push, get_remote_task, follow_remote_task and delete_remote_push, with the new ActivityNames and models.
    • Depends on: 2, 3
    • Requirements: R1.1–1.3, R1.5, R2.3, R2.5, R3.1–3.3, R5.1
    • Test: T1 tests/unit/service/test_remote_out.py; T2
  • [x] 6. RemoteTaskWorkflow: the watch, the remote_push update and its validator, _park, _end
    • Depends on: 5
    • Requirements: R1.4, R2.2–2.4, R2.6, R2.8, R3.4, R4.2–4.4, R5.1–5.3
    • Test: T2 tests/integration/agents/test_remote_updates.py; T8
  • [x] 7. Parent child_done: note only on terminal or interrupted remote states
    • Depends on: none
    • Requirements: R2.7
    • Test: T2 (row moves without a model turn on WORKING)
  • [x] 8. Security tests for every abuse case
    • Include the token-absence check over workflow history, captured logs and the store.
    • Depends on: 4, 6
    • Requirements: abuse cases 1–9
    • Test: T8 tests/security/test_remote_push.py
  • [x] 9. Documentation
    • Update docs/capabilities/remote-agents.md and a2a-server.md (the endpoint), the configuration capability doc and the example config. Record them in evidence/documentation.md.
    • Depends on: 6
    • Requirements: all (capability docs)
    • Test: T12 markdownlint

Dependency graph (DAG) ​

Checkpoints ​

  • After 4: uv run pytest tests/unit tests/contract.
  • After 7: uv run pytest tests/integration -q.
  • After 9: uv run pre-commit run --all-files.