<!-- Authored per the the-loop:writing skill. -->

Tasks: a released work item is launched with the arguments the operator declared

The last spec artifact (bugfix → design → testing plan → tasks). A DAG of implementation tasks derived from the design and the testing plan.

Task list

• [x] 1. One resolver for a harness's launch arguments

  • cli/the_loop/modelchoice.py: launch_args(config, routing_args=None) over harness_args, covering every harness named in either home; the conflict logged once and reported as a config_findings warning
  • Depends on: none
  • Requirements: R1.1, R1.3, R1.4
  • Test: T1 — pytest cli/tests/test_modelchoice.py -k launch_args or both_homes (red→green)

• [x] 2. The regression scenarios, red

  • cli/tests/test_spawn_gate_integration.py: T2 (the report's check, through poller.daemon._build_dispatcher), T3 (deprecated home), T4 (the frozen model reaches the post-gate spawn and the record), T6 (the events carry the argv), T9 (a forged model buys nothing on that path)
  • Run them and record the failures before task 3
  • Depends on: none
  • Requirements: R4.1, R4.2
  • Test: T2, T3, T4, T6, T9 (red)

• [x] 3. Every adapter builder takes the resolver; the choice path starts from the adapter

  • cli/the_loop/poller/daemon.py, cli/the_loop/webhook/daemon.py, Dispatcher.reload: build_adapters(launch_args(cli_config, routing.harness_args), …)
  • cli/the_loop/webhook/dispatcher.py: _adapter_for(work_item, harness, cwd="") with effective_args(adapter.extra_args, …)
  • cli/tests/test_dispatcher_choice.py: the fixture builds its adapter through launch_args, so the existing suite runs against the real wiring
  • Depends on: 1, 2
  • Requirements: R1.1, R1.2
  • Test: T2, T3 (green), T5 — pytest cli/tests/test_dispatcher_choice.py

• [x] 4. The spawn resolves after the gate, and the events carry the argv

  • cli/the_loop/webhook/dispatcher.py: _spawn_for checks the adapter exists, prepares the workspace, asks on_arm, then resolves _adapter_for(…, cwd); _spawn_tmux takes the resolved (model, effort); session.spawned, session.respawned and session.pr_spawned gain the argv fields; the catalogue entries in eventlog.EVENT_TYPES say so
  • Depends on: 3
  • Requirements: R1.5, R2.1, R2.2
  • Test: T4, T6, T9 (green)

• [x] 5. The other readers: standing sessions and the probe

  • cli/the_loop/standing.py and cli/the_loop/core/standing.py: inherit launch_args(config).get(harness) when harnessArgs is omitted
  • cli/the_loop/commands/models_cmd.py: build_adapters(launch_args(config)) for the probe and the findings
  • cli/the_loop/api/routes.py and docs/api-specs/openapi/the-loop.v1.yaml: the harnessArgs description names both homes
  • Depends on: 1
  • Requirements: R3.1, R3.2
  • Test: T7 — pytest cli/tests/test_standing.py cli/tests/test_standing_integration.py -k inherit; T8 — pytest cli/tests/test_models_cmd.py -k launch (red→green)

• [x] 6. Documentation and capability docs

  • docs/config/cli/harnesses-options.md, docs/config/cli/routing-options.md, docs/config/cli/standing-sessions-options.md: harnesses[].args is read by every spawn path, the standing sessions and the probe; routing.harnessArgs is deprecated and loses to it, with the conflict reported
  • docs/capabilities/interactive-sessions.md, docs/capabilities/standing-sessions.md, docs/capabilities/observability.md: behaviour clauses and history rows
  • skills/the-loop/reference/automation.md, docs/capabilities/webhook-triggers.md: the bypass-acceptance sentence names the launch arguments rather than harnessArgs
  • evidence/documentation.md
  • Depends on: 3, 4, 5
  • Requirements: — (ready-to-ship gate)
  • Test: T11

• [x] 7. Verification, reviews and evidence

  • Run every activity of testing-plan.md, fill its results table; write evidence/self-review.md, evidence/critic-review.md, evidence/security-review.md, evidence/final-validation.md, evidence/pull-requests.md
  • Depends on: 6
  • Requirements: R4
  • Test: T10, T11

Dependency graph (DAG)

Checkpoints

After task 2: the five new scenarios fail for the reasons bugfix.md names (empty argv; no model on the post-gate spawn). After task 4: they pass, and the choice suite passes against the rewired fixture. After task 6: make check (lint, format, typecheck, config validation, full suite) is green. Then verification executes the plan and the record is written in testing-plan.md § Verification results.