Implementation Plan: Discipline Delivery Channel Matrix Contract (#4119)

Trecek · claude · web-flow · commit c063618ff97b · 2026-06-27T06:45:59.000Z
## Summary Create `tests/contracts/test_discipline_delivery_matrix.py` — a parametrized 8-case contract test suite that encodes the session-type × backend discipline-delivery channel matrix. The test exercises `build_interactive_cmd`, `build_skill_session_cmd`, `build_food_truck_cmd`, and `_read_full_sous_chef` composition paths to assert that the PRIMARY delivery channel is populated for each (SessionType × backend) combination, making future regressions CI-detectable. Also update `tests/contracts/AGENTS.md` with the new file entry. Closes #4044 ## Implementation Plan Plan file: `/home/talon/projects/autoskillit-runs/impl-20260626-224909-325309/.autoskillit/temp/make-plan/t5_p6_a1_wp1_discipline_delivery_matrix_plan_2026-06-26_230000.md` 🤖 Generated with [Claude Code](https://claude.com/claude-code) via AutoSkillit  ## Token Usage Summary | Step | Model | count | uncached | output | cache_read | peak_ctx | turns | cache_write | time | |------|-------|-------|----------|--------|------------|----------|-------|-------------|------| | plan* | opus[1m] | 1 | 2.5k | 25.1k | 2.3M | 130.1k | 57 | 135.7k | 16m 5s | | verify* | sonnet | 1 | 1.5k | 5.1k | 152.4k | 47.8k | 15 | 50.8k | 6m 53s | | implement* | MiniMax-M3 | 1 | 69.2k | 10.8k | 1.2M | 0 | 50 | 0 | 4m 36s | | audit_impl* | sonnet | 1 | 52 | 12.6k | 199.8k | 42.0k | 16 | 37.8k | 3m 38s | | prepare_pr* | MiniMax-M3 | 1 | 41.7k | 3.1k | 192.0k | 0 | 13 | 0 | 47s | | compose_pr* | MiniMax-M3 | 1 | 36.4k | 1.7k | 140.9k | 0 | 12 | 0 | 32s | | **Total** | | | 151.3k | 58.4k | 4.2M | 130.1k | | 224.4k | 32m 32s | \* *Step used a non-Anthropic provider; caching behavior may differ.* ## Token Efficiency | Step | LoC Changed | cache_read/LoC | cache_write/LoC | output/LoC | |------|-------------|----------------|-----------------|------------| | plan | 0 | — | — | — | | verify | 0 | — | — | — | | implement | 178 | 6686.2 | 0.0 | 60.7 | | audit_impl | 0 | — | — | — | | prepare_pr | 0 | — | — | — | | compose_pr | 0 | — | — | — | | **Total** | **178** | 23701.3 | 1260.6 | 327.9 | ## Model Usage Breakdown | Model | steps | uncached | output | cache_read | cache_write | time | |-------|-------|----------|--------|------------|-------------|------| | opus[1m] | 1 | 2.5k | 25.1k | 2.3M | 135.7k | 16m 5s | | sonnet | 2 | 1.6k | 17.7k | 352.2k | 88.6k | 10m 31s | | MiniMax-M3 | 3 | 147.2k | 15.6k | 1.5M | 0 | 5m 55s | --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
diff --git a/tests/contracts/AGENTS.md b/tests/contracts/AGENTS.md
@@ -26,6 +26,7 @@ Protocol satisfaction, package gateway, and skill contract compliance tests.
 | `test_config_field_coverage.py` | REQ-CONFIG-001: every sub-config dataclass field must be referenced in from_dynaconf |
 | `test_core_public_api_surface.py` | Validates that every symbol in autoskillit.core.__all__ is importable via the public gateway |
 | `test_diagnose_ci_steps.py` | Contract tests for diagnose-ci SKILL.md step numbering and cross-reference integrity |
+| `test_discipline_delivery_matrix.py` | Contract: session-type x backend discipline delivery channel matrix — interactive primary channel, headless prompt + completion marker, skill prompt, and sous-chef composition |
 | `test_docstring_skill_prefix.py` | Contract: source files must not use /autoskillit: prefix for skills_extended skills |
 | `test_environment_setup_design_contracts.py` | Contract tests verifying the environment-setup skill design doc completeness |
 | `test_ephemeral_skill_namespace.py` | Contract: ephemeral SKILL.md bodies use the correct namespace for the session delivery channel |
diff --git a/tests/contracts/test_discipline_delivery_matrix.py b/tests/contracts/test_discipline_delivery_matrix.py
@@ -0,0 +1,158 @@
+"""Discipline delivery channel matrix contract tests.
+
+Session-type x backend parametrized tests asserting that each combination's
+PRIMARY delivery channel is populated via the correct builder.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from autoskillit.core import (
+    SESSION_TYPE_ENV_VAR,
+    SESSION_TYPE_FLEET,
+    SESSION_TYPE_ORCHESTRATOR,
+    SESSION_TYPE_SKILL,
+    DirectInstall,
+)
+from autoskillit.execution.backends.claude import ClaudeCodeBackend
+from autoskillit.execution.backends.codex import CodexBackend  # noqa: F401
+
+pytestmark = [pytest.mark.layer("contracts"), pytest.mark.small]
+
+
+def _assert_interactive_primary_channel(backend, spec) -> None:
+    """Assert the interactive primary delivery channel is populated."""
+    if isinstance(backend, ClaudeCodeBackend):
+        assert "--append-system-prompt" in spec.cmd
+    else:
+        assert any("developer_instructions=" in arg for arg in spec.cmd)
+
+
+class TestFleetInteractive:
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_primary_channel_populated(self, backend) -> None:
+        spec = backend.build_interactive_cmd(
+            system_prompt="Fleet discipline prompt",
+            env_extras={SESSION_TYPE_ENV_VAR: SESSION_TYPE_FLEET},
+        )
+        _assert_interactive_primary_channel(backend, spec)
+
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_session_type_fleet_in_env(self, backend) -> None:
+        spec = backend.build_interactive_cmd(
+            system_prompt="Fleet discipline prompt",
+            env_extras={SESSION_TYPE_ENV_VAR: SESSION_TYPE_FLEET},
+        )
+        assert spec.env.get(SESSION_TYPE_ENV_VAR) == SESSION_TYPE_FLEET
+
+
+class TestOrchestratorInteractive:
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_primary_channel_populated(self, backend) -> None:
+        spec = backend.build_interactive_cmd(
+            system_prompt="Orchestrator discipline prompt",
+        )
+        _assert_interactive_primary_channel(backend, spec)
+
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_no_session_type_assertion(self, backend) -> None:
+        spec = backend.build_interactive_cmd(
+            system_prompt="Orchestrator discipline prompt",
+        )
+        assert not spec.env.get(SESSION_TYPE_ENV_VAR, "")
+
+
+class TestOrchestratorHeadless:
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_orchestrator_prompt_non_empty(self, backend) -> None:
+        spec = backend.build_food_truck_cmd(
+            orchestrator_prompt="Run the pipeline",
+            plugin_source=DirectInstall(plugin_dir=Path("/tmp")),
+            cwd="/tmp",
+            completion_marker="%%DONE%%",
+        )
+        assert any("Run the pipeline" in arg for arg in spec.cmd)
+        assert any("%%DONE%%" in arg for arg in spec.cmd)
+
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_session_type_orchestrator_in_env(self, backend) -> None:
+        spec = backend.build_food_truck_cmd(
+            orchestrator_prompt="Run the pipeline",
+            plugin_source=DirectInstall(plugin_dir=Path("/tmp")),
+            cwd="/tmp",
+            completion_marker="%%DONE%%",
+        )
+        assert spec.env.get(SESSION_TYPE_ENV_VAR) == SESSION_TYPE_ORCHESTRATOR
+
+
+class TestSkillSession:
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_positional_prompt_non_empty(self, backend) -> None:
+        spec = backend.build_skill_session_cmd("/investigate foo", "/tmp")
+        assert any("investigate" in arg for arg in spec.cmd)
+
+    @pytest.mark.parametrize(
+        "backend",
+        [ClaudeCodeBackend(), CodexBackend()],
+        ids=["claude-code", "codex"],
+    )
+    def test_session_type_skill_in_env(self, backend) -> None:
+        spec = backend.build_skill_session_cmd("/investigate foo", "/tmp")
+        assert spec.env.get(SESSION_TYPE_ENV_VAR) == SESSION_TYPE_SKILL
+
+
+class TestSousChefDelivery:
+    def test_sous_chef_in_orchestrator_prompt(self) -> None:
+        from autoskillit.cli._prompts import _build_orchestrator_prompt, _read_full_sous_chef
+
+        sous_chef = _read_full_sous_chef()
+        assert sous_chef, "_read_full_sous_chef must return non-empty content"
+        prompt = _build_orchestrator_prompt("test-recipe", "mcp__autoskillit__")
+        assert sous_chef[:80] in prompt
+
+    def test_sous_chef_in_open_kitchen_prompt(self) -> None:
+        from autoskillit.cli._prompts import _build_open_kitchen_prompt, _read_full_sous_chef
+
+        sous_chef = _read_full_sous_chef()
+        assert sous_chef, "_read_full_sous_chef must return non-empty content"
+        prompt = _build_open_kitchen_prompt("mcp__autoskillit__")
+        assert sous_chef[:80] in prompt
+
+    def test_sous_chef_not_in_fleet_dispatch_prompt(self) -> None:
+        from autoskillit.cli._prompts import _build_fleet_dispatch_prompt, _read_full_sous_chef
+
+        sous_chef = _read_full_sous_chef()
+        assert sous_chef, "_read_full_sous_chef must return non-empty content"
+        prompt = _build_fleet_dispatch_prompt("mcp__autoskillit__")
+        assert sous_chef[:80] not in prompt
