My Agent Core Design - Completion Audit

Completion Audit

This file records the current completion evidence for doc/01 through doc/22. It is intentionally evidence-oriented: each row points to implementation areas, tests, and final validation commands that were checked against the current worktree.

Scope

• Source of truth: doc/01-goal.md through doc/22-codegen-plan.md.
• Highest-priority contract: doc/21-codegen-contract.md.
• Test matrix: doc/22-codegen-plan.md sections A through O.
• Provider test policy: provider tests use ollama / local Ollama. Tests do not exercise OpenAI or Claude providers.

Requirement Evidence

Area	Docs	Implementation Evidence	Test Evidence
Public SDK, package, CLI, errors	01, 02, 03, 21, 22-A	pyproject.toml, src/agent_core/__init__.py, src/agent_core/api/runtime.py, src/agent_core/api/handles.py, src/agent_core/errors/codes.py, src/agent_core/cli.py	tests/unit/test_public_api.py, tests/unit/test_assets_and_cli.py
Config and path layout	03, 18, 21, 22-B	src/agent_core/config/*, src/agent_core/api/runtime.py, src/agent_core/assets/templates/config_default.toml	tests/unit/test_config_loader.py, tests/unit/test_assets_and_cli.py, tests/integration/test_plan_task_tools.py
Provider abstraction and Ollama	02, 05, 19, 21, 22-C	src/agent_core/providers/base.py, src/agent_core/providers/ollama.py, src/agent_core/providers/registry.py, src/agent_core/providers/tool_mapping.py	tests/unit/test_provider_mapping.py, tests/integration/test_runtime_loop.py, tests/e2e/test_ollama_gemma4.py, provider forbidden scan
Runtime loop, event stream, queue, cancel	03, 04, 07, 20, 22-D	src/agent_core/api/runtime.py, src/agent_core/loop/*, src/agent_core/events/*, src/agent_core/runtime/*	tests/integration/test_runtime_loop.py, tests/integration/test_run_queue_cancel.py, tests/integration/test_orchestrator_mode.py
SQLite, replay, artifacts, cleanup	06, 07, 09, 20, 22-E	src/agent_core/storage/*, src/agent_core/api/replay.py, src/agent_core/artifacts/*, src/agent_core/api/runtime.py	tests/integration/test_replay_cleanup.py, tests/unit/test_artifacts.py
Context, instructions, skills	06, 11, 21, 22-F	src/agent_core/context/*, src/agent_core/tools/internal.py, src/agent_core/assets/prompts/system/*	tests/unit/test_instruction_skill_memory.py, tests/integration/test_runtime_loop.py
Code tools and output boundaries	12, 19, 21, 22-G	src/agent_core/tools/builtin_code.py, src/agent_core/tools/output_limits.py, src/agent_core/tools/registry.py	tests/unit/test_code_tools.py, tests/integration/test_runtime_loop.py, tests/integration/test_replay_cleanup.py
Permissions, hooks, declarative tools	12, 14, 18, 20, 22-H	src/agent_core/permissions/*, src/agent_core/hooks/*, src/agent_core/tools/declarative.py, src/agent_core/tools/registry.py	tests/unit/test_code_tools.py, tests/unit/test_hooks_mcp_declarative.py, tests/unit/test_assets_and_cli.py
Agent definitions and multi-agent	15, 19, 21, 22-I	src/agent_core/agents/*, src/agent_core/tools/agent_tools.py, src/agent_core/assets/agents/*, src/agent_core/api/runtime.py	tests/unit/test_agent_definitions.py, tests/integration/test_multi_agent.py
Worker and orchestrator	15, 17, 19, 22-J	src/agent_core/agents/workers.py, src/agent_core/tools/agent_tools.py, src/agent_core/api/runtime.py, src/agent_core/tasks/*	tests/integration/test_worker_dispatch.py, tests/integration/test_orchestrator_mode.py, tests/unit/test_assets_and_cli.py
Task DAG and WAL	06, 17, 19, 21, 22-K	src/agent_core/tasks/*, src/agent_core/storage/task_wal.py, src/agent_core/context/task_board.py	tests/integration/test_plan_task_tools.py, tests/integration/test_worker_dispatch.py
Plan template	11, 16, 21, 22-L	src/agent_core/tasks/tools.py, src/agent_core/assets/templates/plan_default.md, src/agent_core/context/*	tests/integration/test_plan_task_tools.py, tests/integration/test_runtime_loop.py
Compact and memory	08, 10, 11, 19, 22-M	src/agent_core/compact/*, src/agent_core/memory/*, src/agent_core/tools/internal.py, src/agent_core/api/runtime.py	tests/unit/test_compact_memory.py, tests/integration/test_compact_runtime.py, tests/integration/test_memory_extraction_runtime.py
MCP	12, 13, 18, 19, 22-N	src/agent_core/mcp/*, src/agent_core/tools/registry.py, src/agent_core/config/models.py	tests/integration/test_mcp.py, tests/unit/test_hooks_mcp_declarative.py
CLI and end-to-end	03, 18, 19, 21, 22-O	src/agent_core/cli.py, pyproject.toml, src/agent_core/assets/*	tests/unit/test_assets_and_cli.py, tests/e2e/test_ollama_gemma4.py, wheel build and install smoke test

Final Validation

The following commands were run against the current code state:

PYTHONPATH=src python3 -m compileall -q src

Result: passed.

PYTHONPATH=src python3 -m pytest tests/unit tests/integration -q

Result: 320 passed in 5.90s.

SOONG_AGENT_REQUIRE_OLLAMA_E2E=1 PYTHONPATH=src python3 -m pytest tests/e2e -q -rs

Result: 3 passed in 6.14s.

SOONG_AGENT_REQUIRE_OLLAMA_E2E=1 PYTHONTRACEMALLOC=25 PYTHONPATH=src python3 -m pytest tests/unit tests/integration tests/e2e -q -rs -W error::pytest.PytestUnraisableExceptionWarning

Result: 323 passed in 150.36s.

PYTHONPATH=/private/tmp/soong-agent-build-deps python3 -m build --wheel --no-isolation --outdir /private/tmp/soong-agent-dist

Result: wheel built successfully as soong_agent-0.1.0-py3-none-any.whl.

rg -n "provider\s*=\s*['\"](fake|openai|anthropic)['\"]|FakeProvider|fake_provider|OpenAI|Anthropic|openai|anthropic" tests

Result: no matches.

Wheel smoke test:

• Installed the built wheel into a temporary venv with --no-deps.
• Imported agent_core, AgentRuntime, and UserMessage.
• Loaded template.config.default from packaged assets and confirmed it uses provider = "ollama".
• Ran soong-agent --help from the installed console script.

Result: passed.

Current Completion

All documented A-O acceptance areas have direct implementation and test evidence. The final validation suite passes with local Ollama enabled, and provider tests do not exercise OpenAI or Claude providers.

Completion: 100%.