- system-overview-design.md: repaired with P0/P1/P2 gaps resolved, all 24 frozen baselines listed, error taxonomy, global ~/.air, IPC, TaskSpec/WorkerResult, PromptLayer, PermissionEngine, RuntimeEvent, state machines, capability trust, artifact naming, operations - Four cross-verification audit reports (GPT-5, MIMO 2.5, Opus 4.7, DeepSeek V4 Pro) - Three regression reviews (R1: initial repair closure, R2: second pass with PromptLayer L9 wording found, R3: full PromptLayer alignment verified) - AGENTS.md, plan.md, todo.md synchronized Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
11 KiB
Executable File
Implementation Plan
Date: 2026-05-27 Status: Formal architecture-to-V1.0.0 Alpha plan
Read First
AirPlan/AGENTS.mdAirPlan/docs/analysis/requirements.mdAirPlan/docs/architecture/solution-architecture.mdAirPlan/docs/architecture/system-overview-design.mdAirPlan/docs/architecture/c4/module.mdAirPlan/docs/architecture/baselineV1.mdAirPlan/docs/architecture/多模型三视角审查联合评估.mdAirPlan/docs/architecture/gpt5概要设计审查.mdAirPlan/docs/architecture/mimo2.5概要设计审查.mdAirPlan/docs/architecture/Opus4.7概要设计审查.mdAirPlan/docs/architecture/DeepSeek概要设计审查.mdAirPlan/docs/architecture/概要设计修复回归审查.mdAirPlan/docs/architecture/adr/AirPlan/todo.md
Goal
Build AirCoding V1.0.0 Alpha: a Linux-first Bun/TypeScript local AI coding runtime with project-local .air state, SQLite session persistence, event/domain store, scheduler/worker IPC, ToolRegistry/PermissionEngine, provider adapter boundary, complete C++ development workflow, usable plugin/capability foundation, OpenTUI/Solid ProjectionStore UI, Doctor, and release validation gates.
Architecture Baseline
Implementation must follow these canonical V1 documents:
| Area | Document |
|---|---|
| Requirements | AirPlan/docs/analysis/requirements.md |
| Solution architecture | AirPlan/docs/architecture/solution-architecture.md |
| System overview design | AirPlan/docs/architecture/system-overview-design.md |
| C4/module view | AirPlan/docs/architecture/c4/module.md |
| Baseline | AirPlan/docs/architecture/baselineV1.md |
| DB schema | AirPlan/docs/architecture/db-schema-v1.md |
| Event registry | AirPlan/docs/architecture/event-registry-v1.md |
| Tool registry | AirPlan/docs/architecture/tool-registry-v1.md |
| Scheduler state machine | AirPlan/docs/architecture/scheduler-state-machine-v1.md |
| Prompt layering | AirPlan/docs/architecture/prompt-layering-v1.md |
| Provider matrix | AirPlan/docs/architecture/provider-capability-matrix-v1.md |
| Error taxonomy | AirPlan/docs/architecture/error-taxonomy-v1.md |
| Artifact naming | AirPlan/docs/architecture/artifact-naming-v1.md |
| Scope escalation | AirPlan/docs/architecture/scope-escalation-v1.md |
| Security model | AirPlan/docs/architecture/security-model-v1.md |
| Capability trust | AirPlan/docs/architecture/capability-trust-v1.md |
| Platform matrix | AirPlan/docs/architecture/cross-platform-matrix-v1.md |
| Multi-model phase-gate assessment | AirPlan/docs/architecture/多模型三视角审查联合评估.md |
Phase Gate Status
The P0/P1/P2 remediation and UX scope gate from AirPlan/docs/architecture/多模型三视角审查联合评估.md are complete:
- source-of-truth and V1.0.0 Alpha vocabulary cleanup — DONE (T-000A)
- public contract freeze repair — DONE (T-000B)
- persistence/runtime invariant repair — DONE (T-000C)
- UX scope decision — DONE (T-000D): all features scheduled as Alpha; VibeBox removed from main plan
System overview design has been repaired after the GPT-5, MIMO 2.5, Opus 4.7, and DeepSeek overview audits. All P0/P1 and non-decision alignment gaps were folded into system-overview-design.md; 概要设计修复回归审查.md passed the overview regression gate. Detailed design and packages/contracts implementation may proceed.
Constraints
- Runtime: TypeScript on Bun.
- Monorepo: Bun workspaces + Turborepo.
- TUI:
@opentui/solid,@opentui/core,@opentui/keymap. - IPC: NDJSON over stdio.
- DB: project-local SQLite per session.
- Internal message format: Anthropic canonical content blocks.
- Workers: independent Bun child processes.
- V1.0.0 Alpha platform: Linux x86_64 tier 1, Linux arm64/WSL2 tier 2.
- Execution layer must follow Claude Code-style edit/verification discipline.
- All side effects must pass through ToolRegistry and PermissionEngine.
Phases
Phase 0: Repository Skeleton and Contracts
Objective: create a compileable monorepo foundation and shared contract package.
Deliverables:
- Bun workspace and Turborepo config
packages/contracts- core types for runtime, events, IPC, task, worker result, tool, artifact, provider, project, UI, error
- baseline test/lint/typecheck commands
Validation:
bun install
bun run typecheck
bun test
Phase 1: Project and Session Storage Foundation
Objective: implement project-local .air initialization and SQLite V1 session store.
Deliverables:
- project
.air/sharedand.air/localinitializer - session ID/project ID creation
- SQLite schema/migration V1
- SessionStore repositories
- EventStore transactional durable event application
- ArtifactStore temp-write/rename/hash/DB insert
Validation:
bun test packages/runtime
bun run air -- doctor --read-only --project <fixture>
bun run air -- init <fixture>
Phase 2: ToolRegistry, PermissionEngine, Core Tools, and Capability Foundation
Objective: provide safe side-effect execution primitives and a usable plugin/capability foundation for V1.0.0 Alpha.
Deliverables:
- ToolRegistry
- ToolResult envelopes and tool lifecycle events
- realpath path classifier
- command risk analyzer
- PermissionEngine decision records
- filesystem, shell, git, project, artifact, context, permission, doctor core tools
- CapabilityRegistry manifest loading/validation
- local and built-in capability enable/disable config
- namespaced capability tool registration
- Doctor dependency check path for capabilities
Validation:
bun test packages/runtime --filter tool
bun test packages/runtime --filter permission
Phase 3: Provider and Context Layer
Objective: support model calls through canonical context assembly and provider capability validation.
Deliverables:
- provider config loader
- ProviderCapabilityMatrix implementation
- ProviderAdapter interface
- one Anthropic-compatible or OpenAI-compatible adapter path
- ContextAssembler L0-L9 layering
- omission/conflict reporting
- provider conversion report
Validation:
bun test packages/llm
bun test packages/runtime --filter context
Real-provider smoke gated by local config:
bun run air -- provider doctor
bun run air -- provider smoke
Phase 4: Worker IPC and Scheduler V1.0.0 Alpha
Objective: run independent child workers with Scheduler state machine.
Deliverables:
- WorkerManager process spawn/monitor
- NDJSON IPC framing
- Executor/Reviewer/Debugger/Compactor/ExperienceMiner process foundations
- heartbeat and timeout handling
- TaskGraph load/plan/dispatch/collect
- retry budget/failure signature tracking
- workspace create/merge events
- restart recovery for lost/running agents
Validation:
bun test packages/runtime --filter ipc
bun test packages/runtime --filter scheduler
bun run air -- e2e worker-fixture
Phase 5: Complete C++ Development Workflow
Objective: deliver the first complete language workflow, not just build/test smoke coverage.
Deliverables:
packages/toolchain-cppcpp.detectcpp.cmake.configurecpp.buildcpp.testcpp.static.cppcheckcpp.clangd.query- compiler/linker/test diagnostic parser and semantic signatures
- compile_commands generation/discovery
- build-system conflict reporting
- Debugger integration for failed build/test/debug evidence
- Reviewer integration for C++ changes
- automatic scoped repair loop for representative C++ failures
- C++ fixture projects covering configure/build/test/static-analysis/debug/fix/review
Validation:
bun test packages/toolchain-cpp
bun run air -- fixture cpp-build-test
bun run air -- e2e cpp-fix-fixture
bun run air -- e2e cpp-debug-review-fixture
Phase 6: TUI/HUD Projection
Objective: expose runtime state through OpenTUI/Solid UI without coupling UI to DB/EventBus.
Deliverables:
- ProjectionStore hydrate/apply APIs
- TUI startup shell
- session/task/agent/tool/command/artifact views
- permission prompt UI
- blocker report UI
- basic HUD/statusline projection
Validation:
bun test packages/tui
bun run air -- tui-smoke --project <fixture>
Phase 7: Agent Prompt Integration and End-to-End Loop
Objective: connect Main Agent, Architecture Designer, Scheduler, workers, tools, context, and provider call path.
Deliverables:
- built-in prompt resources
- Main Agent request classification
- Architecture Designer impact assessment flow
- Executor WorkerResult loop
- Reviewer report loop
- Debugger failure evidence loop
- Compactor summary loop
- scope escalation integration
Validation:
bun run air -- e2e simple-edit-fixture
bun run air -- e2e cpp-fix-fixture
Phase 8: Release Gate and Packaging
Objective: make V1.0.0 Alpha foundation reproducible and releasable for tier-1 Linux.
Deliverables:
- binary tarball packaging
- resource bundle
- release gate script
- doctor report bundle generation
- documentation of skipped gates/known limitations
Validation:
bun run release:check
bun run release:package
Parallelization Boundaries
Can parallelize after contracts stabilize:
- storage/EventStore and provider/context adapter work
- ToolRegistry/PermissionEngine and ArtifactStore
- C++ workflow tools/debug/review fixtures and TUI ProjectionStore
- worker foundation and Scheduler tests
- capability manifest validation and built-in capability registration
Must serialize:
packages/contractsbefore package implementations.- DB schema before EventStore/SessionStore tests.
- ToolRegistry/PermissionEngine before tools and workers rely on side effects.
- Provider/context contract before agent prompt integration.
- Scheduler IPC before real worker E2E.
Validation Strategy
Minimum per-task validation appears in AirPlan/todo.md.
Release-level validation:
bun install
bun run typecheck
bun test
bun run lint
bun run air -- doctor --read-only
bun run air -- e2e worker-fixture
bun run air -- fixture cpp-build-test
bun run air -- e2e cpp-fix-fixture
bun run air -- e2e cpp-debug-review-fixture
bun run air -- capability validate --all
bun run release:check
If a validation gate cannot run, record:
why skipped
evidence available
risk
follow-up task
Static Analysis Strategy
V1.0.0 Alpha static analysis:
- TypeScript typecheck for all packages.
- ESLint or equivalent once repo config exists.
cppcheckfor C++ fixture/toolchain validation.- Architecture/static consistency reviewed through ADR/C4 updates.
Risk Controls
| Risk | Control |
|---|---|
| contract churn | implement contracts first and version event/tool schemas |
| unsafe edits | enforce read-before-edit and exact edit tools |
| hidden side effects | all tools through PermissionEngine |
| context overflow | ContextAssembler budgets and Compactor |
| worker loss | heartbeat, timeouts, task_attempts, restart recovery |
| provider mismatch | capability matrix and conversion reports |
| evidence gaps | artifact/evidence refs required in WorkerResult |
| platform surprises | Doctor platform report and Linux tier-1 release gates |