How the tools fit together¶
flowchart LR
Scope["Shape<br/>Scope, specs, ticket"] --> Build["Build<br/>Agent, branch, PR"]
Build --> Verify["Verify<br/>Checks, review, merge"]
Verify --> Finish["Finish<br/>Merged CI, close ticket"]| Tool | Responsibility |
|---|---|
| ai-dlc CLI + local MCP server | Setup, configuration rendering, work records, document tools, and completion gates. MCP exposes selected services to agents; machine mutations stay in the CLI. |
| Claude Code / Codex | Interactive development using shared instructions and skills for discovery, PRDs, specifications, and handoffs. |
| uv + mise | Python environments and dependencies, pinned runtimes, and delegated tool installation. |
| Copier | Project templates, recorded template versions, and staged updates. |
| OpenSpec | Formal requirements, scenarios, and archived changes when the work needs a specification. |
| GitHub Issues + Projects / GitHub Actions | Ticket identity and planning, PR review and merge, and CI evidence. |
| Obsidian | Private continuity notes and navigation to canonical repository documents. |
| hangar (separate tool) | Runs coding agents in cloud sandboxes and starts agent jobs from tickets (intake); renders the team harness, which can pin the workflow plugin, at claim time. |
| flightdeck (separate tool) | Evaluates harness changes, such as running with and without the workflow plugin, on runs hangar recorded; ai-dlc's evaluation runner moved there. |
Providers and hangar¶
Provider roles are configurable. GitHub Issues + Projects is the default and recommended tracker with recorded live evidence. The Jira Cloud adapter is available; its live workflow qualification is pending. The Plane adapter is unsupported and unqualified. The Linear adapter is removed (see the migration guide).
Agent jobs enter through hangar intake. Moving a ticket to a status, label or
Projects column starts a hangar job, and hangar writes job progress back to the
ticket. ai-dlc is the local developer loop around that ticket: a work record,
branch binding and specification artifacts, then a gated work finish. The
tracker boundary says which
tool writes what.
For agent access, register ai-dlc mcp serve --root /absolute/path/to/project
as a stdio MCP server in your client. The tool map lists commands,
MCP tools and skills.
Feature details¶
- Repeatable setup. Versioned profiles select runtimes, tools, and personal agent settings. Each machine keeps its own paths and account bindings.
- Project scaffolding and updates. Start or adopt a generic, Python, Node, or Rust project. Copier templates provide shared instructions, documentation, setup commands, and checks, with previews for adoption and updates.
- Shared agent guidance. Render project instructions, skills, hooks, and MCP configuration for supported clients. Subscribe to pinned team repositories, filter by role or tag, and import supported Tencent teamai layouts.
- Tickets connected to delivery. Bind reviewed scope and specifications to a ticket, branch, and PR. Completion verifies the merged revision and its CI receipts before closing the ticket.
- Repository docs and private knowledge. Search project documents, review documentation impact, and check ownership and links. Connect canonical project docs to an existing Obsidian vault while keeping personal notes separate.
- Engagement and design workflows. Scaffold seven-stage forward-deployed engineering (FDE) engagement docs with local stage checks. Optional frontend tooling captures Playwright evidence for design review.