待翻译：LaneGate – Git-native worktree orchestrator for AI agents

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Notifications You must be signed in to change notification settings Fork 0 Star 0 BranchesTags Open more actions menu Latest commit History 11 Commits 11 Commits Folders and files NameName Last commit message Last commit date .claude/commands .claude/commands .github/workflows .github/workflows ci ci docs docs lanegate lanegate prompts prompts scripts scripts tests tests tui tui .gitignore .gitignore .gitleaks.toml .gitleaks.toml .lanegate.yml.example .lanegate.yml.example AGENTS.md AGENTS.md CHANGELOG.md CHANGELOG.md CLAUDE.md CLAUDE.md CONTRIBUTING.md CONTRIBUTING.md LICENSE LICENSE NOTICE NOTICE README.md README.md ROADMAP.md ROADMAP.md SECURITY.md SECURITY.md go.work go.work go.work.sum go.work.sum pyproject.toml pyproject.toml test_safeguards.py test_safeguards.py Repository files navigation Parallel lanes. Protected gates. Every agent gets a lane. Nothing ships without a gate. LaneGate coordinates coding agents (Claude, Codex, Ollama, aider, etc.) working on one repo: every ticket gets its own git worktree, declared touches lock the files a ticket owns so a second agent can't claim them, and nothing reaches main without an approved review plus whatever guards you configure. Tickets are Markdown files in your repo. No SaaS, no external project state. You want this if you've watched an agent merge something half-baked and found out two commits later, or you've had two agents clobber each other's edits in the same checkout. The part most tools skip: pre_merge guards re-run against the merged result, not just the ticket's branch. Two tickets that each passed in isolation but break once combined get caught here — LaneGate resets main to its pre-merge commit and routes the ticket back to review instead of leaving a broken commit behind. LaneGate doesn't plan or implement. Your agents and IDE do that; LaneGate runs the gate around them. Security note. V1 provides git-level isolation and diff inspection. It does not sandbox agents at the OS level. Agents run as host processes. See Security Status and Known Limitations before running it on repositories you care about. Demo When LaneGate Helps keep agent work as local Markdown tickets run each task in its own git worktree reduce edit collisions with explicit file-level touches block out-of-scope or sensitive-file changes from auto-merge run configured safeguards before completion or merge pause for human review before anything lands on main By default, LaneGate commits ticket specs and review verdicts to git as they change. Tickets are a git-native artifact, not local-only state. Executor transcripts, logs, cooldowns, locks, and worktrees stay local under .lanegate/. Set commit_status_changes: false to opt back into fully zero-footprint local state. MCP is built in. Run lanegate mcp to expose LaneGate commands as native tools for any MCP-compatible agent (Claude, Cursor, etc.), with no shell commands required. Concepts in 30 seconds .lanegate/tickets/TICK-007.md ← source of truth (YAML frontmatter) │ ▼ lanegate next ← agent asks: what should I work on? lanegate start TICK-007 ← agent claims it; worktree + branch created ... agent works in .lanegate/worktrees/tick-007/ ... lanegate complete TICK-007 lanegate review TICK-007 lanegate merge TICK-007 ← branch → main, worktree removed lanegate promote production ← staged deploy with guard/pre/post hooks Main command groups: Group Commands Scoping: record work as tickets and ask an executor to infer touches / close_criteria create, analyze Execution: who works on what without colliding board, next, start, complete, review, merge Delivery: how merged code reaches environments promote, pipeline-status, flag That's the happy path. For the full state machine — every status, what blocks each transition, where a ticket lands when a guard or review fails, and how many review rounds run — see docs/lifecycle.md. Security Status LaneGate is experimental software for coordinating coding agents on a local repo. Before letting it run without review, read this section. What LaneGate does to limit agent scope: Runs each agent in a dedicated git worktree, one per ticket Enforces a touches list: files committed outside the declared scope route to needs_review Hard-blocks writes to CI/CD configs, dependency manifests (Python, JS/Node, Rust, Go, Java/Gradle, Ruby, PHP, .NET), and credential-shaped files (.env, *.pem, secrets.*) Scans ticket text for prompt injection signals before dispatch Optionally runs gitleaks, semgrep/bandit, pip-audit, npm-audit, composer-audit, and bundler-audit on agent-produced diffs Runs configured safeguards such as tests or scripts before complete and merge What LaneGate does NOT do: MCP is not a sandbox. lanegate mcp exposes ticket lifecycle commands as native tools, and an agent with access to those tools can claim tickets, create branches, and trigger merges. The MCP server authenticates nothing beyond what the MCP client config enforces. Host executors inherit host permissions. Agents (Claude Code, aider, Codex) run as the invoking OS user with full filesystem and network access. There is no bwrap, seccomp, or container wrapping. The file-based lock is single-machine, single-checkout. Two separate clones on two machines can both claim the same ticket. File-level locks are not semantic dependency analysis. Two tickets can touch different files and still break each other through shared APIs, imports, types, schemas, or runtime behavior. Treat touches as a practical coordination boundary, not a proof of correctness. Recommendation: use --human-review final when running lanegate run on any repository with production code. This stops after implementation and requires lanegate merge to be run manually after you inspect the diff. To make this the safe default without depending on every invocation remembering the flag, set default_human_review: final (or per_ticket) in .lanegate.yml. An explicit --human-review flag still overrides it. Note that a ticket's autonomy field also gates the merge step: the default autonomy: supervised already pauses every approved ticket for a human merge decision regardless of --human-review, and unattended merge only happens when both --human-review none and autonomy: full/green/yellow apply — see docs/config-reference.md for the full breakdown. For the full threat model, executor permissions, and safe usage notes, see SECURITY.md and docs/security-model.md. Known Limitations Lock scope is single-machine, single-checkout. The file-based concurrency lock at .lanegate/orchestrator.lock prevents two lanegate run runs on the same checkout from racing, but does not coordinate across separate clones or machines. No OS-level sandbox in V1. Agents run as child processes with full user permissions (no bwrap, seccomp, or container wrapping), regardless of which Claude Code permission mode is configured (see the next bullet for that separate, application-level choice). LaneGate inspects the git diff after the agent exits. It cannot observe what the agent read or sent over the network during execution. Executor permissions come from the executor runtime, not from LaneGate. lanegate init configures Claude Code with a scoped --allowedTools set by default so the agent can run headless without interactive prompts, while tools outside that list stay gated. You can instead configure flags: ["--dangerously-skip-permissions"], which disables Claude Code's per-action approval prompts entirely. That's a valid choice for setups like a sandboxed CI runner, but it means the agent acts on anything without confirmation. See Security Status for the full set of options. Platform support Platform Status Linux Supported, primary development platform macOS Supported, CI-verified on every push Windows Core functionality works, but projects using .sh executor scripts require WSL Install pip install lanegate lanegate init # scaffold .lanegate/ + .lanegate.yml in your repo lgt and lane are installed alongside lanegate as short aliases for the same command (e.g. lgt board, lane run). Use lanegate run (or lane run) to clear the board; lanegate orchestrate remains a compatibility alias. If you added lane to an existing editable install, run pip install -e . once to create the new executable. Ticket analysis matches candidate files by parsing real symbols — via stdlib ast for Python and built-in tree-sitter support for JS/TS, Go, Rust, Java, Ruby, C, and C++. Multi-language symbol parsing is available with every LaneGate installation. LaneGate is a standalone CLI, not a library you import into a project. pipx is the recommended way to install it, since it isolates the tool's own dependencies from whatever Python environment your project uses: pipx install lanegate lanegate init To run from source: git clone https://github.com/sudheerdvn/lanegate cd lanegate pip install -e ".[dev]" Quick start Commit what lanegate init scaffolded (.lanegate.yml, .gitignore, .lanegate/) before creating your first ticket — ticket worktrees are created with git worktree add, which only ever sees committed content, so an uncommitted .gitignore means the first ticket's worktree has none at all: git add .lanegate.yml .gitignore .lanegate/ git commit -m "chore: scaffold LaneGate" For real repositories, configure safeguards before you let an executor run. Safeguard commands run in your project's own environment, not LaneGate's, so make sure the test runner is actually installed there first (e.g. pip install pytest in a fresh venv) or pre_complete fails before the executor gets a chance to do anything: safeguards: pre_complete: - pytest pre_merge: - pytest post_merge: - pytest # 1. Create a ticket file cat > .lanegate/tickets/TICK-001.md << 'EOF' --- id: TICK-001 title: Add rate limiting to the upload endpoint status: open priority: 1 touches: - src/upload.py - src/middleware.py parallel_safe: true autonomy: supervised close_criteria: POST /upload returns 429 after 100 requests/min per user --- ## Background A single client can saturate the worker pool with rapid uploads. EOF # 2. See the board lanegate board # 3. Agent claims and works lanegate start TICK-001 # ... implement in .lanegate/worktrees/tick-001/ ... lanegate complete TICK-001 lanegate review TICK-001 lanegate merge TICK-001 lanegate validate TICK-001 lanegate done TICK-001 # 4. Deploy lanegate promote production Commands lanegate board # see the ticket board and pipeline snapshot lanegate next # choose the next unblocked ticket(s) lanegate create "Add rate limits" --title "Rate-limit uploads" --autonomy supervised lanegate start TICK-007 # claim; creates branch + worktree lanegate complete TICK-007 # code done → code_complete lanegate review TICK-007 # submit for review → in_review lanegate merge TICK-007 # merge to main, delete worktree → merged lanegate run # clear eligible work with the configured executor That covers daily use. For planning/reporting, the full ticket lifecycle, deployment, feature flags, MCP, the local API/UI, the TUI, monitoring daemons, and setup utilities, see docs/commands.md. lanegate --help-all (or lgt --help-all / lane --help-all) prints the complete flat command list. Configuration .lanegate.yml in your repo root, with walk-up discovery so it works from any subdirectory. Covers the delivery profile (profile: strict — enforced review independence), executor routing (claude, codex, aider, ollama, multi-executor executor_steps, dispatch budget caps), safeguards (pre/post-merge test gates, including the automatic re-verification against the real merged main), UI-ticket visual verification, and deployment environments. The Quick Start above shows a minimal safeguards block to get going. See docs/config-refer [truncated for AI cost control]