Quality gate and pilot for AI coding agents
Nothing ships without proof it doesn't break things.
What gait does
gait is a VS Code extension that runs your pipeline, watches your agents, and blocks bad code from shipping.
|
Gate — lint, typecheck, test in dependency order with early abort. Secret scanning on every commit. Regression detection against per-branch baselines.
Agents — launch Claude or Codex from VS Code. Live token tracking, context bar, pause/resume/kill. Auto-gate after every agent session.
Autofix — failed stage? Click Fix. Agent gets the error, blame context, and source code. Auto-fix loop retries up to 3x.
|
Review — AI code review as a pipeline stage. Structured findings with severity levels. Squiggly lines in your editor.
Memory — agents remember your project: conventions, past corrections, "never do this" rules. Context persists across sessions.
Workflows — multi-agent orchestration. Define agent → gate → agent → gate pipelines in YAML. Snapshot before, restore on failure.
|
Quick Start
# Install from source
git clone https://github.com/arodroz/gait.git && cd gait
npm install && npm run compile
# Press F5 in VS Code to launch
Then:
Cmd+Shift+P → Gait: Initialize Project
Cmd+Shift+G → Run Quality Gate
gait init detects your stack, configures commands, creates scripts, sets up linters, generates prompt templates and example workflows. One command, everything ready.
Features
Pipeline
| Stage |
What happens |
| Lint |
Runs your linter (eslint, go vet, ruff, swiftlint) |
| Typecheck |
Type checker (tsc, go vet, mypy) |
| Test |
Test suite with output parsing for regression detection |
| Audit |
Dependency vulnerability scan (npm audit, pip-audit, go mod verify) |
| Review |
AI code review with structured JSON findings |
Stages run in dependency order. First failure aborts. Profiles switch between quick (lint only) and full (everything).
Agent Integration
Cmd+Shift+P → Gait: Run Agent
- Pick a prompt template or type custom
- Snapshot taken automatically before agent runs
- Dashboard shows: tokens, context %, elapsed, pause/resume/kill
- Live diff updates every 2s as agent writes code
- Auto-gate runs when agent finishes
- Cost tracking with daily budget enforcement
- Memory prepended to every prompt — agents get smarter over time
Autofix
When a stage fails, the dashboard shows a Fix button:
- Click → scoped prompt (error + source + blame) → agent fixes → auto-gate
- Shift+click → auto-fix loop (3 attempts, no human in loop)
- Config →
autofix = true makes it fully automatic
More
| Feature |
Description |
| Secret scanning |
AWS keys, tokens, private keys, high-entropy strings |
| Pre-commit hook |
Gate runs before every commit, modal in dashboard |
| Rollback |
Simulate revert in worktree, verify tests pass, then apply |
| Release |
Conventional commits → semver bump → changelog → tag |
| PR generator |
Structured PR body from git log, push + create via gh |
| Regression detection |
Per-branch baselines, flaky test exemption |
| Coverage detection |
Find untested functions in changed files |
| Test generation |
Agent writes tests for uncovered code |
| Monorepo |
Affected-only testing for go.work, npm workspaces, python |
| Scripts |
.gait/scripts/ with metadata headers and dependency resolution |
| Notifications |
Slack, Discord, webhook on gate/agent/regression events |
| Impact analysis |
Map source files → test files, log affected tests |
| Hooks suite |
pre-commit, pre-push, post-merge, post-checkout |
| AGENTS.md |
Auto-generated agent instructions from config |
| Environment check |
Verify required tools are on PATH |
Commands
31 commands via Cmd+Shift+P:
Pipeline & Gate
| Command |
Keybinding |
| Gait: Initialize Project |
— |
| Gait: Run Quality Gate |
Cmd+Shift+G |
| Gait: Run Lint |
— |
| Gait: Run Tests |
— |
| Gait: Run Typecheck |
— |
| Gait: Run Build |
— |
| Gait: Switch Pipeline Profile |
— |
| Gait: AI Code Review |
— |
| Gait: Audit Dependencies |
— |
Agents & AI
| Command |
Description |
| Gait: Run Agent |
Launch Claude/Codex with template picker |
| Gait: Edit Agent Memory |
Open .gait/context.md |
| Gait: View Memory |
Show corrections, patterns, rules |
| Gait: Generate Tests for File |
Agent writes tests for uncovered code |
| Gait: Cost Summary |
Daily/weekly/monthly spend |
Git & Release
| Command |
Description |
| Gait: Install Pre-Commit Hook |
Gate before commits |
| Gait: Install All Hooks |
pre-commit + pre-push + post-merge + post-checkout |
| Gait: Manage Hooks |
Toggle individual hooks |
| Gait: Rollback Assistant |
Simulate revert in worktree |
| Gait: Release |
Semver bump + changelog + tag |
| Gait: Create Pull Request |
Generate + push + gh pr create |
| Gait: Take Snapshot |
Manual working tree snapshot |
| Gait: Restore Snapshot |
Pick and restore |
Tools
| Command |
Description |
| Gait: Open Dashboard |
Cmd+Shift+D |
| Gait: Run Script |
Execute .gait/scripts/*.sh |
| Gait: List Scripts |
Show all with metadata |
| Gait: Detect Script Patterns |
Find repeated commands → save as scripts |
| Gait: Run Workflow |
Multi-agent orchestration |
| Gait: Generate AGENTS.md |
Agent instructions from config |
| Gait: Environment Check |
Verify tools on PATH |
| Gait: Recover |
Clean stale worktrees/locks |
Configuration
[project]
name = "myproject"
[stacks.typescript]
Lint = "npx eslint src/"
Test = "npx vitest run"
Typecheck = "npx tsc --noEmit"
Build = "npm run build"
[pipeline]
stages = ["lint", "typecheck", "test", "audit", "review"]
timeout = "300s"
autofix = true
autofix_agent = "claude"
daily_budget_usd = 10.00
commit_profile = "full"
[review]
agent = "claude"
block_on = "error"
[notifications]
slack_webhook = "https://hooks.slack.com/..."
events = ["gate.failed", "regression.detected"]
.gait/ directory structure
| Path |
Git |
Purpose |
config.toml |
committed |
All configuration |
scripts/*.sh |
committed |
Repeatable operations |
prompts/*.md |
committed |
Agent prompt templates |
workflows/*.yaml |
committed |
Multi-agent workflows |
context.md |
committed |
Agent project context |
.gitignore |
committed |
State file exclusions |
baseline_*.json |
gitignored |
Test baselines per branch |
flaky.json |
gitignored |
Flaky test tracking |
costs.json |
gitignored |
Agent cost data |
impact-map.json |
gitignored |
Test impact mapping |
snapshots.json |
gitignored |
Snapshot index |
memory.json |
gitignored |
Agent corrections/patterns |
history/*.jsonl |
gitignored |
Action log |
Architecture
src/
├── extension.ts ← 31 commands, lifecycle, event wiring
├── core/ ← 40 modules
│ ├── pipeline.ts ← Stage runner, topo-sort, early abort
│ ├── runner.ts ← Shell executor (spawn + escape + timeout)
│ ├── config.ts ← TOML config, stack detection
│ ├── agent.ts ← Claude/Codex, SIGSTOP/CONT/KILL
│ ├── autofix.ts ← Fix prompt builder, auto-fix loop
│ ├── review.ts ← AI code review, JSON finding parser
│ ├── memory.ts ← Persistent agent context + corrections
│ ├── blame.ts ← Git blame for error root cause
│ ├── secrets.ts ← Regex + Shannon entropy scanning
│ ├── baseline.ts ← Test baselines, regression diffing
│ ├── coverage.ts ← Per-stack coverage detection
│ ├── dep-audit.ts ← npm/go/python vulnerability audit
│ ├── snapshot.ts ← Git tag snapshots, restore, prune
│ ├── profiles.ts ← Pipeline profiles (quick/full)
│ ├── prompts.ts ← Template parser, variable interpolation
│ ├── workflow.ts ← Multi-agent YAML workflows
│ ├── cost-tracker.ts ← Token/cost estimation, budget
│ ├── pr-generator.ts ← PR summary, gh CLI integration
│ ├── notify.ts ← Slack/Discord/webhook notifications
│ ├── ... ← + 15 more (git, hooks, scripts, etc.)
│
├── views/
│ ├── statusbar.ts ← Status bar badges
│ ├── sidebar.ts ← 4 tree views
│ └── dashboard.ts ← Webview panel
│
└── webview/
└── main.ts ← Dashboard UI (safe DOM)
Development
npm install # dependencies
npm run compile # build extension + webview
npm run lint # tsc + eslint
npm test # 216 tests / 33 files
npm run watch # rebuild on change
Press F5 to debug. Package: npx @vscode/vsce package --no-dependencies
License
MIT
Built with Claude Opus 4.6 · github.com/arodroz/gait
