Zigzag968/claude-code-lgtmgate
An LGTM you can trust: a complete pipeline running several issues at once, each in its own git worktree, through a scripted five-agent flow that self-recovers or escalates to you, on the way to a mergeable PR.
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
Copy this rig
# review before running: this installs third-party code
$ npx degit Zigzag968/claude-code-lgtmgate/.claude ./rig-claude-code-lgtmgate # inspect, then merge into .claude/ MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit Zigzag968/claude-code-lgtmgate/.claude/skills/run-probe-evals .claude/skills/run-probe-evals $ npx degit Zigzag968/claude-code-lgtmgate/plugins/backlog/skills/file .claude/skills/file $ npx degit Zigzag968/claude-code-lgtmgate/plugins/backlog/skills/next .claude/skills/next $ npx degit Zigzag968/claude-code-lgtmgate/plugins/backlog/skills/triage .claude/skills/triage $ npx degit Zigzag968/claude-code-lgtmgate/skills/context .claude/skills/context $ npx degit Zigzag968/claude-code-lgtmgate/skills/deliver .claude/skills/deliver $ npx degit Zigzag968/claude-code-lgtmgate/skills/init .claude/skills/init
$ curl -fsSL --create-dirs -o .claude/agents/mia.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/mia.md $ curl -fsSL --create-dirs -o .claude/agents/morgan.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/morgan.md $ curl -fsSL --create-dirs -o .claude/agents/nick.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/nick.md $ curl -fsSL --create-dirs -o .claude/agents/probe.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/probe.md $ curl -fsSL --create-dirs -o .claude/agents/sam.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/sam.md $ curl -fsSL --create-dirs -o .claude/agents/theo.md https://raw.githubusercontent.com/Zigzag968/claude-code-lgtmgate/main/agents/theo.md
Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"permissions": {
"deny": [
"Bash(rm)",
"Bash(rm *)",
"Bash(git clean:*)",
"Read(**/.ssh/*)",
"Read(**/.aws/*)",
"Read(**/.env*)",
"Read(**/*secret*)",
"Bash(git push origin main)",
"Bash(git push origin main:*)",
"Bash(ALLOW_PUSH_TO_MAIN=1 git push origin main)",
"Bash(ALLOW_PUSH_TO_MAIN=1 git push origin main:*)"
]
}
} Skills (7)
Subagents (6)
| Mia model: claude-haiku-4-5-20251001 | Mia (PM) — Product framing agent, generic and reusable on any stack. Spawned by the Lead (via the deliver-pipeline workflow) only when the feature template's pm_review checkbox is checked. Reads the p |
| Morgan model: claude-sonnet-5 | Morgan (Reviewer) — Generic senior reviewer, reusable on any stack. Reviews Nick's PR in the shared worktree against Sam's plan and the project's conventions rule, runs a regression guard, monitors CI |
| Nick model: claude-sonnet-5 | Nick (Dev) — Generic developer agent, reusable on any stack. Works in the task's shared worktree, reads Sam's plan from the issue, implements it with tests, and opens a draft PR against the base branc |
| probe model: haiku | Probe — mechanical copier for the engine's probe-run gate. Runs the one or two given probe-run commands in order and returns the PROBE line and the VERIFY line they printed, verbatim. Never judges, re |
| Sam model: claude-sonnet-5 | Sam (Scout & Planner) — generic scout and implementation planner, reusable on any stack. Works in the task's shared worktree, scans the codebase, produces an anchored impact table + implementation pla |
| Theo model: claude-sonnet-5 | Theo (Diagnose) — Epistemic diagnosis agent, generic and reusable on any stack. Spawned by the Lead (via the deliver-pipeline workflow) before Sam, on EVERY dispatched issue — mandatory gate, no opt-o |
Permissions
deny (11)
Bash(rm)
Bash(rm *)
Bash(git clean:*)
Read(**/.ssh/*)
Read(**/.aws/*)
Read(**/.env*)
Read(**/*secret*)
Bash(git push origin main)
Bash(git push origin main:*)
Bash(ALLOW_PUSH_TO_MAIN=1 git push origin main)
Bash(ALLOW_PUSH_TO_MAIN=1 git push origin main:*)
ask (0)
—
allow (1)
Bash(git rm:*)
Similar rigs
warrendeanlangeveldt/code-kit
Keeps AI coding agents inside your project's rules: lanes, layer rules, protected paths and proof before finish, enforced by Claude Code hooks and in CI
Pragmatist 1.2k tok ·
itsOmidKarami/kraft
A local orchestrator that takes your coding agent from spec to pull request, stopping only when a decision is yours.
Skill Collector 7.9k tok ·
bentleypark/claude-code-mobile-spine
Claude Code plugin marketplace bundling mobile-spine — scaffolds a meta-repo coordinating 4 subagents (api/pm/android/ios) for mobile teams whose Android, iOS, and Backend live in 3 separate repos.
Minimalist 84 tok ·
yongkyung-oh/agent-bootstrap
A minimal workflow bootstrap for AI coding agents — three markdown files that solve context loss between sessions. No framework, no dependencies.
Minimalist 1.7k tok ·