moneymikeMD/night-watchman
Claude Code plugin for unattended operations: tickets as dispatch contracts, self-starting sessions, and cheap models doing the reading while the expensive one decides.
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
CONTEXT TAX · EVERY TURN
up to ~8.5k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Heavy · median rig: 2.3k · breakdown
Copy this rig
# review before running: this installs third-party code $ npx degit moneymikeMD/night-watchman/agents ./rig-night-watchman/agents $ npx degit moneymikeMD/night-watchman/skills ./rig-night-watchman/skills $ npx degit moneymikeMD/night-watchman/hooks ./rig-night-watchman/hooks
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 moneymikeMD/night-watchman/skills/capability-ladder .claude/skills/capability-ladder $ npx degit moneymikeMD/night-watchman/skills/diagnose-and-pr .claude/skills/diagnose-and-pr $ npx degit moneymikeMD/night-watchman/skills/grill .claude/skills/grill $ npx degit moneymikeMD/night-watchman/skills/handoff-docs .claude/skills/handoff-docs $ npx degit moneymikeMD/night-watchman/skills/session-start .claude/skills/session-start $ npx degit moneymikeMD/night-watchman/skills/shell-scripting .claude/skills/shell-scripting $ npx degit moneymikeMD/night-watchman/skills/tickets-protocol .claude/skills/tickets-protocol $ npx degit moneymikeMD/night-watchman/skills/to-issues-mine .claude/skills/to-issues-mine $ npx degit moneymikeMD/night-watchman/skills/to-issues .claude/skills/to-issues $ npx degit moneymikeMD/night-watchman/skills/wave-trail .claude/skills/wave-trail
$ curl -fsSL --create-dirs -o .claude/agents/cost-reviewer.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/cost-reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/diagnose-and-pr.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/diagnose-and-pr.md $ curl -fsSL --create-dirs -o .claude/agents/librarian.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/librarian.md $ curl -fsSL --create-dirs -o .claude/agents/reflector.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/reflector.md $ curl -fsSL --create-dirs -o .claude/agents/researcher.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/researcher.md $ curl -fsSL --create-dirs -o .claude/agents/script-author-lite.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/script-author-lite.md $ curl -fsSL --create-dirs -o .claude/agents/script-author.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/script-author.md $ curl -fsSL --create-dirs -o .claude/agents/script-reviewer.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/script-reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/spec-reviewer.md https://raw.githubusercontent.com/moneymikeMD/night-watchman/main/agents/spec-reviewer.md
This rig commits no guardrails. Here is the community baseline instead — the deny/ask rules most often found across all 16.2k rigs:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(**/.env)",
"Read(~/.ssh/**)",
"Bash(git push --force:*)",
"Read(**/*.pem)",
"Read(.env)",
"Read(~/.aws/**)",
"Read(**/*.key)",
"Read(**/.env.*)",
"Bash(git push --force*)",
"Bash(rm -rf:*)",
"Bash(rm -rf *)",
"Bash(sudo *)",
"Bash(sudo:*)",
"Bash(rm -rf /)",
"Bash(git push -f:*)",
"Read(.env.*)",
"Bash(git reset --hard*)",
"Bash(rm -rf /*)"
],
"ask": [
"Bash(git push:*)",
"Bash(git push *)",
"Bash(gh pr merge *)",
"Bash(git commit:*)",
"Bash(rm *)",
"Bash(gh pr merge:*)",
"Bash(gh pr create *)",
"Bash(git reset --hard:*)",
"Bash(npm publish:*)",
"Bash(git rebase *)"
]
}
} MCP servers (1)
| server | source | est. tokens |
|---|---|---|
| release env: AI_TOOLKIT_ROOT | local / custom | 2.5k |
Skills (10)
capability-ladderdiagnose-and-prgrillhandoff-docssession-startshell-scriptingtickets-protocolto-issues-mineto-issueswave-trail
Subagents (9)
| cost-reviewer model: sonnet | Runs the end-of-wave cost review from the project's cost ledger (default docs/cost.md's ledger, see that doc for the path convention) — appends this wave's row with scripts/claude-cost.py, compares it |
| diagnose-and-pr model: sonnet | Reads a firing failure signal (an alert, a failed check, a monitoring finding), diagnoses the root cause against this repo, and opens a draft PR fixing it on a feature branch. Never pushes to main, ne |
| librarian model: haiku | Maintains a project's docs/ (or knowledge-base) directory and its tickets, per the tickets-protocol skill. Use for filing tickets, moving tickets between stages, recording decisions, updating known-is |
| reflector model: sonnet | Read-only sonnet agent that mines the current session's transcript for durable corrections and proposes routed skill/agent edits, output as Accepted / Rejected / Backlog tables. Use at session wrap-up |
| researcher model: sonnet | Finds external facts on the web — software versions, upgrade paths, provider/collection maturity, feature support matrices, licences. Use when a decision needs a fact that lives outside the repo and t |
| script-author-lite model: haiku | Captures commands already run live in a session into one small, bounded, non-privileged script under scripts/. Use for a mechanical write-up of a proven command sequence — not for new design work, and |
| script-author model: sonnet | Writes and modifies the reusable operational scripts in scripts/, following the project's shared shell-library conventions. Use whenever operational work needs to land as a script, a new API wrapper i |
| script-reviewer model: sonnet | Reviews shell (and Python-lane) scripts in scripts/ against this project's shared-library conventions and shipped-bug history. Runs by default only for scripts touching credentials, hosts, sudo, or sc |
| spec-reviewer model: sonnet | Reviews a finished ticket branch against its own ticket, not against style — runs the repo's required checks and issues.py scope, then reports missing or partial Solution items, behaviour the ticket d |
Similar rigs
martinisef/ai_sdlc
The AI-native SDLC loop in your repo: agents write intent, spec, plan and code; people approve each gate in git; guardrails for Claude Code, Cursor and AGENTS.md agents.
Orchestrator 5.0k tok · · Q23
SteamedFish/dotfiles
mydotfiles
Minimalist 1.1k tok · · Q0
mtangoz/grill
A second opinion on your decisions, from a different AI company than the one you think with.
Pragmatist 2.7k tok · · Q0
tigerless-labs/autoharness
Autoharness — a self-learning skill layer for Claude Code — distills skills from your real sessions, updates them as you work, and prunes the ones that stop getting used. No daemon, no benchmark.
Pragmatist 2.6k tok · · Q0