henryzhangpku/agent-starter
Run a coding agent like a new hire: a written working agreement, guardrails enforced by hooks, a fresh-context reviewer, plan and decision logs. One-line bootstrap.
stack: Python
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
GUARDRAILS
3/5
Blocks destructive commands · Pre-tool screening hook · No YOLO mode · details
Quality 20/100 Guardrails v2
Copy this rig
# review before running: this installs third-party code
$ npx degit henryzhangpku/agent-starter/template/.claude ./rig-agent-starter # 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 henryzhangpku/agent-starter/template/.claude/skills/async-concurrency .claude/skills/async-concurrency $ npx degit henryzhangpku/agent-starter/template/.claude/skills/reliability .claude/skills/reliability $ npx degit henryzhangpku/agent-starter/template/.claude/skills/sql-and-schema .claude/skills/sql-and-schema $ npx degit henryzhangpku/agent-starter/template/.claude/skills/team-repo .claude/skills/team-repo
$ curl -fsSL --create-dirs -o .claude/agents/architect.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/architect.md $ curl -fsSL --create-dirs -o .claude/agents/explorer.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/explorer.md $ curl -fsSL --create-dirs -o .claude/agents/implementer.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/implementer.md $ curl -fsSL --create-dirs -o .claude/agents/reviewer.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/safety-reviewer.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/safety-reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/test-engineer.md https://raw.githubusercontent.com/henryzhangpku/agent-starter/main/template/.claude/agents/test-engineer.md
Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"hooks": {
"PreToolUse": [
{
"matcher": "Edit|Write|MultiEdit|NotebookEdit",
"hooks": [
{
"type": "command",
"command": "python3 \"$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py\" || python \"$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py\""
}
]
},
{
"matcher": "Bash|PowerShell",
"hooks": [
{
"type": "command",
"command": "python3 \"$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py\" || python \"$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py\""
}
]
}
]
}
} Skills (4)
Subagents (6)
| architect model: opus | Decomposes the agreed plan into parallel tasks with disjoint file ownership and fixed contracts. Use after PLAN.md exists and before any parallel work. Writes only TASKS.md and CONTRACTS.md. |
| explorer model: sonnet | Read-only mapping of an existing codebase before any change - how to run, test, deploy; conventions; where a feature would go; who owns what. Use first when dropped into someone else's repository. |
| implementer model: sonnet | Implements exactly one task from TASKS.md inside the paths that task owns, against CONTRACTS.md, until its done-when tests pass. Use one per task; several can run in parallel on tasks with disjoint ow |
| reviewer | Read-only, fresh-context reviewer. Use after a PLAN.md step is implemented and before it is committed. Checks the diff against the plan, CLAUDE.md rules and the tests; may add edge-case tests only in |
| safety-reviewer model: opus | Read-only review of a change for failure handling, secrets, money/identity/compliance paths and data protection. Use on any change that touches external calls, user data, payments or decisions, alongs |
| test-engineer model: sonnet | Writes tests for a task from its spec and contracts, without reading or writing its implementation. Use in parallel with the implementer of the same task so the tests are independent of the code they |
Hooks (5)
| event | matcher | runs |
|---|---|---|
| PreToolUse | Edit|Write|MultiEdit|NotebookEdit | python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py" || python "$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py" |
| PreToolUse | Bash|PowerShell | python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py" || python "$CLAUDE_PROJECT_DIR/.claude/hooks/guard.py" |
| PostToolUse | Edit|Write|MultiEdit|NotebookEdit | python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/run_tests.py" || python "$CLAUDE_PROJECT_DIR/.claude/hooks/run_tests.py" |
| UserPromptSubmit | * | python3 "$CLAUDE_PROJECT_DIR/.claude/scripts/clock.py" hook || python "$CLAUDE_PROJECT_DIR/.claude/scripts/clock.py" hook |
| UserPromptSubmit | * | python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/context_budget.py" || python "$CLAUDE_PROJECT_DIR/.claude/hooks/context_budget.py" |
Slash commands (18)
/capture/change/checkin/demo-prep/dispatch/integrate/make-plan/make-tests/next/onboard/publish/questions/scaffold/ship/show/step/team-plan/update-kit
Similar rigs
yuroka-studio/nightshift-agents
Scheduled Claude Code agents that take one task from a backlog, open a PR with real test evidence, and never merge or deploy on their own.
Fort Knox 1.0k tok · · Q61
ingenierodeatascos-cyber/ai-project-template
AI-First project template for building software with orchestrated AI agents using Spec-Driven Development. Includes router, planner, implementer and reviewer workflows for structured development with Codex, Claude Code or similar AI coding
Orchestrator 196 tok · · Q0
Trolzie/claude-loop-systems-demo
A small demo of Markdown-as-system AI dev loops with Claude Code: issues → agents → skills → verification.
Pragmatist 480 tok · · Q15
homeofe/improvements
Portable, LLM-agnostic framework for multi-model AI workflows: model routing, structured agent handoff (AAHP), and a phased agent pipeline. MIT.
Orchestrator 1.4k tok · · Q24