~ / rigs / dangnhh92 / viber-harness

dangnhh92/viber-harness

My personal Claude Code harness: one orchestrator, pinned-model agents, a review gate, and memory checked against the code.

↗ GitHub ★ 1 MIT updated 15d ago personal setup Claude CodeCodex
stack: Python
share on X
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
CONTEXT TAX · EVERY TURN
up to ~8.4k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Heavy · median rig: 2.3k · breakdown
GUARDRAILS
3/5
Blocks destructive commands · Pre-tool screening hook · No YOLO mode · details
Quality 23/100 Guardrails v2

Copy this rig

# review before running: this installs third-party code
$ claude mcp add agent-device -e AGENT_DEVICE_NO_UPDATE_NOTIFIER=YOUR_AGENT_DEVICE_NO_UPDATE_NOTIFIER -- npx -y agent-device@0.21.28 mcp
$ npx degit dangnhh92/viber-harness/template/.claude ./rig-viber-harness  # 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.

$ claude mcp add agent-device -e AGENT_DEVICE_NO_UPDATE_NOTIFIER=YOUR_AGENT_DEVICE_NO_UPDATE_NOTIFIER -- npx -y agent-device@0.21.28 mcp
$ npx degit dangnhh92/viber-harness/template/.claude/skills/designing-interfaces .claude/skills/designing-interfaces
$ npx degit dangnhh92/viber-harness/template/.claude/skills/device-testing .claude/skills/device-testing
$ npx degit dangnhh92/viber-harness/template/.claude/skills/harness-audit .claude/skills/harness-audit
$ npx degit dangnhh92/viber-harness/template/.claude/skills/iap-entitlement .claude/skills/iap-entitlement
$ npx degit dangnhh92/viber-harness/template/.claude/skills/new-app .claude/skills/new-app
$ npx degit dangnhh92/viber-harness/template/.claude/skills/serving-mobile-ads .claude/skills/serving-mobile-ads
$ npx degit dangnhh92/viber-harness/template/.claude/skills/writing-copy .claude/skills/writing-copy
$ curl -fsSL --create-dirs -o .claude/agents/analyst.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/analyst.md
$ curl -fsSL --create-dirs -o .claude/agents/builder.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/builder.md
$ curl -fsSL --create-dirs -o .claude/agents/coder.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/coder.md
$ curl -fsSL --create-dirs -o .claude/agents/grunt.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/grunt.md
$ curl -fsSL --create-dirs -o .claude/agents/reviewer.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/reviewer.md
$ curl -fsSL --create-dirs -o .claude/agents/runner.md https://raw.githubusercontent.com/dangnhh92/viber-harness/main/template/.claude/agents/runner.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": "Agent|Task|Bash|Edit|Write|MultiEdit|NotebookEdit|Read|Grep",
        "hooks": [
          {
            "type": "command",
            "command": "python3 \"${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber\" guard"
          }
        ]
      }
    ]
  }
}

MCP servers (1)

serversourceest. tokens
agent-device
env: AGENT_DEVICE_NO_UPDATE_NOTIFIER
npm 2.5k

Skills (7)

Subagents (6)

analyst
model: opus
Analysis, research, design and document writing. Turns a vague ask into a concrete answer, a spec, or a written deliverable. Writes no application code.
builder
model: opus
Implementation that is large in scope, touches many files, or is a complex or risky refactor. Builds against an approved spec and the patterns already in the codebase.
coder
model: sonnet
Simple, contained implementation. One well-scoped change against an existing spec and the patterns already in the codebase. Makes no design decisions.
grunt
model: haiku
Bulk mechanical work across many files: renames, extraction, reformatting, transforms. No judgment, no design, no logic changes.
reviewer
model: sonnet
Reads one finished change for defects before it is committed. Reports findings against the scope contract; never edits code. Called by the orchestrator's review gate, not routed.
runner
model: haiku
Executes a written test or capture plan (simulators, emulators, test suites, screenshots) and returns the raw artifacts. Never judges whether the result is good. Called by whoever wrote the plan, not

Hooks (7)

eventmatcherruns
SessionStart*python3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" state
SessionStartcompactpython3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" recompact
UserPromptSubmit*python3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" state
UserPromptSubmit*python3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" check
PreToolUseAgent|Task|Bash|Edit|Write|MultiEdit|NotebookEdit|Read|Greppython3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" guard
SubagentStop*python3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" handoff
PostToolUseSubagentHandbackpython3 "${CLAUDE_PROJECT_DIR:-.}/.viber/bin/viber" handoff

Similar rigs

copied ✓