cc10x

cc10x

Router-Owned Claude Code Harness

Current version: 10.1.17

Recommended: Create ~/.claude/CLAUDE.md (global) so the router is always active across all projects.

1 Router  •  9 Agents  •  13 Skills  •  4 Workflows

An opinionated Claude Code harness for developers who want routing, state, and verification to behave like engineering infrastructure.

---

Why cc10x

cc10x is for developers who like Claude Code, but do not want to babysit it.

It makes the coding loop behave more like a disciplined engineering system:

• one router entry point
• explicit workflows instead of ad-hoc prompting
• specialist agents instead of one overloaded assistant
• memory and workflow artifacts that survive long sessions
• evidence before advancement: LOG FIRST, RED → GREEN → REFACTOR, expected vs actual verification

The point is not "more AI". The point is tighter execution:

You describe the job. cc10x decides the workflow, loads the right context,
brings in narrow specialists, and blocks weak "done" states.

---

What cc10x Is

cc10x is a developer-focused Claude Code harness packaged as a marketplace plugin.

Its job is straightforward:

• take PLAN, BUILD, DEBUG, REVIEW, and RESEARCH requests and route them through a known execution model
• keep orchestration in one place: cc10x-router
• split execution across specialist subagents instead of a single general prompt
• persist enough workflow state to survive long sessions, compaction, and handoffs
• stay native to Claude Code primitives: skills, subagents, hooks, and optional user-configured MCP

Short mental model:

cc10x = a thin orchestration layer on top of Claude Code
      + durable workflow state
      + specialist agents
      + guardrails that fail closed

It is not a hosted product, background daemon, or parallel control plane. It is a Claude Code plugin that makes normal coding sessions more deterministic.

What You Get

With cc10x installed, Claude Code stops behaving like a single assistant session and starts behaving like a routed workflow:

• PLAN that resolves intent, constraints, scenarios, and default decisions before code starts
• BUILD that pushes implementation through RED → GREEN → REFACTOR, then review, then verification
• DEBUG that starts from logs and observed behavior instead of guess-and-patch loops
• REVIEW that is adversarial by design and reports only confidence-backed findings
• VERIFY that checks scenarios, artifacts, and wiring before anything is treated as complete
• RESEARCH that can use Octocode/Bright Data when present, but does not collapse if they are absent

This is one execution model, not a stack of unrelated prompts.

What 10.1 adds

• Decision-grade planning with direct, execution_plan, and decision_rfc modes selected by the router
• Fresh planning review loop with a bounded read-only plan-gap-reviewer pass before final plan acceptance
• Adversarial spec gates that stop BUILD when feasibility, completeness, or alignment is weak
• Proof-oriented BUILD with explicit checkpoint types, expected artifacts, proof states, and no auto-advance on partial evidence
• Stricter VERIFY that checks truths, artifacts, and wiring before any pass verdict
• Stable workflow UUIDs and versioned v10 state under .claude/cc10x/v10/
• Advisory internal skills where explicit user/project standards always outrank CC10X pattern skills

---

Runtime Model

1. Router owns orchestration

cc10x-router is the only orchestration authority.

It decides:

• which workflow to run
• which subagent to invoke next
• when to pause for clarification or scope decisions
• when remediation is required
• when a workflow is allowed to advance
• when memory and workflow artifacts are finalized

Agents do not own workflow state. They return structured results. The router interprets them.

2. Agents are narrow specialists

The shipped subagents are intentionally specialized:

• planner
• plan-gap-reviewer
• component-builder
• bug-investigator
• code-reviewer
• silent-failure-hunter
• integration-verifier
• web-researcher
• github-researcher

Each agent is optimized for one role. This keeps prompts sharper and makes workflow behavior easier to reason about.

3. Skills are reusable local instructions

Skills are the reusable instruction layer that agents and the router depend on.

They provide:

• planning patterns
• TDD rules
• debugging patterns
• review rules
• research synthesis
• memory handling
• verification-before-completion discipline

4. Workflow artifacts are the durable truth

cc10x writes proof-of-work workflow state under:

.claude/cc10x/v10/workflows/{wf}.json
.claude/cc10x/v10/workflows/{wf}.events.jsonl