CC10x Workflow Engine
Give Claude Code’s multi-step development work durable state, explicit evidence, and gated completion.
- Source repo
- romiluz13/cc10x
- Stars
- ★ 165
- Last updated
- 11d ago
- License
- MIT
- Primary language
- Python
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A Claude Code developer building a multi-step feature who wants TDD and independent integration verification
- Not a fit if
- Teams seeking a directly supported runtime outside Claude Code
- Users who do not want workflows to persist state in project files
- Source review
- 54/100 · Major gaps
What does this agent do, and when should you use it?
CC10x is a Claude Code plugin whose sole entry skill, cc10x-router, classifies a request and selects one of eight workflows: BUILD, DEBUG, PLAN, REVIEW, ORIENT, QA, TRIAGE, or CODEBASE-HEALTH. Fourteen specialist subagents carry out the work, while reusable skills define practices such as TDD, planning, review, and verification; hooks provide lightweight audits and selected blocking guards. The router stores memory, phase evidence, verdicts, and events under the project’s .cc10x/ directory so work can resume after a restart or context compaction. Depending on the route, it produces code changes, review findings, plans, QA reports, or a codebase-health HTML report. It installs as a Claude Code plugin through the marketplace or a local plugin directory and requires Python 3.9 or newer; MCP integrations are optional.
After a user describes a development task, cc10x-router routes by primary deliverable and asks for missing requirements when needed. BUILD uses component-builder for RED → GREEN → REFACTOR, runs code-reviewer and failure-hunter in parallel, then asks integration-verifier to validate; DEBUG investigates logs and observed behavior before review and verification. PLAN produces a plan for an independent plan-gap-reviewer check, REVIEW reports findings with file-and-line evidence, and QA surveys code, specifications, tickets, or workflow artifacts, reviews a test plan, provisions an environment, and executes tests. Other routes classify incoming issues or PRs, produce a codebase-health HTML report, or answer orientation questions without writes. Workflows read and update .cc10x/ memory plus .cc10x/workflows/{wf}.json and event logs. If configured, octocode or brightdata MCP servers can augment research; without them, research falls back to Claude Code’s built-in tools.
- A Claude Code developer building a multi-step feature who wants TDD and independent integration verification
- A maintainer investigating a bug from logs and observed behavior and needing regression validation
- An engineering team turning rough requirements into an execution-ready plan with a separate gap review
- A QA team surveying a feature, preparing an isolated test environment, and running an end-to-end test plan
- A maintainer who needs evidence-backed PR findings or a codebase-health scan and HTML report
How do you install or deploy this agent?
CC10x requires Claude Code and Python 3.9 or newer. Install through the Claude Code marketplace:
claude plugin marketplace add romiluz13/cc10x
claude plugin install cc10x@cc10xThen say “set up cc10x for me” in Claude Code and restart. To load a clone for the current session only:
git clone https://github.com/romiluz13/cc10x.git
claude --plugin-dir "$PWD/cc10x/plugins/cc10x"The plugin writes project state under .cc10x/; the README recommends adding .cc10x/ to the project’s .gitignore. The octocode and brightdata MCP integrations are optional.
How do you use this agent?
After installation and setup, restart Claude Code and describe the work you want done. The router selects a workflow. For example:
build a user authentication systemOr:
review this PR for security issuesFor multi-step work such as BUILD and DEBUG, the router resolves required missing information before specialist agents proceed. State and evidence are stored in the project’s .cc10x/ directory. To update the plugin, run:
claude plugin update cc10x@cc10x --scope userThen run /reload-plugins in an open session or restart Claude Code. Replace user with the scope used during installation (project, local, or managed are also documented).
What are this agent's strengths and limitations?
- Persists workflow state, evidence, verdicts, and events to disk so sessions can resume after compaction or restart.
- BUILD uses TDD and a separate integration verifier rather than relying only on the builder’s completion claim.
- Review findings require file-and-line evidence, and the verifier rereads cited lines to drop hallucinated findings.
- Test Honesty Gates flag specific weak-test patterns so those tests alone cannot establish a PASS.
- Octocode and Bright Data MCP integrations are optional; built-in Claude Code tools remain the fallback.
- The documented runtime is a Claude Code plugin; the README does not document direct use in other agent platforms.
- Setup involves marketplace installation, configuration, and a restart; hooks also require Python 3.9 or newer.
- Workflow state is written into the project’s
.cc10x/directory, which should be gitignored and may require file-write permissions. - Use depends on Claude Code and its model service; the README does not state model usage costs or pricing.
How does this agent compare with similar options?
The README contrasts CC10x with asking Claude to handle a complex task directly. Its stated approach is to constrain execution with routing, persisted state, and verification gates; it does not provide a feature comparison with other named tools.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| CC10x Workflow Engine This agent | 54 · Major gaps | Agent plugin / skillFree + model costs | ★ 165 | 11d ago | Python | Claude Code |
| Ring — Skills Library for AI Agents | 72 · Some gaps | Agent plugin / skillFree + model costs | ★ 217 | 5d ago | HTML | Codex · Claude Code |
| Pi Dynamic Workflows | 88 · Good | Agent plugin / skillFree + model costs | ★ 555 | 12d ago | TypeScript | — |
| Aura IDE | 77 · Good | Desktop appFree + model costs | ★ 57 | 1mo ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes workflow state written under project .cc10x/, optional MCP integrations, and approval prompts for outward-facing documents; it also says a git guard denies pushes and forced deletes. These improve visibility into data flow and external effects, though they are primarily documentary claims. Setup instructions recommend editing global CLAUDE.md and granting broad Bash(python3:*) permission, so least privilege is only partial. No credential protections, dependency security review, or state rollback mechanism is shown. The author name and repository path provide limited attribution, but publisher identity is unverified and maintenance responsibility is unclear.
The README explains routing, workflow state, the Python 3.9 prerequisite, and fallback to built-in tools when optional MCPs are unavailable. It also documents troubleshooting for EXDEV, disabled plugins, and missing Python. Deductions reflect the mismatch between the supplied description and README counts for agents, skills, and workflows, and seam_eval.py's claim that text checks guarantee the builder's correct decision, despite also stating that it does not run a builder.
The README covers build, debug, review, and QA scenarios with trigger words and priorities; it identifies ORIENT as read-only, REVIEW and TRIAGE as advisory, and describes a fallback when optional MCPs are absent. Deductions reflect word- and priority-based routing without sufficient evidence about misrouting handling, and environment guidance focused on Claude Code, Python 3.9+, and optional MCPs.
The README provides installation, updates, manual loading, environment variables, troubleshooting, and architecture details. The complete MIT LICENSE supports full marks for licensing. The supplied material has no separate changelog or maintenance policy; the version, repository component counts, and user-supplied description also disagree, reducing confidence in naming and version stability. Quick-start examples show intended flows but are not execution evidence.
The router, specialist agents, disk-backed workflow state, independent review, and verification offer a clear intended benefit for complex Claude Code tasks, and the examples make the expected outputs easier to understand. Deductions reflect that these benefits are asserted in the README rather than demonstrated by the static evidence. Global configuration changes, broad Python permission, and additional state files add setup and maintenance costs that are not quantified.
The README points to the router skill as the source of truth for workflow routing, and seam_eval.py explicitly says it performs text checks without running a builder; these clarify some evidence boundaries. However, the supplied description and README disagree on component counts, and seam_eval.py's assertion that text checks support correct behavioral decisions exceeds what its method establishes. No independent corroborating source is supplied, so traceability, corroboration, and fact-versus-inference separation receive only partial credit.
- The supplied description and README disagree on agent, skill, and workflow counts; the provided static material does not establish which counts apply to the target revision.
- The setup procedure recommends changing global CLAUDE.md and permissions; users should review the breadth of Bash(python3:*) and decide whether the global changes fit their needs.
- seam_eval.py says it checks text and does not run an agent, so its claim to guarantee correct decisions should not be treated as behavioral validation.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v12.8.0 v12.8.0 — cc10x-guide skill + documentation overhaul
Release notes excerpt
v12.8.0 — cc10x-guide skill + documentation overhaul User feedback drove a documentation-first release: the README was a reference manual instead of a landing page, and the plugin shipped no way to ask Claude about cc10x itself. Two behavioral fixes fell out of the setup-correctness audit. Added …View on GitHub ↗ -
v11.1.0 cc10x v11.1.0 — execution-engine harvest + 4 net-new skills
Release notes excerpt
Minor release — all changes are additive over v11.0.0. Driven by a system-level architecture comparison against superpowers + matt-pocock that honestly mapped where cc10x leads (back-half verification: independent verifier + REVERT + test-honesty gates, parallel silent-failure-hunter, fail-closed hooks) and where it trailed (the execution engine). This closes the engine gap without disturbing the per-phase build loop, which stays the default. Execution engine …View on GitHub ↗ -
v11.0.0 cc10x v11.0.0 — de-version + revitalization + upstream harvest
Release notes excerpt
First v11 release. Three campaigns, all validator-gated (4 validators green at every commit; full live harness 4/4 PASS). De-version the state namespace .cc10x/v10/ → .cc10x/. The version identifier now lives only in plugin.json and on GitHub — never in the on-disk state path. Fresh start: no automatic migration of existing .cc10x/v10/ state (orphaned and ignored on upgrade). Revitalization - Complexity gradient restored — trivial BUILD work runs a reduced graph; full chain for standard/planned work, with scope/blocked escalation. …View on GitHub ↗
FAQ
Do I need to install an MCP server?
How does it recover after context compaction?
.cc10x/. Session startup can rehydrate workflow context from those artifacts.Can I run it directly in ChatGPT or another agent platform?
What runtime does installation require, and what does it cost?
Will file writes require permission prompts?
Edit(.cc10x/**) to Claude Code’s permissions.allow to reduce prompts for state files. Writes to outward-facing project documentation still request approval by design.