Dev & Engineering clitaste-learningcoding-agentworkflowsinteractive-mode

Command Code

The coding agent that learns your coding taste

FollowAgents review · FARS-2.1
Not recommended
25/ 100 5-point scale 1.3 / 5
1 2 3 4 5 6
1Trust0 / 29 · 0.0/5

Evidence shows: The repository provides no information on permission model, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. Therefore all trust criteria score 0. Deduction: lack of documentation or implementation for these aspects.

2Reliability3 / 14 · 1.1/5

Evidence shows: README descriptions are consistent with documentation links, but no dependency availability or failure messages are provided. Self-consistency scores 1 for consistency but lack of detail; dependency availability scores 1 for no mention; failure messages scores 0 for no mention.

3Adaptability8 / 18 · 2.2/5

Evidence shows: README clearly identifies target audience (developers) and scenarios (build, fix, refactor), but does not specify capability boundaries, trigger precision, or environment requirements. Audience and scenarios score 2 for clarity; capability boundaries score 1 for lack; trigger precision score 1 for lack; environment fit score 1 for lack.

4Convention6 / 18 · 1.7/5

Evidence shows: README provides clear documentation structure, installation instructions, and version badge, but lacks known limitations, license, changelog, and maintenance responsibility. Information architecture scores 2 for clarity; install notes score 2 for npm command; naming stability scores 1 for no mention; examples and FAQ score 1 for none; known limitations score 0; license score 0; versioning/changelog score 1 for version badge; maintenance responsibility score 1 for no explicit.

5Effectiveness6 / 13 · 2.3/5

Evidence shows: README describes outputs (build software, fix bugs, etc.) but lacks concrete examples or cost-benefit analysis. Output usability scores 1 for vague description; marginal value scores 2 for emphasis on learning; cost-benefit scores 1 for none.

6Verifiability2 / 8 · 1.3/5

Evidence shows: Claims (e.g., 'first frontier coding agent') lack evidence or sources. Claim traceability scores 1 for some links; cross-source corroboration scores 0 for none; fact-inference separation scores 1 for partial distinction.

Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 5c8f1b48c9d6
Safety controls not found in source: least-privilege scoping, confirmation before acting, data-flow disclosure, sensitive-data handling, dependency security, disclosed external effects, rollback or recovery path, verifiable attribution
Before you use it
  • The repository lacks license information; verify legal compliance before use.
  • No security documentation (permissions, data flow, dependency security) is provided; assess risks before use.
  • No known limitations or failure handling are documented; unexpected issues may arise.
Review evidence [1]
See the full review method →

What does this agent do, and when should you use it?

Command Code is a terminal-based coding agent that builds software and continuously learns your coding taste. It ships full-stack projects, features, fixes bugs, writes tests, and refactors, all while learning from your accept, reject, and edit actions. Built on a Meta Neuro-Symbolic AI model called taste-1, it enforces the invisible logic of your preferences. It offers interactive mode with slash commands, Bash mode, and file autocompletion. Taste profiles can be shared across teams via `npx taste push/pull`. The tool is distributed as an npm package, installed globally, and run inside projects. Documentation and community resources are available.

Command Code is a CLI tool installed via npm and launched with cmd in a project directory. It reads project files, executes terminal commands, invokes AI models (supported providers include Anthropic, OpenAI, DeepSeek, GLM, Kimi) to generate and modify code. The centerpiece is the taste-1 model, which records user actions (accept/reject/edit) to build a taste profile that influences subsequent outputs. The interactive session supports slash commands, Bash mode, and file path autocomplete with @. Users can provide feedback via /feedback command.

  1. A developer working on a long-term project who wants an AI assistant that adapts to their personal coding style over time.
  2. A team that wants consistent code style across members; they can share taste profiles using npx taste push/pull.
  3. A developer needing to quickly scaffold a full-stack project or feature; Command Code generates code while learning preferences.
  4. An early adopter willing to try new AI tools and engage with the community via Discord for support and feedback.

What are this agent's strengths and limitations?

Pros
  • Continuously learns from your editing habits, personalizing code generation.
  • Built on the taste-1 model, a Meta Neuro-Symbolic AI that enforces subtle coding preferences.
  • Interactive mode with slash commands, Bash mode, and file path autocomplete.
  • Shareable taste profiles across teams via npx taste push/pull.
Limitations
  • Requires Node.js and global npm installation.
  • Depends on external model providers like Anthropic and OpenAI; API keys may be necessary but not detailed.
  • No explicit license, which may raise concerns for commercial adoption.
  • The taste-learning mechanism is relatively new; effectiveness may require validation.

How do you install or deploy this agent?

Install Command Code globally:

npm i -g command-code

Requires Node.js.

How do you use this agent?

Start in your project directory:

cd your-project
cmd

In the session, type / to open the command menu, ! for Bash mode, and @ for file path autocomplete. Refer to the quickstart guide for onboarding.

FAQ

Which model API keys do I need?
The README does not specify, but it likely requires API keys for providers like Anthropic or OpenAI, as implied by the supported models.
How are taste profiles stored and synced?
Use npx taste push/pull to share profiles, but the storage location is not documented.
Can I use it offline?
It requires network to call AI models; no offline mode is indicated.
How can I report issues?
Use the /feedback command or open a GitHub issue.

Related agents