Approval gates ask “may I?”
Decision requests ask “which way?”
An AI coding agent needs a person at two different moments. One is before it does something risky. The other is when it reaches a choice nobody delegated to it. Tools for the first are common. CruxDeck is built for the second.
Draft. CruxDeck is not yet available. This page describes it as it will be at launch.
| Aspect | Approval gate | Decision request (CruxDeck) |
|---|---|---|
| The question | May I run this command, write this file, merge this change? | Which of these options should we take, and why? |
| The answer | Yes or no. | A chosen option or a written answer, with any explanation. |
| Who answers | The developer running the agent, or a rule set in advance. | The person who owns the call, such as a tech lead, architect, engineering manager, product owner or CTO. |
| What the agent sends | The action it is about to take. | The question and its background, with options and their pros and cons, its recommendation, and screenshots or files when it includes them. |
| Where it is answered | Wherever the tool asks: the agent’s own session, a chat message or an email. | In the team’s Slack or Microsoft Teams channel, once connected, where people with access to the channel can see it, or in CruxDeck. |
| When | Before a risky action runs. | When the agent reaches a judgment call it should not make alone. |
| What is kept | Whatever the tool logs. | A decision record, with any explanation, that the team and later agents can read back. |
They answer different needs
An approval gate stops an agent doing something risky. A decision request gives an agent somewhere to send a choice that is someone else’s to make, rather than settle it quietly.
Approval gates are already built in
Claude Code’s permission settings and hooks, and Codex’s approval and sandbox settings, decide what an agent may do without asking and what it can reach. For risky actions, they are the first line.
Choices are where the courier work is
Without a place to send a choice, the developer copies the question into Slack or a ticket, chases the answer, and pastes it back into the session. The options and the reasoning are easily trimmed on the way.
How CruxDeck fits
Through MCP
Claude Code or Codex files a Decision Request through the CruxDeck MCP server, an open-source server any coding agent that can run a local MCP server can use.
A person decides, where the team works
Once the team connects Slack or Microsoft Teams and chooses a channel, the request is posted there, and it can be assigned to the reviewer who owns the call. A person chooses an option or answers, from the thread or in CruxDeck. Only a person can resolve a request.
The agent carries on
The agent reads the outcome back through MCP and continues. The decision stays on record.
When CruxDeck is not the right tool
A yes or no on a command
Use your agent’s own permission settings, or a gateway in front of the systems it can reach.
A team vote
A person resolves each request. CruxDeck does not run polls or count votes.
Planning work
CruxDeck is not a tracker. Tasks and backlogs stay in GitHub or Jira, and decisions link to them.
For tech leads, architects and CTOs: what reaches you, and what you do. Setting an agent up to ask: the escalation guide.
Bring your judgment calls to one place.
CruxDeck isn’t available yet. We’d like to hear how your team handles agent decisions today.