Skip to content
@control-decks

Control Decks

Composable card decks for explicit, human-directed control of AI agents.

Control Decks

One object. Multiple decks.

Control Decks gives humans a small, explicit interface for directing agent behavior. Each deck has its own job. Every compatible card can receive the same Working Object, add a result or scoped annotation, and pass it to the next deck.

HACP — Human-Agent Control Protocol
Cards are the interface. Control is the protocol.

Think It Through Work This Way Reality Check
Think It Through
Explore, decide, plan, and explain.
Work This Way
Set visible session controls and act.
Reality Check
Verify sources, assumptions, and concrete targets.

Cards compose across decks

These are recipes, not dependencies between decks. Cards compose through HACP kinds, predicates, traits, and annotations rather than foreign card IDs.

THINK PROPOSE → REALITY CHECK ASSUMPTIONS → REALITY CHECK SOURCES

A proposal becomes the Working Object. Reality Check annotates its assumptions and supporting evidence without replacing the proposal.

WORK ASK FIRST
→ THINK TO PLAN
→ REALITY CHECK SOURCES
→ WORK IMPLEMENT
→ THINK EXPLAIN

One object moves through planning, evidence checking, guarded execution, and presentation. The user sees one useful final result, not five disconnected reports.

The shared grammar

Binding + Working Object + Active Controls
  • binding chooses what the cards apply to.
  • control activates, qualifies, or clears visible state.
  • operation transforms, annotates, or acts on the object.
  • presentation changes representation without changing substance.

Cards resolve in written order. The full stream is prevalidated before its first effect. No hidden reordering, hidden duration card, or implicit permission grant is allowed.

Install HACP once

The HACP plugin loads the shared Draft 0.4 payload when a session starts. Cards remain explicit human commands; the agent cannot infer, invoke, or replay them.

# Codex
codex plugin marketplace add control-decks/human-agent-control-protocol
codex plugin add hacp@hacp

# Claude Code
claude plugin marketplace add control-decks/human-agent-control-protocol --scope user
claude plugin install hacp@hacp --scope user

Then install only the decks you want.

Decks

Repository Current release Purpose
Think It Through 0.13.0 Thinking operations, bindings, and presentation
Work This Way 0.3.0 Session controls and implementation
Reality Check 0.3.1 Read-only verification annotations
HACP Draft 0.4 · plugin 0.3.0 Required session adapter and authoring plugin

Each deck is an independent Codex and Claude Code plugin, but full HACP conformance requires a compatible session adapter. The canonical HACP plugin loads that payload once and also explains, creates, and audits cards.

Start here

  • Want to think something through? Install Think It Through.
  • Want to change how the agent works in this session? Install Work This Way.
  • Want to verify a result before relying on it? Install Reality Check.
  • Want to build a compatible deck? Read HACP Draft 0.4 or play HACP AUTHOR CARD and HACP AUTHOR DECK.

Created by thevzion.

Pinned Loading

  1. think-it-through think-it-through Public

    A conversation-card deck for exploring, challenging, planning, and explaining ideas with AI.

    3

  2. human-agent-control-protocol human-agent-control-protocol Public

    The semantic protocol behind Control Decks. Cards are the interface. Control is the protocol.

    JavaScript 2

  3. reality-check reality-check Public

    A verification deck for checking sources, assumptions, and concrete targets without replacing the result.

  4. work-this-way work-this-way Public

    A control deck for setting visible, session-scoped rules for how AI agents work.

Repositories

Showing 5 of 5 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…