intelligent-terminal: Skill for Claude Code

Add ACP Agent Support

Integrate a new built-in agent across WTA, Terminal UI, settings, policy, telemetry, documentation, and validation without leaving agent-specific behavior inconsistent between layers.

When to Use This Skill

• Add a native ACP-speaking agent CLI to Intelligent Terminal.
• Add an ACP adapter for a CLI that does not speak ACP itself.
• Promote a custom ACP command into a first-class built-in agent.
• Add or repair ACP authentication, model probing, delegate, icon, or policy support for an existing agent.
• Add or repair session-tracking hooks, hook onboarding, Settings status, or hook runtime behavior for a built-in agent.

Prerequisites

• Work on a feature branch, never directly on main or master.
• Read the repository and area-specific instructions before editing, including tools/wta/AGENTS.md and the Rust instruction files for WTA changes.
• Obtain the agent's official CLI documentation and install a pinned or known version for live verification.
• Confirm that the CLI either exposes a long-running ACP server over stdio or has a maintained ACP adapter. A normal interactive TUI is not an ACP server.

Workflow

Track the work with a TODO list because the Rust and C++ registries, UI, policy, tests, and live verification must stay synchronized.

1. Define the agent contract. Complete the capability matrix in integration-map.md: executable, ACP command, native-vs-adapter ownership, auth, model selection, interactive delegate syntax, session resume, installation, and official branding.
2. Prove ACP behavior before editing. Run the exact ACP command and verify stdio initialization, session creation, model discovery, prompt streaming, cancellation, and shutdown. Record the tested CLI/adapter version.
3. Inventory current integration surfaces. Search for every existing built-in agent ID and follow the current patterns. Do not rely on old line numbers or assume the Rust registry is the only source of truth.
4. Implement the WTA profile and command behavior. Add the profile, ACP launch command, auth flow, model behavior, install guidance, and resume metadata. Wire the canonical ID through the session subsystem when session listing or resume is supported. Add delegate support only when the agent has a true interactive initial-prompt invocation.
5. Wire Terminal surfaces. Keep C++ built-in registries, ACP command resolution, settings/model probing, telemetry sanitization, branding, and policy-facing lists consistent. Use the detailed file map in integration-map.md.
6. Decide and implement session hooks. Treat hooks as a separate capability: verify the CLI's official hook/plugin API, map lifecycle events to WTA, suppress hook emission from the ACP process to avoid duplicate sessions, and wire safe install/status/uninstall/upgrade behavior through onboarding and Settings. If the CLI has no suitable hook API, document session tracking as unsupported instead of inferring it from ACP support.
7. Add focused tests. Cover profile lookup, ACP command construction, identification, session source round-trips and resume dispatch, auth command generation, hook lifecycle and ownership behavior when implemented, and every delegate shell path that applies: direct Windows, PowerShell 7, Windows PowerShell 5.1, and WSL.
8. Update user and administrator documentation. Document support, installation/auth requirements, limitations, delegate behavior, and the AllowedAgents identifier. Do not advertise hooks or history integration unless they were implemented and tested.
9. Validate end to end. Follow validation.md, including the WTA test suite, explicit-target WTA build, Terminal build, package deployment, live ACP prompt, auth path, model selection, hook-backed session tracking when supported, delegate behavior, GPO filtering, and log inspection.
10. Prepare the PR. Keep the diff scoped, cite the tested agent version and ACP command, describe native-vs-adapter and hook support, list unsupported features, and link the tracking issue with a closing keyword when appropriate.