Add Agent Computer delegation skills

Co-authored-by: Codex <noreply@openai.com>

README.md

@@ -0,0 +1,38 @@
+# agentcomputer-delegate
+
+Skills for controlling Agent Computer machines and remote agent sessions through the `aicomputer` CLI.
+
+## Install
+
+Install all skills:
+
+```bash
+npx skills add harivansh-afk/agentcomputer-delegate -g -y
+```
+
+Install only the main fleet-control skill:
+
+```bash
+npx skills add harivansh-afk/agentcomputer-delegate -g -y --skill computer-fleet
+```
+
+Install only the ACP bridge skill:
+
+```bash
+npx skills add harivansh-afk/agentcomputer-delegate -g -y --skill computer-acp
+```
+
+## Skills
+
+- `computer-fleet`: Use `computer` / `aicomputer` to list machines, create remote agent sessions, prompt them, watch progress, cancel, interrupt, and close.
+- `computer-acp`: Use `computer acp serve` to expose a remote machine agent as a local ACP-compatible stdio command.
+
+## Notes
+
+- These skills assume the `aicomputer` CLI is installed or can be installed with `npm install -g aicomputer`.
+- Authentication uses the normal browser flow: `computer login`.
+- The recommended install source is this repo directly:
+
+```bash
+npx skills add harivansh-afk/agentcomputer-delegate
+```

skills/computer-acp/SKILL.md

@@ -0,0 +1,79 @@
+---
+name: computer-acp
+description: Bridge a remote Agent Computer machine agent into a local ACP-compatible stdio process using `computer acp serve`. Use when a local agent client or ACP tool needs to talk to Claude or another remote machine agent through one local command.
+allowed-tools: Bash(npm:*), Bash(npx aicomputer@latest:*), Bash(aicomputer:*), Bash(computer:*), Bash(acpx:*)
+---
+
+# Computer ACP
+
+Use this skill when a local ACP-capable client should treat a remote machine agent as if it were a local ACP server.
+
+## What this does
+
+`computer acp serve` starts a local stdio ACP bridge. The remote machine stays in the cloud, but the local client speaks ACP to the bridge process.
+
+That means you can point an ACP client at a command like:
+
+```bash
+computer acp serve <machine> --agent claude --name review
+```
+
+## Setup
+
+1. Ensure the CLI exists.
+
+```bash
+if command -v computer >/dev/null 2>&1; then
+  COMPUTER=computer
+elif command -v aicomputer >/dev/null 2>&1; then
+  COMPUTER=aicomputer
+else
+  npm install -g aicomputer
+  COMPUTER=computer
+fi
+```
+
+2. Verify auth.
+
+```bash
+$COMPUTER whoami || $COMPUTER login
+```
+
+3. Pick a machine and agent.
+
+```bash
+$COMPUTER ls --json
+$COMPUTER agent agents <machine> --json
+```
+
+## Start the bridge
+
+```bash
+$COMPUTER acp serve <machine> --agent claude --name review
+```
+
+Keep that process running. The local ACP client should launch or connect to this command over stdio.
+
+## Recommended pattern
+
+- Use one stable machine per bridge.
+- Use one stable session name per task thread.
+- Prefer a named session like `review` or `debug` so reconnects stay simple.
+- Use the normal `computer-fleet` skill if you only need CLI control and not ACP interoperability.
+
+## Example
+
+If `acpx` is installed, a typical shape is:
+
+```bash
+acpx --agent "computer acp serve <machine> --agent claude --name review" "fix the failing tests"
+```
+
+## Limits
+
+- This bridge is for agent chat/session control, not direct filesystem or terminal passthrough.
+- If the bridge process exits, the ACP client loses the connection and should reconnect by starting the command again.
+
+## Protocol notes
+
+Read [references/acp-flow.md](references/acp-flow.md) for the supported ACP flow and expected behavior.

skills/computer-acp/references/acp-flow.md

@@ -0,0 +1,23 @@
+# ACP Flow
+
+The `computer acp serve` bridge is intended to feel like a local ACP agent command while proxying work to a remote machine agent.
+
+## Basic shape
+
+1. The local client starts `computer acp serve <machine> --agent <agent> --name <session>`.
+2. The bridge handles ACP initialization on stdin/stdout.
+3. The bridge maps ACP session operations onto Agent Computer session APIs.
+4. Remote session updates are streamed back as ACP `session/update` notifications.
+
+## Good commands
+
+```bash
+computer acp serve yinnar-south --agent claude --name review
+computer acp serve airport --agent claude --name debug
+```
+
+## Recommended client behavior
+
+- Reuse a stable session name when reconnecting.
+- Treat the bridge as the single local command for remote-agent interaction.
+- Use the normal `computer agent ...` commands outside ACP when you need manual inspection or cleanup.

skills/computer-fleet/SKILL.md

@@ -0,0 +1,102 @@
+---
+name: computer-fleet
+description: Control Agent Computer machines and remote agent sessions from a local agent using the aicomputer CLI. Use when you need to list machines, inspect installed agents, create or resume remote sessions, send prompts, stream progress, cancel or interrupt work, or close sessions cleanly.
+allowed-tools: Bash(npm:*), Bash(npx aicomputer@latest:*), Bash(aicomputer:*), Bash(computer:*)
+---
+
+# Computer Fleet
+
+Use the `aicomputer` CLI as the control plane for remote machine agents.
+
+## When to use
+
+Use this skill when the user wants to:
+
+- work on one of their Agent Computer machines from a local agent
+- delegate coding or research work to a remote Claude/Codex-style agent
+- inspect remote agent status or stream progress
+- stop or clean up remote agent sessions
+
+Do not use SSH, VNC, or browser URLs when simple agent control is enough.
+
+## Quick start
+
+1. Ensure the CLI exists.
+
+```bash
+if command -v computer >/dev/null 2>&1; then
+  COMPUTER=computer
+elif command -v aicomputer >/dev/null 2>&1; then
+  COMPUTER=aicomputer
+else
+  npm install -g aicomputer
+  COMPUTER=computer
+fi
+```
+
+2. Verify auth.
+
+```bash
+$COMPUTER whoami || $COMPUTER login
+```
+
+3. List running machines.
+
+```bash
+$COMPUTER ls --json
+```
+
+4. Inspect available agents on a machine.
+
+```bash
+$COMPUTER agent agents <machine> --json
+```
+
+5. Create or reuse a named session.
+
+```bash
+$COMPUTER agent sessions new <machine> --agent claude --name review --json
+```
+
+6. Send prompts and stream output.
+
+```bash
+$COMPUTER agent prompt <machine> "fix the failing tests" --agent claude --name review
+```
+
+## Default workflow
+
+Prefer this flow:
+
+1. Pick a running machine from `computer ls --json`.
+2. Check `computer agent agents <machine> --json` and choose an installed agent with credentials.
+3. Reuse a stable session name for a task thread like `review`, `debug`, or `migration`.
+4. Use `computer agent prompt ...` for the work.
+5. Use `computer agent status ...` or `computer fleet status --json` when you need current state.
+6. Use `computer agent cancel ...` for normal stop.
+7. Use `computer agent interrupt ...` only if normal cancel is not enough.
+8. Use `computer agent close ...` when the session is no longer needed.
+
+Prefer `--json` whenever you need to parse output programmatically.
+
+## Session naming
+
+Use short stable names:
+
+- `review`
+- `debug`
+- `backend`
+- `frontend`
+- `release`
+
+Reuse the same session name when continuing the same line of work on the same machine. Start a new name when the task changes meaningfully.
+
+## Cancellation rules
+
+- `cancel` is the normal stop path.
+- `interrupt` is the hard stop path when the session does not settle quickly enough.
+- `close` is cleanup after the task is done or abandoned.
+
+## Commands reference
+
+Read [references/cli-cheatsheet.md](references/cli-cheatsheet.md) when you need the exact command set.

skills/computer-fleet/references/cli-cheatsheet.md

@@ -0,0 +1,51 @@
+# CLI Cheatsheet
+
+## Authentication
+
+```bash
+computer whoami
+computer login
+```
+
+## Machine discovery
+
+```bash
+computer ls --json
+computer fleet status --json
+```
+
+## Agent discovery
+
+```bash
+computer agent agents <machine> --json
+```
+
+## Sessions
+
+```bash
+computer agent sessions list <machine> --json
+computer agent sessions new <machine> --agent claude --name review --json
+computer agent status <machine> --session <session_id> --json
+```
+
+## Prompting
+
+```bash
+computer agent prompt <machine> "inspect the repo and summarize the failing tests" --agent claude --name review
+computer agent prompt <machine> "continue from the previous task" --session <session_id>
+```
+
+## Streaming and control
+
+```bash
+computer agent watch <machine> --session <session_id>
+computer agent cancel <machine> --session <session_id> --json
+computer agent interrupt <machine> --session <session_id> --json
+computer agent close <machine> --session <session_id>
+```
+
+## Good defaults
+
+- Prefer machine handles over machine ids when both are available.
+- Prefer `--name` for human-meaningful persistent sessions.
+- Prefer `--json` when another program or agent needs to read the result.