Amplifier

Delegate complex tasks to Amplifier's multi-agent framework. Use when: (1) research/comparison needing multiple perspectives, (2) multi-file code projects, (...

Security Scan

The skill is internally consistent: its name/description match the CLI it requires and the install points to a GitHub package that produces the declared binary, but there are a few operational/installation details you should verify before installing.

Purpose & Capability

The skill advertises delegation to Amplifier and requires the amplifier-openclaw binary; the install entry creates that binary from a GitHub package. Required binaries, bundles, and example commands align with the described purpose.

Instruction Scope

SKILL.md instructs the agent to run the amplifier-openclaw CLI (including background runs, sessions, and model selection). It does not instruct the agent to read unrelated files or credentials, but it assumes the CLI will be available and configured. The file delegates execution to a local binary and instructs killing background processes — normal for a CLI integration but gives the skill control to spawn persistent background work.

Install Mechanism

Install uses uv to fetch a package from a GitHub URL (git+https://github.com/microsoft/amplifier-app-openclaw@v1.0.5) which produces the named binary. Installing code from a GitHub package is common and the host is a well-known domain (github.com/microsoft), but it still writes/extracts code and creates a binary on disk, so review the upstream repo or run in a sandbox if you need strong assurance.

Credentials

The skill declares no required environment variables or credentials, which is proportionate to the textual instructions. However, the CLI supports a --model flag and will likely read its own config or model/API credentials from the environment or config files at runtime; those credentials are not declared here. Expect the underlying tool might require separate API keys or config outside the skill manifest.

Persistence & Privilege

always:false and normal autonomous invocation settings are used. The skill runs and may spawn background processes via the CLI (documented in SKILL.md), but it does not request permanent inclusion or modify other skills' configs.

Assessment

This skill appears to do what it says: it drives a local Amplifier CLI. Before installing, verify the upstream package/repo (https://github.com/microsoft/amplifier-app-openclaw@v1.0.5) so you trust the code that will be installed. Understand that the installer will write a binary to disk and the CLI can run background processes; run the install in a controlled environment if you are cautious. Check whether the CLI requires model/API credentials or config files (the skill doesn’t declare any), and avoid running the installer as root. If you need higher assurance, inspect the repository/tag, prefer an official release artifact or signed release, or run the CLI in a container or sandbox first.

Amplifier — Multi-Agent Delegation

Amplifier is a multi-agent AI framework. Delegate tasks that benefit from specialist agents, structured workflows, or parallel investigation.

When to Delegate

High confidence → delegate immediately:

• "Research X and compare approaches"
• "Build a Python tool that does X"
• "Review this code for security and design"
• User says "amplifier", "deep dive", "thorough", "comprehensive"
• Task has clear subtasks benefiting from parallel agents

Medium confidence → offer the choice:

• "I can do a quick analysis, or delegate to Amplifier for a thorough multi-agent review."

Low confidence → handle yourself:

• Simple Q&A, quick code edits, casual conversation, anything needing immediate response

Usage

Basic Delegation

exec command:"amplifier-openclaw run 'Research the top 3 Python web frameworks' --bundle foundation" background:true timeout:600

With Model Selection

Pass --model to override the default model:

exec command:"amplifier-openclaw run --model your-preferred-model 'Deep code review' --bundle foundation" background:true timeout:600

Tip: Pass --model matching the model from your OpenClaw Runtime line so Amplifier uses the same one.

Bundles

amplifier-openclaw bundles list

Bundle	Best For
foundation	General: research, analysis, planning (default)
superpowers	Multi-agent brainstorm, deep investigation
coder	Code generation, refactoring, debugging

Session Persistence

# Start a named session
exec command:"amplifier-openclaw run --session-name my-project 'Start building the auth module' --bundle foundation" background:true

# Resume later
exec command:"amplifier-openclaw run --resume --session-name my-project 'Now add unit tests'" background:true

Modes

Amplifier supports slash-command modes in prompts. Modes do not carry over between runs — include the mode at the start of each prompt:

# Brainstorm mode (uses all agents)
exec command:"amplifier-openclaw run --bundle superpowers '/brainstorm How should we architect the new API?'" background:true

# Research mode
exec command:"amplifier-openclaw run --bundle foundation '/research Latest advances in RAG'" background:true

JSON Output

{
  "response": "The analysis found...",
  "usage": {
    "input_tokens": 28566,
    "output_tokens": 1800,
    "estimated_cost": 0.12,
    "tool_invocations": 3
  },
  "status": "completed"
}

Cost Tracking

exec command:"amplifier-openclaw cost --period week"

Report costs only when asked or when notable (>$1).

Interpreting Results

• response: Present to the user (the main output)
• error: Report in plain language, don't dump raw JSON
• usage.estimated_cost: May be 0.0 — don't alarm about zero
• status: "completed", "cancelled", or error state

During Active Delegation

• "stop"/"cancel" → kill the background process
• Unrelated questions → answer yourself, don't interrupt Amplifier
• Follow-up → tell user you'll pass it along when current task finishes

Install

If not already installed:

uv tool install "amplifier-app-openclaw @ git+https://github.com/microsoft/amplifier-app-openclaw@v1.0.5"