Agent Skills

Agent Skills package Squasher-specific workflows for AI coding assistants. Use them when an assistant needs durable context for onboarding, the Squasher CLI, the hosted MCP server, or observability workflows. Skills are helper documentation for agents, not the product contract. The stable contracts are the public API, CLI behavior, MCP tool schemas, and Code Mode operation types. Install skills to help an agent choose the right Squasher surface and avoid stale setup or query guidance.

Available skills

Skill	Use when	What it prevents
`squasher-onboard`	The user wants to install Squasher SDKs, log drains, OTLP, AI telemetry, or setup verification in a repo.	Partial installs, wrong public package names, hardcoded keys, skipped services, and fake verification.
`squasher-telemetry-audit`	The user wants to find and repair weak, noisy, costly, or unsafe application telemetry.	Unapproved edits, secret exposure, unbounded queries, and volume increases used as a debugging shortcut.
`squasher-cli`	The user asks for `squasher` commands, says “squash CLI”, or wants to inspect Squasher data from a terminal.	Invented `squash` commands, stale flags, missing `--project` scoping, and unsafe API-key handling.
`squasher-mcp`	The user wants to connect Squasher to Claude Code, Cursor, VS Code, Codex CLI, or another MCP client.	Wrong MCP URLs, incorrect OAuth/API-key setup, made-up tool names, and unbounded observability queries.

Direct skill URLs

Agents can read all four skills and their linked reference files directly:

https://skills.squasher.ai/squasher-onboard/SKILL.md
https://skills.squasher.ai/squasher-telemetry-audit/SKILL.md
https://skills.squasher.ai/squasher-cli/SKILL.md
https://skills.squasher.ai/squasher-mcp/SKILL.md

The dashboard setup prompt includes this URL after signup so a coding agent can map the repo, choose the right Squasher SDK or integration, make the code changes, and smoke-check the setup.

One-command setup

For local agent environments, install the Squasher skills and register the hosted Squasher MCP server in one command:

bash <(curl -fsSL https://squasher.ai/install.sh) --agents -y

The installer adds squasher-onboard, squasher-telemetry-audit, squasher-cli, and squasher-mcp from the public skills source. It also registers https://mcp.squasher.ai/mcp for detected agent CLIs such as Codex CLI, Claude Code, Gemini CLI, and Amp. Omit -y to approve each command, or use --skills-only / --mcp-only for a narrower install.

Install

Install from the public skill index. No GitHub access is required.

npx skills add https://skills.squasher.ai --skill squasher-onboard squasher-telemetry-audit squasher-cli squasher-mcp

The index at /.well-known/agent-skills/index.json lists every skill and its reference files. The installer downloads the same files served by the direct URLs. For project-local installation without a package manager, copy the skill folders into your repository:

.agents/
`-- skills/
    |-- squasher-onboard/
    |   |-- SKILL.md
    |   `-- references/
    |-- squasher-telemetry-audit/
    |   |-- SKILL.md
    |   `-- references/
    |-- squasher-cli/
    |   |-- SKILL.md
    |   `-- references/
    `-- squasher-mcp/
        |-- SKILL.md
        `-- references/

If you already have a local checkout that contains these folders, install from the local path:

npx skills add ./.agents/skills/squasher-onboard
npx skills add ./.agents/skills/squasher-telemetry-audit
npx skills add ./.agents/skills/squasher-cli
npx skills add ./.agents/skills/squasher-mcp

Use the skills

After installation, start a new agent session and ask for the skill explicitly:

Use $squasher-onboard to install Squasher in this repo and verify one real runtime event.

Use $squasher-telemetry-audit to review this service's telemetry. Show findings and a repair plan before you edit files.

The telemetry audit starts with a read-only review and requires approval for its repair plan. It does not enable future capture, increase sampling, retention, or log levels, start paid model runs, or generate synthetic production traffic.

Use $squasher-cli to show me the command for listing unresolved errors in this project.

Use $squasher-mcp to connect Claude Code to Squasher with OAuth.

Most assistants also activate a skill implicitly when your prompt matches its description. Explicit invocation is still useful when you are testing a new install.

Verify

Run your assistant’s skills command, if available:

/skills

Confirm that squasher-onboard, squasher-telemetry-audit, squasher-cli, and squasher-mcp appear in the list. Then test with a prompt that should not require live project data:

What public package should I install for the Squasher Node SDK?

The answer should include:

@squasher-ai/node

Get Started

SDKs

Integrations

Migrations

Cloud Log Drains

Features

API Reference

Available skills

Direct skill URLs

One-command setup

Install

Use the skills

Verify

​Available skills

​Direct skill URLs

​One-command setup

​Install

​Use the skills

​Verify

​Related docs

Available skills

Direct skill URLs

One-command setup

Install

Use the skills

Verify

Related docs