> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pandaprobe.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Skill

> Teach AI coding agents to trace, evaluate, and debug with PandaProbe

The **PandaProbe Agent Skill** teaches AI coding agents — Claude Code, Cursor, and other
tools that support the open [Agent Skills](https://docs.claude.com/en/docs/claude-code/skills)
standard — how to work with PandaProbe. With it installed, your agent knows how to set up
PandaProbe, instrument your app, and drive the `pandaprobe` CLI without you spelling out the
details every time.

It's open source and lives at
[github.com/chirpz-ai/pandaprobe-skills](https://github.com/chirpz-ai/pandaprobe-skills).

## Why use it

Coding agents produce significantly better results with the skill installed, because it
conditions them to follow PandaProbe's best practices — work doc-first instead of guessing
APIs from memory, use the CLI for data access instead of hand-building requests, and never
echo API keys into the chat.

The skill is self-contained: a `SKILL.md` entry point plus focused reference docs.

```
SKILL.md
references/
  ├── setup.md            # guided onboarding playbook
  ├── instrumentation.md  # SDK tracing across frameworks and providers
  └── cli.md              # the pandaprobe CLI command tree and recipes
```

It uses a **progressive disclosure** model to keep context usage low. The agent only sees the
skill's name and description until a task matches; it then reads `SKILL.md`, and pulls in a
specific reference file (`setup.md`, `instrumentation.md`, or `cli.md`) only when that topic
comes up.

## What it can do

<CardGroup cols={2}>
  <Card title="Onboard a project" icon="rocket">
    Guided, interactive setup that builds and runs a first traced agent example from scratch
    and verifies the traces landed — ideal for a new or empty project.
  </Card>

  <Card title="Instrument your app" icon="route">
    Add PandaProbe tracing via an agent-framework integration, a provider wrapper, or manual
    decorators — working doc-first against the live documentation.
  </Card>

  <Card title="Read & evaluate data" icon="terminal">
    Use the `pandaprobe` CLI to inspect traces, sessions, spans, and scores, and to run
    evaluations.
  </Card>

  <Card title="Look up the docs" icon="book-open">
    Fetch the latest PandaProbe documentation — the source of truth that all instrumentation
    relies on.
  </Card>
</CardGroup>

## Install

The skill is distributed through [`npx skills`](https://github.com/vercel-labs/skills), the
open agent-skills CLI.

<Tabs>
  <Tab title="Ask your coding agent">
    The simplest path is to ask your coding agent to install it. Paste a prompt like:

    > Install the PandaProbe agent skill from the `chirpz-ai/pandaprobe-skills` GitHub
    > repository.
  </Tab>

  <Tab title="Manual Installation">
    Pick the scope that fits your workflow:

    ```bash theme={null}
    # Local — install into the current project (all detected agents):
    npx skills add chirpz-ai/pandaprobe-skills --skill '*' --yes

    # Global — install for all your projects (all detected agents):
    npx skills add chirpz-ai/pandaprobe-skills --skill '*' --yes --global

    # Target a specific agent (e.g. Claude Code), installed globally:
    npx skills add chirpz-ai/pandaprobe-skills --agent claude-code --skill '*' --yes --global
    ```

    `--skill '*'` selects every skill in the repo (there is one, `pandaprobe`). Omit
    `--agent` to install to all detected agents, or pass `--agent <name>` to target one.

    <Accordion title="Install from a local checkout">
      To test changes before pushing, install from a local clone:

      ```bash theme={null}
      npx skills add ./ --agent claude-code
      ```
    </Accordion>
  </Tab>
</Tabs>

## Prerequisites

<Note>
  The skill installs the instructions, not the tools. Depending on what you ask it to do,
  you'll also need:
</Note>

* **The `pandaprobe` CLI** — for data and evaluation access. Install with:
  ```bash theme={null}
  curl -fsSL https://cli.pandaprobe.com/install.sh | sh
  ```
* **A PandaProbe account and API key** — sign up at
  [app.pandaprobe.com](https://app.pandaprobe.com), then run `pandaprobe auth login`.
  Self-hosted deployments can configure an endpoint and key manually instead.
* **Python 3.10+** with `pip install pandaprobe` — for SDK instrumentation.

## Usage examples

Once installed, the agent picks the skill up automatically for prompts like:

* "Set up PandaProbe in this project and show me my first traces."
* "Add PandaProbe tracing to my LangGraph agent."
* "Use pandaprobe to list my failed traces, then drill into the error in trace `tr_…`."
* "Run a coherence eval over yesterday's completed traces."

## Resources

<CardGroup cols={2}>
  <Card title="Skill repository" icon="github" href="https://github.com/chirpz-ai/pandaprobe-skills">
    The open-source `pandaprobe-skills` repo with the full `SKILL.md` and references.
  </Card>

  <Card title="CLI reference" icon="terminal" href="/introduction/cli">
    Install, authenticate, and use the `pandaprobe` CLI.
  </Card>

  <Card title="Tracing overview" icon="route" href="/tracing/overview">
    The three layers of instrumentation the skill uses.
  </Card>

  <Card title="Evaluation" icon="drafting-compass" href="/evaluation/get-started/concepts">
    How metrics, scores, and monitors work.
  </Card>
</CardGroup>