Agent Skills

AI-first observability. Your agent does the debugging.

Your agent sets up TracePath, queries production telemetry, and finds the root cause.

Works with Claude Code, Cursor, Codex, and any agent that reads SKILL.md.

View on GitHub →

See it in action

Watch an agent set up TracePath, pull production telemetry through the CLI, and walk a bug all the way to its root cause.

0:00 / 0:00
Debug

From bug report to root cause

Paste the bug report as written. The agent pulls the matching exceptions, the logs around the failure, and endpoint stats, then opens the failing code.

  • Grouped exceptions with full stack traces
  • Logs correlated by trace id
  • First-seen times point at the breaking deploy
  • Ends with a proposed fix in your code
Claude
Claude Code v2.1.173
Fable 5 · Claude Max
~/shop-api
❯/tracepath users report 500s on checkout since this morning
Bash(tracepath exceptions list --since 24h --search checkout)
⎿ TypeError: cart is null · 412 events · first seen 09:14
Bash(tracepath logs query --trace-id 9f2c41d8)
⎿ 12 log records · cart expired 09:13, reused 09:14
Read(src/checkout/session.ts)
⎿ Read 86 lines
Root cause: repeat purchase reuses an expired cart. Fix ready for review in src/checkout/session.ts:42.
❯
The CLI

A command line designed for agents first

Built for non-interactive use: JSON output, stable errors, and nothing that hangs waiting for input. Humans on a TTY still get tables.

  • JSON when piped, tables on a TTY
  • --fields trims responses to what was asked
  • Stable error identifiers and exit codes
  • Mutations require --yes, nothing hangs
bash · tracepath-cli
1$ tracepath exceptions list --since 24h | head -1
2{"hash":"82b58892","type":"TypeError","count":412}
3$ tracepath exceptions show 82b58892 --fields type,stacktrace
4{"type":"TypeError","stacktrace":"src/checkout/session.ts:42 …"}
5$ tracepath exceptions archive 82b58892
6{"error":"confirmation_required","hint":"re-run with --yes","exit_code":2}
7# ✓ predictable for agents, readable for humans
Setup

Set up by the agent, not by the docs

/tracepath-setup picks the right path for each part of the stack: OTel for backends, TracePath SDKs for browser and mobile. Then it verifies data is actually arriving.

  • Detects frameworks and services from the repo
  • OTel for backends, SDKs for web and mobile
  • Covers background tasks, AI traces, and source maps
  • Verifies grouped endpoints in the dashboard
TracePath endpoints grouped by route pattern with percentiles
The skills

Two skills, one install.

Plain Markdown playbooks your agent loads on demand. One sets up your project, the other queries and debugs it.

01 · Setup

/tracepath-setup

Reads your repo and wires it up: OpenTelemetry for the backend, TracePath SDKs for web and mobile. Verifies data arrives clean and grouped.

OTel backendsWeb + mobile SDKsSource mapsVerification

02 · Debug

/tracepath

Installs the tracepath CLI, then uses it: exceptions, logs, endpoints, and metrics, from bug report to root cause.

CLI installTelemetry queriesDebuggingIssue triage
Compatibility

One format, every agent

Skills are plain Markdown in the open SKILL.md format. No marketplace, no lock-in. They live in the same MIT-licensed repo as TracePath itself.

Claude CodeCursorCodexOpenCodeGemini CLICopilot
Also available

Prefer live tools? TracePath speaks MCP.

Connect Claude Code, Cursor, or any MCP client straight to your instance over OAuth. The same telemetry, exposed as MCP tools, prompts, and resources.

Explore the MCP server

Give your agent production context.

Install the skills and let your agent take the next bug.

Star on GitHubStart for free
FAQ

Questions about agent skills

What exactly is an agent skill?

A SKILL.md file: Markdown instructions your agent loads when a task matches. TracePath ships two, one that sets up a project and one that queries and debugs it. Both live in the open-source repo.

Which agents are supported?

Anything that reads SKILL.md: Claude Code, Cursor, Codex, OpenCode, Gemini CLI, Copilot, and more.

npx skills add tracepathhq/tracepath installs the skills for every agent it finds on your machine.

Can an agent damage my production data?

No. The CLI is read-only apart from archiving exception groups, and that requires an explicit --yes flag.

Does this work with self-hosted TracePath?

Yes. tracepath login --url works against any instance, and profiles let one machine use several.

Do I need the CLI installed first?

No. The /tracepath skill installs and authenticates it before running its first query.

Is there an MCP server too?

Yes. TracePath also ships a Model Context Protocol server: remote over OAuth at https://your-instance/mcp, or local on stdio via tracepath mcp. Skills drive the CLI; MCP gives the agent live tools.

See the MCP server →