Skip to main content
skill CX Codex

warden-service — Claude Code skill: query Warden API

Skill for reading Warden Service data: search and export runs, findings, costs and metrics via personal token. Works across AI code editors.

Apache-2.0 189 installs ★ 426 updated 2026-10-24

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add getsentry/warden --skill warden-service -a claude-code
Open on GitHub

What warden-service does

This skill wraps the Warden Service read-only API, letting agents search, filter and summarize runs, findings, costs, outcomes, repositories, skills and memory. Requests use a personal token stored in the WARDEN_PAT environment variable, never exposed in logs or chat.

The skill follows seven steps: verify the environment and token, read API docs from references/read-api.md, pick the route and filters, URL-encode all parameters, send a GET request, handle cursor pagination for /api/v1/runs and /api/v1/findings, and return a summary with the route and filters used. The token is never cached, logged, or passed as a command literal.

You need access to Warden Service and an exported personal token. The skill is read-only; write operations (POST, PUT, PATCH, DELETE) are not supported. If the API cannot answer the question, the skill says so instead of trying substitutes through other operations.

What it helps with

  • fetch all high-severity findings in a repository over a time range
  • calculate spend across runs, skills or cost categories
  • export memory, outcomes or metrics data for reporting
  • check the status of recent runs and list errors
  • filter findings by skill, severity or storage

Example prompts for the agent

  • Show all findings with severity high from the past week
  • Find runs for the security-review skill and display their spend
  • Export memory data from warden-service as JSON

How to install warden-service

1Skills CLI (Vercel)

npx skills add getsentry/warden --skill warden-service -a claude-code

2Manually: put the files in place

~/.claude/skills/warden-service/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/warden-service/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about warden-service

How do I install warden-service in Claude Code?

Run npx skills add getsentry/warden --skill warden-service -a claude-code, then open ~/.claude/skills/warden-service/SKILL.md. For other editors, change the flag: -a codex for Codex, -a cursor for Cursor, -a gemini-cli for Gemini CLI, -a opencode for OpenCode. For GitHub Copilot, use gh skill install getsentry/warden warden-service.

What do I need to use warden-service?

Create a personal token in Warden Service (API access menu in the dashboard) and export it as an environment variable: export WARDEN_PAT=<your_token>. Do not paste the token into a command or chat. The skill reads the token from WARDEN_PAT on each request.

What operations does warden-service support?

Read-only: search runs, findings, costs, outcomes, repositories, skills and memory via GET requests to the read API. Write operations (POST, PUT, PATCH, DELETE) are not supported. Pagination uses an opaque cursor passed automatically in the next request.

Source text

SKILL.md · Apache-2.0
---
name: warden-service
description: Queries and summarizes the Warden Service read API. Use when asked to inspect, search, summarize, or export Warden runs, findings, costs, outcomes, repositories, skills, or memories through Warden Service.
spec_hash: c2fea93e547a
---

# Warden Service

Query Warden Service with a read-only personal token and return a bounded, evidence-based answer.

## Workflow

1. Identify the Warden Service origin and confirm `WARDEN_PAT` is set without displaying its value.
2. If the token is missing, direct the user to **API access** in the Warden Service dashboard. Have them create a personal token and export it as `WARDEN_PAT`; never ask them to paste it into chat or a command literal.
3. Read `references/read-api.md` to select the exact route, supported filters, pagination behavior, and response fields for the question.
4. Narrow the request to the user's repository, time range, skill, severity, or other stated scope. Use a bounded `limit` for runs and findings.
5. Send a `GET` request with the token read from the environment. URL-encode every query value.
6. Follow `nextCursor` only while more results are needed. Preserve all original filters and stop when the requested scope is satisfied or the response omits `nextCursor`.
7. Summarize the relevant JSON fields. State the route and filters used, and distinguish an empty result from a failed request.

## Request Pattern

Use `--get` with `--data-urlencode` instead of assembling a query string manually:

```bash
curl --fail-with-body --silent --show-error --get \
  -H 'Accept: application/json' \
  -H "Authorization: Bearer ${WARDEN_PAT}" \
  --data-urlencode 'severity=high' \
  --data-urlencode 'skill=security-review' \
  --data-urlencode 'limit=30' \
  "${WARDEN_SERVICE_URL%/}/api/v1/findings"
```

Keep the token in `WARDEN_PAT`. Do not enable verbose or trace output that could expose the authorization header.

## Pagination

Only `/api/v1/runs` and `/api/v1/findings` use cursor pagination. Treat `nextCursor` as opaque and pass it back unchanged through URL encoding:

```bash
curl --fail-with-body --silent --show-error --get \
  -H 'Accept: application/json' \
  -H "Authorization: Bearer ${WARDEN_PAT}" \
  --data-urlencode "cursor=${NEXT_CURSOR}" \
  --data-urlencode 'severity=high' \
  --data-urlencode 'limit=100' \
  "${WARDEN_SERVICE_URL%/}/api/v1/findings"
```

## Errors

Use the HTTP status and the JSON `error.code` and `error.message` together:

| Status | Meaning | Action |
| --- | --- | --- |
| 400 | Invalid query filters | Correct the parameter names, values, or RFC 3339 timestamps. |
| 401 | Missing, invalid, or expired authentication | Verify the origin and replace the personal token through API access. |
| 403 | Insufficient role or disallowed personal-token operation | Keep the request read-only and within the token's repository scope. |
| 404 | Unknown or unauthorized route/resource | Verify the documented route or ID without assuming the resource exists. |
| 429 | Rate limited | Wait and retry later; do not create a tight retry loop. |

## Boundaries

- Use personal tokens only for `GET` or `HEAD`. Never attempt `POST`, `PUT`, `PATCH`, or `DELETE` with them.
- Never reveal, log, persist, embed as a command literal, or ask the user to paste a token.
- Respect the token's tenant, role, and repository restrictions. Never attempt to bypass them.
- Use only routes, filters, and response fields documented in `references/read-api.md`.
- Fetch and display only the data required for the user's stated scope. Use the export route only when the user explicitly requests an export.
- If the read API cannot answer the question, say so. Do not substitute an ingest, memory-recall, token-management, retention, or deletion request.

© getsentry · Apache-2.0 · getsentry/warden

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set3

Project sponsors

Companies that keep this analytics open