Skip to content

Agent Skill

Warden ships with agentskills.io skills that let AI coding agents run Warden and query Warden Service on your behalf.

Terminal window
npx @sentry/dotagents add getsentry/warden

dotagents is the recommended installer. If your environment already uses Vercel Skills, use the Vercel Skills tab instead.

The /warden and /warden-service skills are available immediately.

The /warden skill teaches agents how to:

  • Run Warden against current branch changes before review.
  • Interpret findings and fix the code.
  • Configure skills in warden.toml.
  • Create custom skills.

Agents load it automatically when they detect trigger phrases like “run warden”, “check my changes”, or “review before PR”.

The /warden-service skill teaches agents how to:

  • Authenticate with a read-only personal token without exposing it.
  • Query runs, findings, costs, outcomes, repositories, skills, and memories.
  • Follow cursor pagination and summarize only the requested data.
  • Diagnose structured API errors without crossing the token’s read-only boundary.

Agents load it for Warden Service API questions such as “show recent Warden findings” or “group Warden cost by repository”.

Ask an agent that supports the spec to use Warden before opening a pull request:

> Use /warden before opening a PR

The agent runs warden, reads the output, and fixes what it finds.

To query the backing service, create a personal token from API access in the dashboard, export it as WARDEN_PAT, and ask the agent a scoped question:

> Use /warden-service to show high-severity findings from the last week

The skill follows progressive disclosure. SKILL.md contains the core workflow, and reference files load when the agent needs deeper detail.

FileLoaded when
SKILL.mdAlways, for the core workflow.
references/cli-reference.mdFull option details and per-command flags.
references/configuration.mdEditing warden.toml, triggers, and troubleshooting.
references/config-schema.mdExact field names, types, and defaults.
references/creating-skills.mdWriting custom and remote skills.

Agents discover the skill by scanning known directories for SKILL.md files. The name field in frontmatter must match its directory name:

.claude/skills/warden/
.claude/skills/warden/
├── SKILL.md
└── references/
├── cli-reference.md
├── configuration.md
├── config-schema.md
└── creating-skills.md

The primary workflow:

  1. You make code changes, or an agent makes them for you.
  2. Before committing, the agent runs warden.
  3. The agent reads findings and fixes issues.
  4. The agent commits the clean code.

One pass. The skill instructs agents to run Warden once, fix what it finds, and move on.