Dependency Outage Triage

August 18, 2026 ยท View on GitHub

An installable agent skill for checking vendor-published incident evidence before you roll back a deployment, rotate credentials, or change working code.

It uses OutageDeck as a read-only evidence source for current status, affected services, incident timelines, and source links across cloud and SaaS vendors. Public checks need no account or API key.

Install

For agents that support the Agent Skills format:

npx skills add https://github.com/outagedeck/triage-dependency-outages --skill triage-dependency-outages

For OpenClaw:

openclaw skills install git:outagedeck/triage-dependency-outages@v0.1.0

The OpenClaw command pins the tested v0.1.0 release. The generic installer reads the skill directly from this public repository.

What the skill does

When an application, CI job, deployment, API, authentication flow, or network call starts failing, the skill:

  1. identifies the external providers implicated by the evidence;
  2. checks service-scoped status and active incidents through OutageDeck MCP or its public REST API;
  3. compares vendor timestamps and affected services with the observed failure;
  4. returns a compact evidence table and one conservative verdict;
  5. recommends reversible next diagnostics without treating provider status as proof of causation.

It explicitly distinguishes a vendor-reported incident from regional, account-specific, newly emerging, and local failures.

Example prompts

Is GitHub Actions down, or did our deployment workflow regress?

Check OpenAI, Anthropic, AWS, and Cloudflare before I change this retry logic.

Build an incident timeline for the providers implicated by these 503 errors.

Runtime access

The skill prefers an already configured OutageDeck MCP server. Without MCP, it falls back to the keyless public endpoints documented in SKILL.md, including:

GET https://outagedeck.com/api/v1/providers?q=<name>
GET https://outagedeck.com/api/v1/providers/<slug>
GET https://outagedeck.com/api/v1/incidents?provider=<slug>&state=active&limit=10

The public API is suitable for lightweight checks. Teams that want ongoing email, Slack, Teams, Discord, or webhook notification can configure provider alerts.

Safety

  • No OutageDeck key is required for public checks.
  • The skill never asks users to paste credentials into chat.
  • It does not change code, roll back deployments, or disable security controls based only on external status.
  • Account-changing OutageDeck tools require an explicit user request; provider removal also requires confirmation.

License

MIT