global-context-builderlisted
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# Global Context Builder
Without this, scenario generation produces plausible rubbish — well-formed tests
against endpoints that were never built. With it, tests bind to real contracts.
It is also the fastest onboarding available: thirty minutes after it runs, a newcomer
has a dossier of services, roles, endpoints and data.
Expect the first build to yield roughly 70–80% of what is eventually needed. The
remainder accumulates during real work — `scenario-data-fix` finds gaps in
`test_data.yaml` on the first scenario run and offers to fill them. Do not chase
completeness on day one.
## Disabled integrations
Follow `qa-conventions` → *Disabled integrations* for the wiki, the tracker or the git host
(`confluence`, `jira`, `github`): read `INTEGRATIONS`, skip a system that is not listed, report it
as `➖ <system> disabled for this project`, never route around a `qa-guard` block.
**What stays available:** repositories listed in `GC_SOURCE_REPOS` and pages under `GC_SOURCE_DOCS` are declared sources and may be cloned or read even when their host's integration is off; the guard allows read-only access to exactly those addresses and nothing else on the host. Confluence and Jira context needs the integration; without it, record the gap in the context and continue in degraded mode.
## Modes
| Mode | What it does | When |
|---|---|---|
| `build` | Full assembly from scratch | New project |
| `update` | Adds new findings, shows a diff, asks per change | New service, new endpoints,