integrations-lifecycle

The pipeline that turns source metadata into the integration pages, the umbrella pages, and the in-app catalog. This skill owns the mechanics and the delivery rules; what a collector metadata.yaml field should say is .agents/skills/collectors-metadata-yaml/, and config_schema.json is .agents/skills/collectors-go-design/config-schema.md.

Apply AGENTS.md#skill-selection. For review, select the affected pipeline or artifact contract below and inspect source plus existing validation evidence. Loading recipes does not require regeneration, dependency installation, commits or publication. Authorized validation uses current inputs and preserves existing outputs through isolation.

Shared delivery facts:

• metadata.yaml is the input. Hand-authored for most integrations; generated for ibm.d modules (from contexts.yaml, config.go, module.yaml) and for the NPM catalog (from SNMP profiles by integrations/gen_npm_catalog.py). Edit the authoritative source, never a generated file.
• Every <dir>/integrations/<slug>.md, every README that is a symlink or whose <!--startmeta block carries a message: "DO NOT EDIT..." line (a <!--startmeta block alone is not the marker; hand-authored Learn pages have one too), and the umbrella pages src/collectors/COLLECTORS.md, SECRETS.md, SERVICE-DISCOVERY.md (no banner) are generated. Do not edit them.
• Source PRs carry sources; generated documentation arrives through the post-merge integrations-regen PR. Validate with gen_integrations.py and the tests during authorized changes; preview pages in isolation using how-tos/preview-collector-page.md (selected collectors or its non-collector route). Preserve existing generated-file edits and keep them out of the source commit. The one place this boundary is written out, with its ibm.d runtime-output exception, is consistency.md.
• The collector taxonomy (taxonomy.yaml, gen_taxonomy.py) is a dormant early implementation kept for later work; nothing runs it. Do not author, extend, or seed taxonomy content (consistency.md, "The dormant collector taxonomy").

You edit metadata or a collector and need to validate and deliver

You change the generators, schemas, templates, or CI

Citations in this skill name files and symbols, never line numbers; open the file and search for the symbol.

Live how-to rule

Capture timing and authorization follow AGENTS.md#knowledge-capture. Pipeline recipes live in how-tos/ and are listed in ./how-tos/INDEX.md; prefer updating an existing guide when it already covers most of the question.

Path discipline

.agents/sensitive-data-discipline.md applies: repository files repo-relative, sibling Netdata-org repositories as ${NETDATA_REPOS_DIR}/<repo>/..., no workstation paths.

Related skills

• collectors-authoring: read first when authoring a brand-new collector; this skill covers the integration side.
• collectors-metadata-yaml: the words on the page, field by field.
• collectors-go-design/config-schema.md: the DynCfg form.
• docs-learn-site-structure: how the generated pages reach learn.netdata.cloud (docs/.map/map.yaml, ingest, redirects).
• collectors-prometheus-profiles: the profile side of the Prometheus catalog.