Capabilities overview
Core delivery is on after pipeline-kit init. Everything in the second table is opt-in.
Pipeline Kit ships two kits. Kit mode is the markdown pack. Orchestrator mode is the Python engine. Same first-party workflow names. Adding a process is an extension, not a fork.
The two kits
| Surface | What it is |
|---|---|
| Two kits — pack and orchestrator | Pick one per project at init. Do not mix. |
| Extensions | Add a process in kit mode (JSON + skill) or orchestrator mode (pipeline_extensions) |
| Repository layout | Where each kit lives in this repo |
Always available after init
| Capability | What it is |
|---|---|
| Workflows | Named procedures with chains and allowlists |
| Planning gates | Human sign-off on requirements, architecture, BA |
| Loader | Per-step allowed_reads |
| Wiki | Trigger-routed agent memory |
| Feature flags | Named on/off keys that mirror config.json |
Opt-in
| Capability | How you turn it on |
|---|---|
| Knowledge base | pipeline-kit knowledge init |
| Memory bank and MCP | uv tool install -e ".[memory]", then pipeline-kit memory link |
| Plugins | External Graphify/Archify (plugins install) and bundled observability (obs install) |
| Graphify | Official CLI + skill registration |
| Archify | Pinned Agent Skill v2.16.0 |
| Agent-run observability | Bundled add-on: pipeline-kit obs install (not plugins install) |
| App telemetry | features enable telemetry or RUN_TELEMETRY |
| Enterprise portal | Separate product and repository (enterprise plan): sign-in, roles, analysis, license management |
:::warning Two different “observability” words Agent-run observability traces what the coding agent did (tools, steps, scores). App telemetry is a pipeline specialist that extracts product analytics events from a spec. Do not mix them. :::
Workflow names
ask, feature-development, jira-story, jira-epic, jira-bug, test-knowledge-bootstrap.
Deep dives: Ask, Feature development, Jira, Knowledge bootstrap.