Pipelines
A pipeline is a phased, gated recipe for producing one kind of artifact. Each pipeline declares a domain, an ordered list of phases (each phase names a role), and a gate. Pipelines live as YAML in pipelines/ and cover both software and knowledge work.
Routing
Routing to a pipeline is deterministic. muster route "<outcome>" matches the outcome against each pipeline's match keywords on word boundaries; if nothing matches, it falls back to the domain default.
npx -y @adnova-group/muster route "draft a PRD for a referral program"
npx -y @adnova-group/muster pipeline prdThe gate: a floor principle
Gating uses a floor principle: the weakest dimension must clear the gate's floor, and the total must clear a pass threshold. A strong average cannot rescue one weak dimension. Scoring is deterministic and fails loud on non-finite inputs. The model only estimates the per-dimension scores; the code decides pass or fail.
Humanize
Human-facing pipelines end with a humanize phase. The muster-humanizer built-in strips em-dashes, banned AI-tell words, and robotic cadence. Machine-facing documents (the design-doc and QA test-plan pipelines) are exempt, to preserve technical precision.
The set
There are 19 pipelines, spanning software and knowledge work. Ask for any of them by id, or list them all with muster pipeline --list:
# print one pipeline's phases, roles, and gate
npx -y @adnova-group/muster pipeline runbook
npx -y @adnova-group/muster pipeline release-notespipeline also accepts a domain (pipeline ops, pipeline pm), which resolves that domain's default pipeline. The set:
- Product: PRD (it can also render an existing intent and spec for stakeholders), epic, roadmap (outcome-framed, with intent stubs for its Now items)
- Business: business-case, executive-summary, OKRs, competitive-battlecard
- Launch and comms: launch-plan, release-briefing (
release-notes.yaml; three audience-split artifacts -- technical changelog, user briefing, announcement post -- from one draft phase), case-study - Content: blog-post, social-post, newsletter, lead-magnet, video-content, book (fiction and non-fiction)
- Engineering specs: design doc (
design; the old idai-implementation-specstill resolves), human QA test plan (ai-test-plan) - Operations: runbook
Content pipelines that resolve an audience or a voice read named profiles from docs/profiles/: AUDIENCES.md (depth, jargon, altitude), VOICE.md (register, rhythm, anti-patterns), and BRAND.md (a visual anchor for the image/video roles) -- created or extended on first use, not hand-authored up front. Blog-post, lead-magnet, and social-post also carry an optional_phases: [publish] phase: skippable, terminal, and scoped to assembling the ready-to-publish packet and evidence -- never the deploy/post/send act itself, which stays behind the action-class fence.
Roadmap prioritization
One pipeline worth calling out. Goals go in, a prioritized Now/Next/Later roadmap of outcomes comes out. The model estimates the factors with evidence-backed rationale; muster prioritize does the arithmetic and fails loud on bad input. RICE is the default, with ice, wsjf, and weighted also available. See prioritization models.
The muster-roadmap built-in it reads can render the ranking to a doc, optionally file the top-N initiatives as GitHub issues, and optionally push them onto a GitHub Project board with status columns by tier. Each external step degrades gracefully, skipping with a note when gh access is unavailable.
Running a pipeline by name
/muster:go and /muster:plan route an outcome to a pipeline by its keywords. /muster:pm run <pipeline> [outcome] runs any pipeline by id, through the same muster pipeline-args and muster:pipeline workflow; /muster:pm run alone lists them. The user-story pipeline is retired: /muster:pm tickets <SPEC.md> turns a spec into backlog items that cite its acceptance-criterion IDs. muster doctor checks that every pipeline stays reachable this way and that its sources are in vendor/manifest.yaml.
Next: the full Architecture.