From pm-skills
Guides a contributor from a workflow idea to a complete Workflow Implementation Packet with overlap analysis, Why Gate, and skill sequencing. Use when creating a new multi-skill workflow or promoting an ad-hoc chain into a durable one.
How this skill is triggered — by the user, by Claude, or both
Slash command
/pm-skills:utility-pm-workflow-builderThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
<!-- PM-Skills | https://github.com/product-on-purpose/pm-skills | Apache 2.0 -->
This skill creates new multi-skill workflows for the pm-skills library. It
produces a Workflow Implementation Packet - a draft _workflows/<name>.md
file, a draft commands/workflow-<name>.md command, and the cross-cutting
update checklist - in a staging area for review before promotion to
canonical locations. The builder authors; it never executes a chain and it
never promotes its own output.
/chain run (orchestrator Mode B) proved reusable and its completion
output suggested promoting the chain to a durable workflow; hand this
skill the exact chain expression from that suggestion/chain
command or utility-pm-workflow-orchestrator directlyutility-pm-skill-builder,
utility-pm-skill-validate, or utility-pm-skill-iterate_workflows/<name>.md
directly; the overlap analysis below will point you there when coverage
is highAccept the idea in any of three entry forms, all converging on the same downstream flow:
measure-experiment-results -> discover-stakeholder-summary), usually
pasted from the orchestrator's completion suggestion, plus the context
the run used.If the idea is vague, ask ONE clarifying question (the recurring trigger and the final artifact), then proceed. Do not interrogate.
Scan the _workflows/ directory at run time for the current inventory
(list every *.md except README.md). Never rely on a remembered count
or a hardcoded list. Compare the idea against ALL existing workflows:
Kill gate (>70% coverage overlap): recommend, in order, (a)
customizing the existing workflow, (b) adding a step to it, or (c) just
using /chain for the occasional run. Do not proceed.
Why Gate (any meaningful overlap): ask for 2-3 specific scenarios where the existing workflows fail to produce what is needed. Do not pass the gate without that evidence or an explicit user override (record the override in the packet's Decision section).
Select and order the steps with the user. Apply the chain-expression
contract's validation rules at authoring time
(skills/utility-pm-workflow-orchestrator/references/PARSE-CONTRACT.md,
Mode B Chain Expression Contract):
skills/<name>/SKILL.md
by EXACT name. Never approximate or auto-correct; on a miss, refuse the
step and OFFER the closest real names (an offer, never a substitution).utility-pm-skill-*,
utility-pm-release-conductor, utility-pm-changelog-curator,
utility-update-pm-skills, foundation-prioritized-action-plan),
dispatch skills that fan out to sub-agents (utility-pm-critic,
utility-pm-workflow-orchestrator, and peers), and existing workflows
(a workflow never nests a workflow). Self-reference is impossible by
construction: this builder is itself Tier-3.Write the complete Workflow Implementation Packet to
_staging/workflows/<name>/ (gitignored; the same review model as the
skill builder). Never write to canonical locations.
_staging/workflows/<name>/
├── workflow.md <- draft _workflows/<name>.md
├── command.md <- draft commands/workflow-<name>.md
└── PACKET.md <- decision, overlap analysis, checklist, promotion steps
Use references/TEMPLATE.md as the packet format. The workflow draft must
carry the full section inventory the current _workflows/*.md files share
(see the TEMPLATE's embedded skeleton); the command draft mirrors the
existing commands/workflow-*.md shape, including one literal
skills/<name>/SKILL.md path per step and the $ARGUMENTS footer. Derive
a linear mermaid context-flow diagram from the final sequence (step names
as nodes, execution order as edges).
Present the packet and stop. The builder itself never promotes. On
approval, the user (or a follow-up session) moves the drafts to their
canonical paths and works the packet's Cross-Cutting Checklist, which
names every count and documentation surface that adding a workflow trips,
including the validator-blind .github/workflows/release.yml release-note
template. The builder never edits those cross-cutting files itself.
The packet follows references/TEMPLATE.md:
_workflows/<name>.md contentcommands/workflow-<name>.md contentSee references/EXAMPLE.md for one complete worked packet.
Before presenting the packet, verify:
_workflows/ directory (no remembered list, no hardcoded count)skills/<name>/SKILL.md path and ends with the $ARGUMENTS footer_staging/workflows/<name>/; no canonical path touched.github/workflows/release.yml release-note rowreferences/TEMPLATE.mdreferences/EXAMPLE.mdskills/utility-pm-workflow-orchestrator/references/PARSE-CONTRACT.md_workflows/sprint-planning.md, _workflows/customer-discovery.mdcommands/workflow-sprint-planning.mdskills/utility-pm-skill-builder/SKILL.mdnpx claudepluginhub product-on-purpose/pm-skills --plugin pm-skillsComposes valid looplia v0.7.0 workflow YAML/Markdown files from skill recommendations and user preferences. Final step for /build commands, workflow creation, or automation pipelines.
Guides designing workflow-based Claude Code skills with numbered phases, decision trees, subagent delegation, and progressive disclosure. For sequential pipelines, routing patterns, safety gates, task tracking, phased execution, or refactoring skills.
Guides PM skill creation from idea to complete Skill Implementation Packet with gap analysis, Why Gate validation, and draft file generation.