Ub Workflow

UB Workflow

Overview

Use this skill as the portable workflow layer for work that is too large, risky, cross-cutting, or stateful to run from chat history alone.

The default operations root is ./.ub-workflows/ unless the host deliberately chooses another root.

The normalized model is product-agile hybrid:

Product Vision -> Product Options -> Outcome Waves -> Initiatives -> Discoveries -> Sprints

The model combines:

1. dual-track agile: discovery validates options before delivery pulls work;
2. goal-oriented roadmapping: product vision stays broad while waves and

initiatives carry current outcomes;

1. Scrum-like sprint execution: one sprint goal, inspect/adapt, and explicit

start approval when reviewed mode is active;

1. Kanban-style flow controls: explicit policies, WIP limits, pre-commitment

options, feedback loops, and stale-candidate revalidation;

1. Shape Up influence: lightweight bet framing, appetite, circuit breakers,

and no backlog sludge;

1. appetite-boxed forecast calibration: candidate counts are forecasts, and

expansion requires operator choice after scoped options and tradeoffs.

Canonical Layout

PROJECT_ROOT/
  AGENTS.md
  SOURCE_ATLAS.md
  .ub-workflows/
    AGENTS.md
    vision.md
    options.md
    status.md
    WORKFLOW_ATLAS.md
    SOURCE_PACK_ATLAS.md
    source-packs/
      .gitkeep
      YYYY-MM-DD-slug/
    waves/
      .gitkeep
      wNN-wave-slug/
        wave.md
        discoveries/
          .gitkeep
          wNN-dNN-slug.md
        source-packs/
          .gitkeep
          YYYY-MM-DD-slug/
        initiatives/
          .gitkeep
          iNN-initiative-slug/
            initiative.md
            options.md
            roadmap.md
            index.md
            discoveries/
              .gitkeep
              wNN-iNN-dNN-slug.md
            sprints/
              .gitkeep
              wNN-iNN-sNN-sprint-slug/
                sprint.md
                decision-log.md
                closeout.md
                evidence/
                  .gitignore
                  index.md

Naming rules:

1. wave IDs are project-sequenced: w01, w02, etc.;
2. initiative IDs reset per wave: i01, i02, etc.;
3. wave-level discoveries use wNN-dNN-slug.md and reset per wave;
4. initiative discoveries use wNN-iNN-dNN-slug.md and reset per initiative;
5. sprints use wNN-iNN-sNN-sprint-slug/ and reset per initiative;
6. source packs use YYYY-MM-DD-slug/ using creation, original research,

earliest known history, or migration date with a note.

Lanes

Choose the lane that lets the accepted objective be completed, validated, and recovered without unrelated expansion:

1. direct bounded work: a small single-session change with no durable planning

surface needed;

1. wave/initiative workflow: multi-session, staged, risky, product-shaping,

or cross-cutting work that needs options boards, discovery, roadmaps, sprints, closeouts, and resumable evidence.

Promote lanes when the current lane no longer provides enough durable surface to finish and prove the accepted objective. Record the promotion reason in the artifact that now owns the work.

Embedded Contract

1. Keep reusable workflow rules in this skill, not in one repository overlay.
2. Keep repository overlays thin: project facts, current pointers, local

validation commands, domain boundaries, and repository-specific constraints.

1. vision.md owns the adaptable product north star.
2. Root options.md owns curated product-level, future-wave, and

unknown-owner options before commitment. It is not a backlog ledger, completion history, or execution authorization surface.

1. status.md owns current product posture, wave sequencing, active pointers,

blockers, WIP state, candidate tracks, retained-context routes, and next allowed action.

1. SOURCE_ATLAS.md owns project-root source routing. Bootstrap seeds it with

a one-time scan; later updates are event-based when source boundaries move.

1. Root AGENTS.md owns the local agent overlay. Bootstrap creates or patches

only a small managed workflow-routing section.

1. A workflow-root guide, split product/live-state startup files, and a root

specs lane are not canonical workflow surfaces.

1. Outcome waves are adaptive product slices; they may split, merge, grow, or

shrink through reviewed discovery.

1. Initiatives are thematic wave-local bets; they may pause and resume.
2. Initiative-local options.md owns possible insertions before that

initiative closes. Product-level or later-wave options move to root options.md.

1. Discoveries are normal upstream steering work, not exceptional ceremony.
2. Discoveries and sprint previews record user or operator evidence status as

used, not triggered, or deferred so technical work still states whether real-user or operator feedback shaped the decision.

1. Sprints are delivery slices pulled from accepted discovery or a reviewed

sprint preview.

1. Default WIP is constrained dual-track: one active delivery sprint and one

active discovery per active initiative; wave discovery is active only for activation, transition, or reroute.

1. Future work may live in root or initiative-local options boards before

commitment. Options are ordered by document order within horizon lanes, but are not delivery commitments, status ledgers, or execution queues.

1. Remove an option after it is promoted, rejected, merged, or completed and

the receiving artifact owns the durable trace. Do not maintain a Done lane in options boards.

1. Before activating a new wave or initiative, review root options and

unresolved local options from the closing initiative.

1. Before closing an initiative, promote, move, reject, or remove every local

option. Do not archive completed cards in the options board itself.

1. Run options validation at wave or initiative transition, option promotion,

initiative closeout, terminal audit, and when a sprint preview is pulled from an option.

1. Reusable workflow-system changes that affect artifact ownership,

lifecycle gates, scaffold output, transition policy, or recovery context require a compact workflow-improvement discovery or equivalent accepted decision record.

1. Revalidate candidates on touch: any candidate not prepared from current

evidence must pass fresh discovery or preview before execution. Use discovery when route-changing uncertainty exists; use reviewed preview only when a registered candidate is being freshly revalidated for start.

1. Discovery Triage is a fail-closed Routing Preflight, not a preview

appendix. Before choosing a reviewed preview, record trigger categories and one outcome: preview_ok, discovery_required, operator_decision_needed, or not_triggered. Trigger categories include autonomy, authority, runtime-boundary, policy, evidence, interoperability, agent-behavior changes, agent loops, harnesses, loop continuation, goal judgment, budgets, and checkpoints. If triggered evidence may change route, contract, scope, validation, runtime boundary, or implementation path, record discovery_required; preview_ok is valid only when the route is already decided and trigger evidence cannot change those decisions. Use not_triggered only with a concrete reason.

1. Discovery-driven sequence changes must be promoted into the owning

roadmap.md before acceptance; closeout-driven next-route changes must be promoted before sprint closeout.

1. Sprint previews identify repo-owned operational surfaces when triggered:

owner, lifecycle, visibility, safety policy, validation path, and inventory/topology/registry/route-map impact. If uncertainty about those facts may change scope, architecture, risk, or acceptance criteria, promote discovery before execution.

1. Each wave and initiative records qualitative Outcome Signals for product

or user value, delivery flow, quality or stability, and context or evidence cost. These are routing signals, not a metrics ledger.

1. Each initiative and sprint records lightweight bet framing: appetite,

success evidence, circuit breaker, non-goals, and deferral path.

1. Waves and initiatives use Appetite-Boxed forecasting by default: forecast

counts are not commitments, sequence expansion is not automatic adaptation, and adding work requires the agent to present options, tradeoffs, and a recommended path for explicit operator decision.

1. Wave and initiative charters record Forecast And Appetite; roadmaps record

Forecast Control; discoveries and reviewed previews record Forecast Impact when they change sequence.

1. Sprint previews record Product Increment Contribution as direct,

enabling, or audit; enabling sprints must name the visible increment they unblock and why a direct slice is not viable.

1. Two consecutive enabling or prerequisite sprints trigger a route review:

propose shipping a vertical proof, cutting/deferring scope, rerouting, or explicitly buying more enabling work.

1. Major capability candidates decompose prerequisite risks and the first

usable product increment before acceptance.

1. In reviewed mode, a request to move on opens a preview only; execution

starts only after a later explicit approval.

1. New sprint packs include decision-log.md, closeout.md, and

evidence/index.md.

1. evidence/index.md is the T4 claim-to-proof router. It records validated

claims, evidence files, required objective gates, optional or not-triggered gates, redaction posture, promotion targets, and read policy without becoming a narrative closeout.

1. Generated runtime state under sprint evidence (any

project-specific scratch directory) is local scratch by default. Commit only reviewed no-secret evidence files or an explicitly approved export.

1. Sprint closeout includes a required outcome and learning review with four

prompts: what did this sprint achieve, how did it make the project better, what could have been done better, and whether the learning changes the next route, stays local, or suggests a workflow improvement.

1. Sprint closeout records Forecast Delta: planned versus actual, hidden

prerequisite discovered, remaining forecast impact, and whether roadmap, index, or status were updated.

1. Material closeout claims must be backed by evidence/index.md. Required

objective proof cannot be deferred while closing the original objective as passed; missing or failing required proof keeps the sprint active or blocked until the operator explicitly changes scope or the proof passes.

1. Sprint closeout includes a focused mini-retro: process friction, evidence

cost, context cost, decision latency, workflow adjustment if needed, and a retro evidence check covering objective signal, controllability, and repeated friction.

1. Repeated or structural workflow friction is promoted into this skill;

one-off repo friction stays local.

1. Project evolution mode defaults to forward-only unless a host records a

reviewed compatibility-preserving decision.

1. When the active evolution mode requires forward migration, broad boundary

changes must include an impact inventory and must update, remove, or explicitly defer each affected owned surface before closeout.

1. Source packs are retained context, not execution authorization, live state,

or backlog.

1. Trace tokens are owner-only lookup anchors. Use triggered T3 initiative

index.md trace routes before broad workflow search, and do not add trace tokens to discoveries, sprint packs, closeouts, decision logs, or evidence indexes.

Interaction Modes

1. reviewed is the default: preview before execution, explicit later start

approval, post-execution report, and pause after closeout.

1. flow: short pre-execution note, post-execution report, manual advancement.
2. auto: internal pre-execution analysis and automatic advancement unless a

blocker, conflict, or path-shaping decision requires interruption.

1. continuous / yolo: no routine pauses, but all gates, artifacts, and

interruption rules still apply.

Mode changes visibility and pause behavior. It does not weaken readiness, evidence, or writeback requirements.

Objective-Complete Rule

Choose an objective-complete action: implement and validate all work required for the accepted sprint objective and exit criteria. Avoid unrelated expansion, but do not cut, defer, or split required proof just to preserve scope or budget.

When a sprint, preview, or closeout uses words like smallest, narrow, or only as needed, interpret them as the smallest objective-complete vertical slice, not the smallest patch that makes the first focused test pass. Every repo-owned affected surface named by the sprint scope, every required evidence gate, and every exit criterion remains in scope unless the operator explicitly changes the objective.

Load References By Trigger

• [phase:lifecycle-detail] Read references/workflow-contract.md for

detailed operating rules, WIP policy, reviewed-mode previews, and recovery.

• [phase:artifact-create|artifact-validate] Read

references/artifact-contracts.md when creating or validating workflow artifacts.

• [edge:context-management|frontmatter|context-budget] Read

references/context-management.md when interpreting context_tier, summary_budget_lines, phase read budgets, atlas routes, context receipts, retained-context reads, or evidence/writeback receipt shapes.

• [edge:trace-token|trace-lookup|workflow-search] Read

references/trace-tokens.md when searching workflow history, interpreting trace IDs or tags, adding trace metadata, or deciding where trace anchors belong.

• [phase:gate-eval|closeout|readiness] Read

references/validation-and-completion.md when evaluating readiness, closeout, archive, or completion.

• [edge:helper-use] Read references/scaffold-helper.md before using or

explaining the deterministic helper.

• [edge:strict-placeholder-validation|options-validation] Read

references/placeholder-contract.md when strict placeholder validation is relevant. Read references/validation-and-completion.md when options-board validation, transition checks, or closeout checks are relevant.

• [edge:governance-escalation] Read references/governance-bridge.md only

when explicit governance mapping is active.

• [edge:authoring-conventions] Read

../ub-authoring/references/authoring-conventions.md when shared routing, naming, or choice-question authoring is being changed.

Bundled Assets

Use assets/operations-root/ and assets/initiative-template/ as canonical templates. Use scripts/scaffold_workflow.py for deterministic bootstrap, wave, initiative, discovery, source-pack, sprint, and archive operations. Use scripts/check_workflow_options.py for options-board transition, closeout, terminal-audit, and stale-card validation.

Output Requirements

For non-trivial workflow work, report:

1. lane and scale decision;
2. active artifact owner;
3. WIP and gate state;
4. chosen path and rejected alternative when path choice matters;
5. validation and evidence expectation;
6. next allowed action.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: robert-hoffmann
• Source: robert-hoffmann/uncle-bob
• License: MIT
• Homepage: https://robert-hoffmann.github.io/uncle-bob/

Install and usage instructions live in the source repository linked above.