Authors the Business Central spec folder (index, brief, plan, acceptance-criteria, change-log) under its module/functional area so a developer can implement without guessing. Use when: create a spec, write the spec folder, draft brief and plan, prepare a feature for development, spec out an AL change, technical specification for BC.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Authors the Business Central spec folder (index, brief, plan, acceptance-criteria, change-log) under its module/functional area so a developer can implement without guessing. Use when: create a spec, write the spec folder, draft brief and plan, prepare a feature for development, spec out an AL change, technical specification for BC.
BC · Author a Spec
Create a complete, developer-ready specification for a BC feature. The spec is the contract
between PLAN and BUILD. This is the core procedure of the bc-spec agent.
Output: the spec folder
Specs are organised by module / functional area (mirroring how the code and the business
think about the app), not by ticket hierarchy. One folder per work item lives under its area:
specs/
└── <Area>[/<Sub-area>…]/ # e.g. Finance, Sales/Invoicing — free-form, any depth
├── index.md # area overview: story list + status + links + high-level diagrams (optional)
├── roadmap.md # phased / epic narrative that may span stories & areas (optional)
└── <type>-<ID>-<slug>/ # one folder per work item
├── index.md # story overview: status, nav to the docs, optional diagram
├── brief.md # business request, value, scope (non-technical)
├── plan.md # technical approach + AL objects + task breakdown (for the developer)
├── acceptance-criteria.md # Given/When/Then, testable
└── change-log.md # version history of the spec itself
Two story shapes. Small, single-deliverable stories stay flat (the five docs above). A large
or multi-part story is broken into tasks: the story keeps index.md, brief.md, roadmap.md
and change-log.md, and the technical detail moves down into numbered task sub-folders:
└── <type>-<ID>-<slug>/ # e.g. us-2117-country-division-bank-data
├── index.md # story overview + task table (status / approved / link)
├── brief.md # business request, value, scope (whole story)
├── roadmap.md # task sequencing: order, dependencies, approved-now vs future
├── change-log.md # spec history (whole story; per-task logs optional)
├── 01-<task-slug>/ # one folder per task, ordinal-prefixed to fix build order
│ ├── plan.md
│ ├── acceptance-criteria.md
│ └── change-log.md # optional — omit and use the story change-log
└── 02-<task-slug>/
├── plan.md
└── acceptance-criteria.md
Story folder name = <type>-<ID>-<slug>:
<type> — work-item type token: us (user story), bug, feat (feature), epic, task, spike.
<ID> — the work-item ID, bare number (e.g. 2117).
<slug> — 3–5 words, kebab-case, from the ticket title.
⚠️ Never put the app/ticket prefix in a spec path. The folder is us-2117-…, notNOB-2117-… / ABC-2117-…. The prefix belongs only to git branches (feature/NOB-2117-…) and
functional-doc folders — never to specs/. If you find yourself typing the prefix into a spec
folder name, stop.
Area folder is optional overview scaffolding. Create index.md (and roadmap.md when useful)
the first time an area gains a story, so there is a navigable overview. brief.md/change-log.md
at the area level are optional — the area index.md usually covers the "what/why", and per-story
change-logs plus the roadmap cover history.
Epics / features that span areas are tracked in markdown — in an area (or top-level)
roadmap.md/index.md that cross-links the child story folders wherever they live — not by
nesting story folders inside an epic folder. This keeps each story next to the module it touches.
Breaking a story into tasks. Split a story into task sub-folders when it has several distinct
deliverables, is large enough to build/test in independent slices, or when some parts are
approved now and others are future/parked.
⚠️ Do this IN PLACE on the existing story folder. Breaking down is an edit, not a rebuild.
Keep the existing specs/<Area>/<type>-<ID>-<slug>/ folder exactly as it is — same area, same
bare-ID name, same brief.md. Do not rename it, move it, add a prefix, or re-scaffold it from
the template. You are only adding subfolders and a roadmap.md and relocating the plan /
acceptance content down into the tasks. If a spec folder for this ID already exists, always edit it
in place — never create a second one.
When the user asks to break a story down, follow the full step-by-step (locate the existing folder,
keep brief/index/change-log, add roadmap.md, relocate the plan/acceptance content into
NN-<task-slug>/ folders, wire dependencies, fix nav bars) in
references/task-breakdown.md — it includes a worked us-2117
before/after. In short: keep the existing folder, add subfolders + roadmap, relocate the detail down.
Don't over-split: a change that touches a handful of objects is one flat story, not three tasks.
Tasks are a spec-side breakdown; they may or may not mirror ADO child Tasks under the User Story.
Finding a spec by ID: glob recursively — specs/**/*-{ID}-*/ matches the story folder at any
depth regardless of type. Agents use this to locate a ticket's spec.
Cross-navigation. Every document must be reachable from every other one — a reader who opens
brief.md should be able to jump to plan.md without going back through index.md. Start each
document (immediately under its H1 title) with a one-line nav bar that links its siblings and bolds
the current file:
Story docs:**Story:** [Index](index.md) · [Brief](brief.md) · [Plan](plan.md) · [Acceptance](acceptance-criteria.md) · [Change Log](change-log.md) (the story index.md also adds · [↑ Area](../index.md); a broken-down story swaps Plan/Acceptance for [Roadmap](roadmap.md)).
Area docs:**Area:** [Index](index.md) · [Roadmap](roadmap.md).
The specs/_TEMPLATE/ files already include these bars — keep them when you copy the template.
Diagrams (mermaid): keep high-level ones (module ERD, process flow, decision tree) in the area
index.md/roadmap.md; keep story-specific ones in that story's plan.md. Add them where they aid
understanding — they are optional, not mandatory.
Language. Write every spec document in English — the spec is the engineering contract and may
be read by any developer. The one exception: you may keep a verbatim customer quote in the
Customer Request section in the customer's own language (e.g. German) when that customer is
non-English-speaking — but add an English summary alongside it. Everything else (business value,
open questions, plan, acceptance criteria, change log) is English. Summarise any foreign-language
source material (tickets, call notes) into English rather than copying it wholesale.
Spec depth — full vs lightweight. Decide up front (ask the user if unclear) how detailed the
plan.md needs to be:
Full spec (default) — the complete plan below: exact objects + reserved IDs, architecture
decisions, dependency map, labels/localization. Use when the design is non-trivial or the
developer is new to the area.
Lightweight spec — a high-level plan.md: the technical approach in prose, the affected
objects at a coarse level (types + rough responsibility; IDs optional), and a task breakdown —
but not the field-by-field / event-by-event design. Use for small, well-understood changes,
or when a senior wants a junior to make the detailed design decisions during BUILD. Flag it at the
top of plan.md:
> **Spec depth:** lightweight — detailed object design is completed during BUILD and recorded back here.
The brief.md and acceptance-criteria.md are still required and still fully testable; only
the technical detail in plan.md is deferred. During BUILD the developer fills in the exact
objects, IDs, and decisions and updates plan.md.
Procedure
1 — Gather inputs & context
Ticket ID, title, description, priority, requestor, target release, attachments. Read the ticket
from the configured work-item backend — Azure DevOps (azure-devops/*) or GitHub Issues
(github/*) — otherwise ask the user to paste it. If a bc-plan-user-story output already exists,
start from it. Walk the hierarchy: if the ticket is a child, read its parent Feature/Epic and
sibling stories so scope boundaries are informed; fold in any spike/investigation findings.
Pick the area & create the folder. Decide which module/functional area the work belongs to
(ask the user if unclear — it should match how the codebase/business is organised).
First check whether a spec for this ID already exists — glob specs/**/*-{ID}-*/. If it
does, edit that folder in place (keep its name, area, and existing docs); do not create a
second folder, rename it, or add a prefix. Only create a new folder for a genuinely new ticket.
For a new ticket, copy the scaffold: the story docs from specs/_TEMPLATE/_story/ into
specs/<Area>/<type>-<ID>-<slug>/ (bare ID, no app/ticket prefix), and — if the area has no
overview yet — specs/_TEMPLATE/_area/ into specs/<Area>/. Add the new story to the area
index.md table.
Clarify interactively — one question at a time. Resolve ambiguities by asking the user a single
focused question at a time, each with a short list of suggested answers they can pick or override
(the ask-questions tool where available — VS Code's askQuestions — otherwise ask inline; the same
conversational style bc-plan uses) — never dump a wall of questions or invent answers.
Unanswered points are not a blocker to drafting: record each under Open Questions in
brief.md, flagged awaiting customer with who owns it and by when, then continue with a
clearly-labelled assumption in plan.md. A spec can ship to review with open questions still open,
as long as they are explicit and the dependent decisions are marked as assumptions.
Align before you write. Before committing an approach to plan.md, confirm the technical
direction with the user: surface the constraints and alternative approaches you found, state your
key assumptions, and validate the ones that carry real design risk. If the answers materially change
scope, loop back to gathering — and if the requirement itself changes (not just the design), hand
back to bc-plan. Keep it proportionate (a one-line check for a small change; confirm the key ADRs
for a non-trivial design).
2 — brief.md (non-technical)
Ticket reference & URL · customer request in their words (with an English summary if it was in
another language) · business value · priority (P1/P2/P3) · requested by · target release ·
open questions (each flagged answered or awaiting customer with an owner) · out of scope.
3 — plan.md (for the developer)
One-paragraph technical summary, then the affected AL objects table (Object Type · ID · Name ·
Action · Notes), followed by: new objects (with reserved IDs) · existing objects to modify (with
reason) · technical approach · dependencies · risks · minimum BC version · performance notes.
For a full spec, follow references/plan-authoring.md for the
object table, ID reservation (Object ID Ninja), architecture decisions (ADRs), the dependency map,
labels/localization, and diagrams. For a lightweight spec, keep this section high-level (approach
coarse object list) and let BUILD fill in that detail.
Always include, inline, regardless of depth:
3e — Scope boundaries
State explicitly what this spec does and does not cover, referencing sibling stories where
responsibility lands elsewhere (e.g. "reporting is handled by us-1235, not here").
3f — Task breakdown
End plan.md with an ordered, checkable task list — the breakdown BUILD executes:
## Task Breakdown- [ ] Create table 50100 "…" with fields …
- [ ] Add event subscriber in codeunit 50103 …
- [ ] Add page extension 50101 to surface the field
- [ ] Write tests for AC-01…AC-03
For a large or multi-part story, promote this checklist into task sub-folders instead (see
references/task-breakdown.md): each NN-<task-slug>/ gets its own
plan.md (with its own Task Breakdown) + acceptance-criteria.md, the story keeps
brief/roadmap/index/change-log, and the roadmap sequences the tasks. Keep the flat
single-plan.md form for everything else.
4 — acceptance-criteria.md
Given/When/Then, numbered AC-01…. Add edge cases, error scenarios (expected messages),
testing notes (prerequisite data/setup), and a customer sign-off placeholder.
5 — change-log.md
Seed with v1.0:
| Version | Date | Author | Change | Requested By |
|---|---|---|---|---|
| 1.0 | YYYY-MM-DD | <you> | Initial spec created | <requestor> |
6 — index.md (story overview)
A short navigation page: status (Draft / Spec review / In development / In test / Done), a link to
the work item, one-paragraph overview, links to the sibling documents, and an optional diagram.
Also add/update the row for this story in the area index.md so the module overview stays current.
7 — Present & refine
Show the user a short summary (approach, key decisions, acceptance criteria, open questions) and
handle their response: changes → revise the affected docs and keep the whole spec in sync
(plan, acceptance-criteria, index status, a change-log row); questions → clarify
one-at-a-time; alternative → loop back to align (or to bc-plan if the requirement changed);
approval → run the quality gate and open the spec PR. Don't open the PR until the user approves
or explicitly asks you to commit the draft as-is.
Quality gate — before opening the spec PR
(For a lightweight spec, the object-detail items — reserved IDs, architecture decisions,
dependency map, labels/localization — are deferred to BUILD; the depth marker, task breakdown,
brief, and acceptance criteria are still required.)
Story documents present — flat story:index, brief, plan, acceptance-criteria, change-log; broken-down story:index, brief, roadmap, change-log + each NN-<slug>/ with plan + acceptance-criteria.
Story lives under the right area folder; area index.md lists it.
plan.md ends with a task breakdown (per task, for a broken-down story); broken-down tasks are sequenced in roadmap.md with status + approval state.
Every document opens with its nav bar (story / task / area).
Object IDs reserved from the assigned range (Object ID Ninja, or marked "verify").
No base-table modifications planned (use table extensions).
≥ 3 acceptance criteria.
Architecture decisions recorded for any non-obvious choice.
