Use when migration discovery needs an evidence-backed legacy-system BRD for one capability, a daily-delivery BRD draft with exception-only review, or an internal POC BRD draft from sparse/partial context so stakeholders can review scope early. Produces SME-reviewable observed behaviors, inferred rule seeds, open gaps, traceability, and BRD-stage validation scenario seeds. This is the primary near-term old-system discovery artifact; it does not compare against the new system, create an SDD handoff, or mandate target-system changes.
インストール
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Use when migration discovery needs an evidence-backed legacy-system BRD for one capability, a daily-delivery BRD draft with exception-only review, or an internal POC BRD draft from sparse/partial context so stakeholders can review scope early. Produces SME-reviewable observed behaviors, inferred rule seeds, open gaps, traceability, and BRD-stage validation scenario seeds. This is the primary near-term old-system discovery artifact; it does not compare against the new system, create an SDD handoff, or mandate target-system changes.
Legacy BRD Writer
Skill Card
Field
Notes
Problem solved
Turns legacy analysis evidence into a stakeholder-readable old-system BRD for migration discovery and SME review.
Legacy BRD Package with behaviors, open questions, validation scenario seeds, and review status.
Core prompt strategy
Separate observed behavior from inferred rules, keep target-system comparison out, and write reviewable business language with evidence tags.
Upstream skill
legacy-ibmi-module-analyzer or an accepted module-first context package.
Downstream consumer
SMEs, gap-analysis / old-vs-new reviewers, legacy-spec-writer, and decision writers.
Validation standard
BRD sections complete enough for the requested mode, evidence strength declared, unresolved gaps visible, and approval/spec gates not bypassed.
Known risk
Mistaking old-system description for a mandate to preserve every behavior in the target system.
Practical example
Given an approved order-entry module overview with Program Flow and Data Flow evidence, produce a BRD package that lists observed hold-release behavior and questions for SME confirmation.
Purpose
Synthesize one legacy-system Business Requirements Document (BRD) Package
from an approved module analysis, making the distinction between observed
behaviors, inferred rules, SME decisions, open gaps, BRD-stage validation
scenarios, and traceability visible and reviewable by non-technical
stakeholders.
In the current migration-discovery operating model, the BRD is the primary
near-term output. Its job is to prepare the old system's business
understanding, not to compare old and new systems or force a handoff to
delivery. Old-vs-new comparison, No-gap / Gap1 / Gap2 disposition,
spec-writing, and SDD handoff are downstream activities that happen after the
legacy BRD is approved and after new-system context is available.
The BRD is a business synthesis layer, not a technical specification or
handoff package. It shows:
The business process story in SME language: triggering event, actors,
customer/account impact, business state changes, normal path, exception paths,
controls, and unresolved policy questions
What the legacy system demonstrably does (BEH-*)
What business rules we infer from evidence (BR-* seeds from module, status
needs_sme_review or draft)
What questions remain unresolved (TBD-*)
How confident each claim is, derived from linked evidence records
Which business validation scenarios (VAL-*) SMEs and downstream teams
should use to review the BRD scope
The BRD consumes module analysis, flow analyses, program analyses, and SME /
BA legacy context. It does not produce formal acceptance criteria, formal
TC-* test cases, modernization decisions, target-platform choices,
old-vs-new comparison, target-system requirements, or SDD handoff files.
VAL-* entries are scenario seeds only; AC-* belongs to
legacy-spec-writer, and formal TC-* belongs to
legacy-golden-master-test-planner.
For the standard migration-discovery path, the BRD is code-backed. A
module-first context package or document-normalized draft can inform
the BRD, but it does not replace 01_inventory/object-map.md, per-program
analysis, or flow analysis. A context-only BRD is allowed only as a
non-approved draft with named owner risk acceptance and visible TBD blockers.
For internal POC validation, generate the BRD earlier when the requester
asks for speed, a POC artifact, or a draft for stakeholder alignment. Missing
object maps, program analyses, flow analyses, SME sign-off, OCR, Markdown, or
source eligibility do not block drafting. They block only BRD approval, spec
promotion, SDD handoff, and strong evidence claims. The package status must be
poc_draft, every weak/candidate statement must carry a low-confidence source
label or TBD-*, and the review artifacts must say the BRD is for internal POC
discussion only.
For daily delivery, generate the BRD after the automated context /
inventory / program / flow / module / data-model chain has run as far as the
available authorized evidence allows. Daily delivery uses
mode: daily_delivery, status: delivery_draft,
evidence_mode: daily_delivery, and review_policy: exception_only. Missing
or weak non-critical evidence becomes TBD-*, review questions, or
delivery-risk-summary rows instead of forcing per-skill approvals. Blocking
exceptions still stop delivery: unresolved authorization/redaction, unknown
sensitivity, scope ambiguity, high-risk contradictions, or a request to use the
draft as spec / SDD / audit-baseline input.
BRD writing follows a source-of-truth firewall:
confirmed_by_sme and approved code_backed evidence may become BRD
conclusions.
source_documented evidence may support source mapping or SME questions
until reviewed.
candidate_only, generated_draft, questions_only, and missing inputs
must become TBD-*, review questions, or deferred evidence gaps.
A generated or weak module-context flow never makes a BRD section covered by
itself.
The BRD body follows the SME-required functional analysis shape. Sections 1-9
are mandatory in brd.md: Function Purpose, Business Scenarios / Use Cases,
Channels, User Interface / User Touchpoints, System Interfaces, Process Flow,
Validation Rules, Error Handling, and Dependencies. Sections 10-12 are optional
and evidence-backed only: Security / Authentication Requirements, Supporting
Workflow or Design Notes, and Source Document Mapping.
Document-quality or readiness criteria do not belong in brd.md. Checks
such as "the BRD clearly explains the business boundary" evaluate the artifact,
not the business capability; keep them in brd-review.md or traceability
validation instead of adding a generic Success Criteria section to the BRD.
When to Use
Trigger on any of these signals:
Module analysis is approved or approved_with_non_blocking_tbd and
stakeholders need a business-readable artifact for scope review
Internal POC stakeholders need an early, explicitly non-approved BRD draft
from sparse source documents, document-intake metadata, context packages, RAG,
screenshots, diagrams, or partial module notes so they can validate direction
before strict discovery gates are complete
Daily delivery stakeholders need a BRD iteration from the best available
code-backed evidence with one consolidated review pack instead of separate
approval stops for inventory, program, flow, module, and data-model artifacts
SME owner wants to validate business scope and rules before forwarding to
legacy-spec-writer
SME / BA / vendor stakeholders need test-aware scenario seeds to discuss
coverage, SOW scope, and missing runtime evidence without creating formal
test cases yet
You are orchestrating a discovery phase and need a non-technical bridge
artifact that explains the old system before any gap-analysis or forward-SDLC
decision
Migration stakeholders need an as-is legacy BRD that can later be compared
against a new-system BRD or product design in a separate comparison /
gap-analysis process
Legal, compliance, or product stakeholders need clear visibility into what
rules are observed vs. inferred vs. uncertain
When NOT to Use
Do not trigger when:
The output is code or target-platform implementation (use
legacy-spec-writer then build-agent-skill)
The requester wants an SDD handoff package or Atlas-ready delivery bundle
(use legacy-brd-to-sdd-handoff only after approved BRD + approved spec)
The requester wants old-vs-new No-gap / Gap1 / Gap2 classification or
target-system scope decisions in the BRD itself. Finish the legacy BRD first,
then route comparison to the post-BRD migration disposition process.
The requester has an explicitly approved technical-spec-only bypass and
accepts the missing BRD review as a documented risk (route to
legacy-spec-writer with that bypass recorded)
No SME is available to review and approve the BRD, unless the requester
explicitly asks for an internal POC draft or daily delivery draft. POC and
daily drafting may continue, but formal approval remains blocked.
The module analysis is below approved_with_non_blocking_tbd status and
the requester expects an approved/code-backed BRD. For internal POC requests,
continue with status: poc_draft and carry missing module evidence as
TBD-* approval blockers.
The selected module / capability is context-only and lacks
01_inventory/object-map.md, required program analyses, or required flow
analyses, unless the requester explicitly asks for daily delivery, an internal
POC, or context-only draft and accepts that it cannot be approved as
code-backed
This skill is a migration discovery and business synthesis layer. If you
find yourself writing formal acceptance criteria, minting formal TC-* test
cases, minting DEC-* ids, specifying target architecture, or producing an
Atlas/SDD package, you are in the wrong skill. Route to
legacy-spec-writer, legacy-golden-master-test-planner, or
legacy-brd-to-sdd-handoff as appropriate.
It is also the wrong output shape if the BRD reads like a program walkthrough.
Runtime chains, object lists, file-copy details, and call-sequence summaries are
evidence, not the main BRD narrative. Keep program names in evidence references,
traceability, or a short appendix when needed; do not make them the reader's
primary path through the business requirement.
Role
You are the business requirements synthesizer for one capability.
You must:
translate module / flow / program evidence into business-process language
before drafting the BRD body
write an as-is business narrative that a SME, BA, operations owner, product
owner, or compliance reviewer can discuss without knowing IBM i object names
extract observed behaviors (BEH-*) from flow and program analyses without
invention or inference
aggregate inferred rules (BR-* seeds) from the module analysis; keep their
status as needs_sme_review (do not promote to approved)
fill all SME-required BRD sections 1-9, using TBD-* where required evidence
or SME confirmation is missing
in internal POC mode, draft sections 1-9 from the best available authorized
context, but mark weak or document-only claims as low-confidence candidate
statements or TBD-* instead of blocking the BRD artifact
in daily delivery mode, draft sections 1-9 from available code-backed and
source-linked evidence, carry non-critical gaps to the daily review pack, and
stop only for blocking exceptions
enforce the BRD source-of-truth firewall: only confirmed SME facts and
code-backed evidence can become BRD conclusions
include optional sections 10-12 only when security/auth details, workflow or
design notes, or source document mappings are available from evidence or SME
input
distinguish knowledge type (observed_behavior / inferred_business_rule)
on every claim, and link each claim to evidence records whose strength is
recorded in the evidence index per
docs/evidence-and-knowledge-taxonomy.md
surface unresolved items (TBD-*) with category and resolver
draft BRD-stage validation scenario seeds (VAL-*) that map to existing
BEH-*, BR-*, and EV-* references
frame unclear scope as SME-answerable boundary questions, not as a generic
problem statement about document fragmentation or analysis-process risk
keep artifact-readiness checks in brd-review.md; do not put generic
document-quality criteria in the BRD body
refuse to produce formal acceptance criteria, modernization decisions, or
platform choices — those belong in legacy-spec-writer
refuse to classify legacy behavior as No-gap / Gap1 / Gap2 inside the BRD;
that classification belongs after BRD approval when new-system context is
available
refuse to produce formal TC-* test cases or invented exact expected outputs
— those belong in legacy-golden-master-test-planner after spec approval
require SME sign-off before the BRD leaves in_review status
keep internal POC BRDs at poc_draft until the project explicitly reruns the
standard code-backed / SME-reviewed approval path
keep daily delivery BRDs at delivery_draft until a consolidated review
records accepted_for_daily_delivery; do not convert that decision into
approved
You must not:
present the BRD as a direct runtime chain, call graph, program inventory, file
movement list, or object-by-object analysis
create a standalone Problem Statement section that mixes business scope,
evidence gaps, technical coupling, and downstream rework risk
create a generic Success Criteria, Success Criiteria, Document Success Criteria, or similar section in brd.md; those checks belong in
brd-review.md, not the BRD body
invent channels, user interfaces, system interfaces, security requirements,
diagrams, source documents, or dependencies just to satisfy the BRD shape
invent business rules beyond what the module analysis suggests + SME
confirmation
use AI-organized module context, generated-draft diagrams, or
candidate-only module rows as BRD conclusions
invent a new-system comparison, gap classification, risk assessment, or
gap-analysis disposition
promote a BR-* seed to approved status in the BRD; SME confirmation is
recorded as review input, and legacy-spec-writer performs the final rule
promotion in spec.yaml
generate formal TC-* test cases or exact expected outputs
include target platform or modernization decisions
include SDD handoff package content
mark No-gap, Gap1, or Gap2 status inside the BRD
collapse observed behavior into inferred rules without marking the
distinction
treat high-confidence inferences as facts without SME validation
Inputs
Accept:
Approved module analysis (04_modules/<MODULE-SLUG>/ with
module-overview.md, 03-program-flow.md, 04-data-flow.md, and
module-review-checklist.md at approved or
approved_with_non_blocking_tbd)
Prefer module analyses whose module-overview.md includes the BRD
Functional Analysis Input Crosswalk and BRD Source Eligibility Crosswalk.
Use those crosswalks to populate SME required sections 1-9 only where
Source Eligibility is brd_conclusion_allowed; carry
needs_sme_review, questions_only, missing, generated-draft, and
candidate-only areas as TBD-* instead of rediscovering them from program
or flow details.
Approved code evidence backbone for standard BRDs:
01_inventory/inventory.yaml, 01_inventory/object-map.md, in-scope
02_programs/<MODULE>/<OBJ>/program-analysis.md, and in-scope
03_flows/<MODULE>/flow-<FLOW-SLUG>.md
Daily delivery inputs: authorized module context, RAG/source metadata, a
program-flow seed, or partial code-backed artifacts that have already run as
far as the orchestrator can safely take them. The BRD must carry
non-critical gaps into the daily review pack and must not claim formal BRD
approval.
One or more capability seeds — selected from the module overview's
Capability Seeds; the SME has confirmed each is a distinct, in-scope business
capability
Capability-owner SME (required for sign-off)
Optional: BAU notes or additional context from the SME
Internal POC draft inputs: any authorized, known-sensitivity combination
of module context package, document-intake manifest, source-document metadata,
RAG/code-knowledge output, screenshots/diagrams, PDF/PNG metadata, SME notes,
BA notes, or partial module notes. These can start a poc_draft BRD when
code-backed evidence is not ready.
Stop and require clarification if:
Module status is below approved_with_non_blocking_tbd and the requested
output is an approved/code-backed BRD → route back to
legacy-ibmi-module-analyzer. For daily delivery, continue as
delivery_draft only when blocking exceptions are absent. For internal POC,
continue as poc_draft.
Standard BRD requested but the code evidence backbone is missing
object-map.md, required program analyses, or required flow analyses → create
an internal POC/context-only draft only if the user accepts non-approved
status; otherwise route to the earliest missing code-backed skill before BRD
approval
Context-only draft requested without named owner risk acceptance → record the
current requester / project POC owner as draft-only risk acceptance when the
user explicitly says this is an internal POC; otherwise request risk
acceptance or route back to source/object evidence collection
A required BRD section is supported only by candidate_only,
generated_draft, questions_only, or unreviewed source_documented module
rows → create TBD-* / SME questions; do not draft it as a conclusion
The selected capability seed has blocking TBDs in the module analysis →
escalate to SME before approval; in daily delivery or internal POC mode,
include the TBDs and continue drafting only if they do not block the current
delivery iteration
No SME owner is assigned → keep owner as TBD and draft only in daily
delivery or internal POC mode; approval remains blocked
The capability boundary is ambiguous (does flow X belong here or to another
capability?) → in daily delivery or internal POC mode, draft the most likely
boundary as a review hypothesis only when the ambiguity is not high-risk, and
add explicit boundary TBD-*; otherwise SME must decide
The user asks the BRD writer to classify old-vs-new No-gap / Gap1 / Gap2 or
decide target-system scope → explain that the BRD must first capture the
legacy baseline; comparison/disposition is a post-BRD process
templates/brd.md, templates/brd-review.md,
templates/validation-scenarios.md, templates/traceability.md as starting
structure
references/synthesis-rules.md for how to extract behaviors and aggregate
rules
references/anti-hallucination.md for what the BRD must refuse to invent
Follow:
../../docs/id-conventions.md for stable IDs (CAP-*, BEH-*, BR-*,
TBD-*, VAL-*, etc.)
../../docs/evidence-and-knowledge-taxonomy.md for knowledge-type /
evidence-strength distinction
Use only these evidence strength values in evidence records:
confirmed_from_code, observed_in_runtime, confirmed_by_sme,
strongly_inferred, weakly_inferred, needs_sme_review,
contradictory, missing
../../docs/data-collection-and-redaction.md for evidence sensitivity
checks
../../docs/input-readiness-rubric.md for input readiness scoring
Post-BRD comparison note:
The legacy BRD is the baseline input for later old-vs-new comparison.
No-gap / Gap1 / Gap2 classification, risk assessment, and formal gap analysis
are outside this skill and must not be written into brd.md,
brd-review.md, validation-scenarios.md, or traceability.md.
Examples:
examples/brd-positive/ — one approved BRD (Credit Limit Enforcement from
CARD-AUTH module)
examples/incomplete-input-negative/ — BRD with non-blocking TBDs (awaiting
SME decision on open questions)
Step Contract
This skill is one step in the Legacy Spec Factory reverse chain. It conforms to
the canonical Step Contract shape — see
../legacy-step-contract/SKILL.md and
../legacy-step-contract/references/step-contract.md for full field-level rules.
The summary below is normative for this skill.
Input
Required for standard code-backed BRDs: approved
04_modules/<MODULE-SLUG>/ (module-overview.md, 03-program-flow.md,
04-data-flow.md, and module-review-checklist.md at approved or
approved_with_non_blocking_tbd, evidence_mode: code_backed); one or more
CAP-* capability seeds from the module overview, validated by SME; named
capability-owner SME; all referenced flow-<FLOW-SLUG>.md and
program-analysis-<OBJ-ID>.md at approved or
approved_with_non_blocking_tbd; approved 01_inventory/inventory.yaml;
01_inventory/object-map.md.
Required for context-only BRD drafts: module overview or context package
marked evidence_mode: context_only; named owner risk acceptance; missing
inventory, object-map, program-analysis, and flow-analysis artifacts carried
as TBD-*; source eligibility labels preserved; BRD status limited to
draft or in_review.
Required for daily delivery BRD drafts: explicit delivery_mode: daily_delivery; authorized, known-sensitivity inputs; a capability/module
scope or program-flow seed accepted by the orchestrator; all blocking
exceptions either absent or recorded with owner decision; non-critical gaps
carried as TBD-*, review questions, or daily review pack rows; BRD status
limited to delivery_draft until formal baseline promotion.
Required for internal POC BRD drafts: an explicit user or project-owner
request for internal POC output; at least one authorized, known-sensitivity
source record or context artifact; a capability/module slug or provisional
capability name; and a visible note that the BRD is non-approved and cannot
feed spec, SDD handoff, or delivery decisions until standard gates pass.
Missing inventory, object-map, program-analysis, flow-analysis, SME owner,
and source eligibility become TBD-* approval blockers, not draft blockers.
Optional: BAU notes, supplemental legacy-system context, source document
pointers, runtime observations, and policy notes from SMEs.
Input readiness scoring:
0-5 blocked: evidence authorization unresolved; unknown/unauthorized
sensitivity; no recognizable capability/module scope; request asks for
approved/spec/handoff use despite missing required gates; or no authorized
source/context artifact exists.
6 minimum_pass: approved code-backed module, selected SME-confirmed
capability seed, named capability-owner SME, approved upstream analyses,
and approved object map are present. Daily delivery, context-only, and
internal POC drafts can reach minimum_pass for draft review only when
authorization is clear and missing evidence is visible as TBD-*; they
cannot reach BRD approval.
7-8 usable: BAU notes, supplemental SME context, and related open-TBD
history are supplied.
9-10 strong: examples of real scenarios, exception cases, runtime
observations, policy notes, and downstream reader context are also supplied.
Missing supplemental context does not block BRD drafting; it should produce
clearer SME questions instead of invented rules.
Readiness checks: module and all upstream analyses at required status for
standard approval; object map exists for standard BRDs; no sensitivity: unknown evidence in scope; SME owner available to approve BRD. For daily
delivery and internal POC, these become approval-readiness checks, not
drafting prerequisites.
Stop conditions: unresolved evidence authorization/sensitivity; no
recognizable capability/module scope; request asks BRD writer to classify
No-gap / Gap1 / Gap2 or decide target-system scope; or the requester demands
approved/code-backed BRD use while standard gates are missing. Missing module
approval, object map, program analyses, flow analyses, risk acceptance, SME
owner, or clear boundary does not stop a daily delivery or internal POC draft
unless it is a blocking exception; otherwise it creates explicit TBD-*
approval/spec blockers.
Execution
Procedure: see the Workflow section below (9 ordered steps).
Allowed inference: lifting BEH-* from flow control flow, program branch
points, and error handling (factual statements about what the legacy system
does); aggregating BR-* seeds from module overview and cross-checking
against program/flow context; computing confidence levels based on evidence
strength; surfacing contradictions as TBDs; converting questions-only module
rows into SME questions.
Forbidden assumptions: inventing business rules beyond module BR-* seeds +
SME confirmation; promoting a BR-* seed to approved status (only
legacy-spec-writer may do that); generating formal acceptance criteria;
generating formal TC-* test cases or invented exact expected outputs;
specifying target platform or modernization decisions; adding old-vs-new
comparison or target-system disposition; reading raw IBM i source code
(consume upstream analyses only); treating weak inferences, generated-draft
context, candidate-only rows, or questions-only crosswalk rows as facts.
Do not label a context-only document or RAG claim as confirmed_from_code.
TBD handling: unconfirmed rule → BR-* seed with status: needs_sme_review
(marked in BRD); contradictory evidence → TBD-* with category: contradictory_evidence and resolver; missing context → TBD-* with
category: sme_questions or category: evidence_gaps; legacy behavior that
needs later comparison or promotion review may be referenced by a TBD-*
with category: downstream_handoff_blockers and blocking: no for BRD
approval unless the SME marks it business-critical. In daily delivery mode,
non-critical missing evidence should become daily review questions or
delivery-risk-summary rows; high-risk contradictions remain blockers. In
internal POC mode, missing evidence, missing SME owner, missing code-backed
artifacts, and ambiguous boundaries should usually be blocking: yes (approval/spec only), not blockers for producing the POC draft.
Required sections/fields (see templates/brd.md): sections 1-9 in the
SME functional-analysis shape: Function Purpose, Business Scenarios / Use
Cases, Channels, User Interface / User Touchpoints, System Interfaces,
Process Flow, Validation Rules, Error Handling, and Dependencies. Sections
10-12 (Security / Authentication Requirements, Supporting Workflow or Design
Notes, Source Document Mapping) are optional and evidence-backed only.
Required IDs: mints BRD-* for the document, VAL-* for BRD-stage
validation scenario seeds, and TBD-* for open
questions. Reuses CAP-*, OBJ-*, EV-*, BEH-*, BR-* seeds,
MODULE-*, FLOW-* from upstream. If a new candidate rule appears during
BRD review and has no upstream BR-*, record it as a TBD-* requiring
module/spec review instead of minting a new BR-* here. Does NOT mint
DEC-*, AC-*, IN-*, OUT-*, STEP-*, TC-*, or new BR-*.
Review status: status: poc_draft for internal POC,
status: delivery_draft for daily delivery, otherwise draft →
in_review → approved (SME sign-off).
The approved BRD is ready as the legacy-system discovery baseline.
legacy-spec-writer consumes it only after a separate post-BRD comparison /
promotion decision says the capability or selected legacy behavior should
move beyond discovery. Direct module-to-spec generation is an exception that
requires an explicit technical-spec-only bypass with approver and risk
acceptance.
A context-only BRD draft cannot become approved until the Code-Backed
Analysis Gate passes; keep missing object map, program analysis, and flow
analysis as blocking TBD-* items in brd-review.md and traceability.md.
A daily delivery BRD may be accepted for the current iteration through
review-decision.yaml.decision: accepted_for_daily_delivery, but it cannot
become approved or feed spec / SDD handoff until the approved-baseline path
reruns the standard gates.
An internal POC BRD cannot become draft/in_review for formal SME review or
approved until its POC blockers are either resolved or explicitly converted
into a context-only / standard BRD path.
Validation
Validation has three distinct layers, in order. None substitutes for another.
4a. Mechanical Validation
What can be checked by a script, schema, or deterministic linter:
required files exist at expected paths
all referenced IDs resolve (no dangling EV-*, BEH-*, BR-*, CAP-*,
VAL-*, TBD-*)
ID prefixes match docs/id-conventions.md
every claim has at least one linked evidence item (EV-* or BEH-*)
no sensitivity: unknown in evidence references
traceability table is complete and consistent (all claims appear in
traceability.md)
standard BRDs cite 01_inventory/object-map.md, in-scope program analyses,
and in-scope flow analyses; daily delivery, context-only, and internal POC
drafts visibly block formal approval but may remain useful as review material
no BRD section marked as a factual conclusion is sourced only from
candidate_only, generated_draft, questions_only, or unreviewed
source_documented rows
BRD does not include acceptance criteria, modernization decisions, or target
platform details
validation-scenarios.md does not mint AC-* or TC-*
internal POC BRDs declare status: poc_draft, evidence_mode: internal_poc, a POC-only limitation note, and approval blockers in
brd-review.md / traceability.md
daily delivery BRDs declare status: delivery_draft, evidence_mode: daily_delivery, review_policy: exception_only, and consolidated delivery
review requirements in brd-review.md / traceability.md
Mechanical validation must be reproducible. If it depends on judgment, move
it to AI semantic review.
4b. AI Semantic Review
What an LLM (or careful reviewer) can check by reading the artifact against
upstream evidence:
observed behaviors (BEH-*) are factual statements about what the code / data /
logs show, not inferences
the section 6 process flow is a business process narrative, not a program
call chain; program and file names appear only when they are essential
identifiers or evidence references
inferred business rules (BR-* seeds) are supported by BEH-* and linked
EV-* content (not just ID reference)
knowledge type (observed_behavior / inferred_business_rule) is correctly
marked for every claim
evidence support is not overstated (no weak evidence record used as if it
were direct code/runtime/SME confirmation)
context-only module or document evidence is not presented as code-confirmed
behavior
internal POC candidate statements remain visibly low-confidence and are not
presented as approved BRD conclusions
daily delivery candidate statements are not presented as approved BRD
conclusions; items needed for this delivery iteration are visible in the
daily review pack
module crosswalk rows marked questions_only or needs_sme_review become
TBDs/review questions, not BRD conclusions
no invented IBM i facts (object names, fields, programs, jobs)
no scope creep into other capabilities
no acceptance criteria or platform decisions hidden in prose
validation scenarios map to existing BEH-*, BR-*, and EV-* references
without introducing new rules
TBDs are explicit, not hidden inside prose
AI semantic review must call out uncertainty rather than smooth it over.
4c. SME / Human Approval
What only a domain expert can decide:
are the observed behaviors accurate to legacy system reality?
are the inferred rules actually business rules, or are they implementation
artifacts?
is the capability scope and boundary correct?
are there unspoken business rules the code doesn't show?
do the VAL-* scenario seeds cover the important happy path, exception,
boundary, and manual review cases for this BRD?
are TBDs blocking or non-blocking for BRD approval or later comparison,
gap-analysis, or spec-writing?
is the BRD safe to promote to approved as a migration-discovery artifact?
SME approval is a control point, not a rubber stamp. The Step Validation
Report must record the SME's name (or role), the date, and the specific
decision.
Workflow
Confirm Capability Scope
Take one or more CAP-* seeds from the module overview's Capability Seeds
Validate with SME: is each a distinct capability worth its own BRD?
Assign brd_id and confirm capability.{id, name, slug, owner}
Define scope.in_scope and scope.out_of_scope from SME
Record evidence_mode: code_backed for standard BRDs, daily_delivery
when the orchestrator selected exception-only day-to-day delivery,
context_only
when named risk acceptance exists and the BRD will remain non-approved, or
internal_poc when the requester explicitly wants early POC output from
partial context. For daily_delivery, set status: delivery_draft and
review_policy: exception_only. For internal_poc, set
status: poc_draft and record the requester/project POC owner as
draft-only risk acceptance.
Collect Evidence Bundle
Gather every EV-* referenced by flows / programs / module that touch this
capability
Confirm the code-backed evidence backbone exists for standard BRDs:
01_inventory/object-map.md, in-scope program analyses, and in-scope
flow analyses
Populate evidence index with source, type, sensitivity, and confidence
Confirm sensitive flag is set on all evidence; if any sensitive: unknown, stop and request redaction review
If daily delivery, carry non-critical evidence gaps into TBD-* rows,
review questions, and the daily-delivery review pack; stop only for
blocking exceptions.
If context-only, create TBD-* blockers for each missing code-backed
artifact and prevent approval in brd-review.md
If internal POC, register source metadata even when extraction is partial
or unavailable. Missing Markdown, OCR, object-map, program analysis, flow
analysis, and SME owner become approval/spec blockers, not draft blockers.
Read the module BRD Source Eligibility Crosswalk. Partition input into
BRD conclusions allowed, needs SME review, and questions only.
Translate Technical Evidence into Business Process Language
Create a short business-first process flow before writing BEH-*
Describe the triggering business event, primary actor or system party,
business object/state affected, normal outcome, exception outcomes,
operational controls, and handoffs
Use domain nouns (cardholder, replacement request, address verification response, exception queue) before implementation nouns
(program, file, library, commit, copy)
If source materials are scattered or the capability boundary is unclear,
do not write a standalone Problem Statement. Capture the SME decision
needed under Scope Clarification Need and create explicit TBD-* items
for actors, triggers, state transitions, handoffs, or in/out-of-scope
boundaries.
Do not add Success Criteria or document-readiness bullets to brd.md.
If those checks are useful, place them in the author/synthesizer preflight
section of brd-review.md.
Populate the SME-required sections explicitly:
Function Purpose, Business Scenarios / Use Cases, Channels, User
Interface / User Touchpoints, System Interfaces, Process Flow, Validation
Rules, Error Handling, and Dependencies.
Populate a section as factual BRD prose only from
brd_conclusion_allowed rows. For needs_sme_review or
questions_only rows, write a TBD-* or review prompt instead of a
conclusion. In internal POC mode, you may include directional draft text
from source-documented or candidate context only when the sentence is
explicitly marked as a POC hypothesis / low-confidence review prompt and
linked to source metadata or a TBD-*.
Include Security / Authentication Requirements, Supporting Workflow or
Design Notes, and Source Document Mapping only when evidence or SME input
supports them; otherwise omit the optional section or create a TBD-* if
confirmation is required.
If the only available description is a runtime chain, convert it into 3-6
business phases and create TBD-* questions for any phase whose business
purpose is unclear
Keep program names, file names, and object IDs in evidence references,
traceability, or appendix notes unless a SME must recognize the object to
confirm scope
Lift Observed Behaviors (BEH-*)
From flow analyses' control flow points, branch conditions, error handlers
(factual statements about legacy system behavior)
Each BEH must trace to ≥1 EV-*
These are factual — what the system does, not why or whether it's correct
Phrase each BEH as business-visible behavior first; implementation details
may appear after the business behavior only as supporting context
Aggregate Business Rules (BR-*)
From module analysis's BR-* seeds in module-overview.md / Capability
Seeds / BRD crosswalk
Cross-check against flow / program analyses
Keep each BR-* at status needs_sme_review (do NOT promote to approved);
that is spec-writer's job
Reuse upstream BR-* seeds only. If BRD review reveals a candidate rule
with no upstream BR-*, record a TBD-* for module/spec review instead
of minting a new BR-*
Each BR must:
Reference ≥1 BEH-* it abstracts
Reference ≥1 EV-* that supports it
Be marked knowledge_type: inferred_business_rule
Draft Validation Scenario Seeds (VAL-*)
Convert approved observed behaviors and inferred rule candidates into
SME-reviewable business validation scenarios
Each VAL-* must map to at least one existing BEH-* or BR-* and at
least one EV-*
Cover happy path, exception, boundary, and manual-review cases where the
evidence supports them
Mark readiness as ready_for_spec, needs_sme_review, or
needs_runtime_evidence
Do not invent exact expected outputs, formal AC-*, formal TC-*, target
system behavior, or new business rules
If a useful scenario cannot be drafted safely, put it in Deferred
Scenarios with the evidence gap and resolver
Surface Open Questions (TBD-*)
Contradictory evidence → TBD-* with category contradictory_evidence
Missing context → TBD-* with category sme_questions
Ambiguous scope → TBD-* with category sme_questions
Legacy behavior that may need later comparison or promotion review →
TBD-* with category downstream_handoff_blockers only when the blocker
is visible from legacy evidence
Each TBD must name a resolver and indicate what it blocks: BRD approval,
later comparison, gap analysis, spec-writing, or SDD handoff
Build Traceability
Generate traceability.md cross-reference table
Every BEH-* and BR-* must have ≥1 supporting EV-*
Every VAL-* must map back to BEH-* or BR-* and supporting EV-*
Every TBD must be listed with category and resolver
Verify complete coverage (no claim is missing from the table)
Prepare for SME Approval
Mark status: poc_draft for internal POC output; mark
status: delivery_draft for daily delivery output; otherwise mark
status: in_review
For daily delivery, generate or reference the consolidated
07_sme_reviews/<CAPABILITY-SLUG>/daily-delivery-review-v1/ pack and
record that final acceptance is accepted_for_daily_delivery, not
approved
Generate brd-review.md checklist
Generate validation-scenarios.md for SME scenario coverage review
Capability owner SME approves the BRD; BRD is then status: approved
Keep BR-* review status as needs_sme_review in the BRD even when SME
notes confirm it for later spec promotion
If SME finds issues, mark status: blocked with specific findings
Workflow State Write-Back (history-only discovery gate)
This is the standard business-review gate for the migration-discovery phase.
It produces a business-facing legacy BRD after module analysis. It does NOT
advance the numeric stage_id; it records BRD review status and blocking
metadata until approval. It does NOT mutate current_focus.
history:-at:<ISO8601>
skill: legacy-brd-writer
capability_id: <CAP-* from current_focus>
stage_after: <UNCHANGED stage_id>
artifact: <path to brd.md, e.g. 05_brds/<CAPABILITY-SLUG>/brd.md>
note: "BRD authored for <CAP-*> — status: poc_draft | delivery_draft | draft | in_review | approved"
Also overwrite project.last_updated_at / project.last_updated_by.
Permitted side-effect: if BRD authoring surfaces inferred business rules or
open gaps not already in
capabilities[<CAP-*>].blocking.*, you MAY append them to
blocking.sme_pending (rule IDs) or blocking.tbds. You MUST NOT change
stage_id, last_artifact, or last_skill — the linear stage owner remains
legacy-ibmi-module-analyzer unless a later stakeholder decision routes the
capability to legacy-spec-writer.
If workflow-state.yaml does not exist, this skill does NOT create it.
Anti-Hallucination Rules
The BRD is the first business-facing artifact. The temptation to "smooth out"
gaps or invent business logic is high. Discipline is essential.
Observed behavior is ground truth. A BEH can only be claimed when it is
grounded in:
Tier 1 (upstream flow / program analysis) — for what the system does, or
Tier 2 (named, date-stamped SME confirmation) — for context the code
doesn't show (manual procedures, business policy)
Do NOT:
Invent business rules beyond what module BR-* seeds suggest + SME
confirmation
Promote a BR- seed to approved* (only legacy-spec-writer may do that)
Generate formal TC-* test cases or invented expected outputs
(golden-master planner's job after spec approval and runtime evidence review)
Include target platform or modernization decisions (spec-writer's job)
Add old-vs-new comparison or target disposition to the BRD. No-gap,
Gap1, Gap2, risk assessment, and formal gap analysis belong after BRD
approval when new-system context is available.
Add generic BRD success criteria to brd.md; artifact-readiness checks
belong in brd-review.md, not in the business requirements document
Invent optional functional-analysis details such as channels, UI screens,
security rules, diagrams, or source documents when they are not evidenced
Claim behavior from field names or comments alone (factual evidence
required)
Pretend ambiguity is resolved by picking one interpretation when SME
hasn't confirmed → mark as TBD instead
Include any fact that doesn't trace to an EV-* in the evidence index
Instead:
If a BR seed is plausible but SME hasn't confirmed → keep it at
needs_sme_review, do not hide uncertainty
If evidence is contradictory → create a TBD-*, mark the category, and name
a resolver
If a field's meaning is unclear → create a TBD-* on semantics; do not guess
If scope is contested → create a TBD-* on boundary; let SME decide
If a test-like scenario needs runtime data or expected output evidence →
create a VAL-* with readiness: needs_runtime_evidence or defer it; do not
invent the expected result
If later comparison is needed → finish the legacy BRD first, then route the
approved BRD into the separate comparison / gap-analysis process
Quality Checklist
Before marking the BRD approved, confirm:
All four files exist at correct paths (brd.md, brd-review.md,
validation-scenarios.md, traceability.md)
Every claim in brd.md appears in traceability.md
Section 6 Process Flow is business-readable and does not read as a direct
runtime chain, object inventory, or call graph
No standalone Problem Statement section mixes business scope,
evidence gaps, technical coupling, and delivery/rework risk
No generic Success Criteria or document-quality/readiness section is
included in brd.md
Required sections 1-9 are present: Function Purpose, Business Scenarios /
Use Cases, Channels, User Interface / User Touchpoints, System
Interfaces, Process Flow, Validation Rules, Error Handling, Dependencies
Optional sections 10-12 are included only when evidence-backed or
explicitly SME-confirmed; missing optional details are omitted or tracked
as TBD-*
Every BEH-* and BR-* links to ≥1 EV-*
Every VAL-* maps to existing BEH-* or BR-* and ≥1 EV-*
validation-scenarios.md contains no formal AC-*, formal TC-*,
target architecture, or invented exact expected output
No No-gap / Gap1 / Gap2 classification, risk assessment, gap-analysis
disposition, target-system requirement, or SDD handoff content appears in
the BRD Package
No invented IBM i facts; all object names come from upstream artifacts
Knowledge type (observed_behavior / inferred_business_rule) is marked for
every claim
BR-* seeds carry status: needs_sme_review (not promoted to approved)
No acceptance criteria, modernization decisions, or platform details in
BRD
All TBD-* items carry a category (missing_inputs, evidence_gaps,
contradictory_evidence, sme_questions, downstream_handoff_blockers)
SME has reviewed and approved the BRD
Capability scope and boundaries are validated by SME
For a daily delivery BRD, the approval checklist above is not required before
daily delivery drafting. Instead, confirm:
mode: daily_delivery, status: delivery_draft,
evidence_mode: daily_delivery, and review_policy: exception_only are
visible in the BRD package.
Blocking exceptions are either absent or listed with owner / resolver in
brd-review.md, traceability.md, or the daily review pack.
Non-critical gaps are visible as TBD-*, review questions, or
delivery-risk-summary.md rows.
review-decision.yaml uses accepted_for_daily_delivery for final daily
acceptance, not approved.
The package explicitly says it cannot feed spec writing, SDD handoff,
audit baseline, or trusted knowledge publication until the
approved-baseline gates pass.
For an internal POC BRD, the approval checklist above is not required before
drafting. Instead, confirm:
status: poc_draft and evidence_mode: internal_poc are visible in all
BRD package files.
Missing module approval, object-map, program analyses, flow analyses, SME
owner, OCR, Markdown, or source eligibility are listed as TBD-*
approval/spec blockers.
Low-confidence statements are marked as POC hypotheses or review prompts,
not approved BRD conclusions.
The package explicitly says it cannot feed spec writing, SDD handoff, or
delivery decisions until standard gates pass.
Relationship to Other Skills
legacy-ibmi-module-analyzer (upstream): produces module analysis with
BR-* seeds and capability seeds. BRD consumes this output.
legacy-spec-writer (downstream): consumes module analysis plus the
approved BRD Package only after a separate post-BRD comparison / promotion
decision says the capability or selected legacy behavior should move beyond
discovery. The BRD is the business context layer for later rule promotion and
acceptance criteria; it is not itself a mandate to implement legacy-only
behavior. Direct module-only spec writing requires an explicit
technical-spec-only bypass.
legacy-golden-master-test-planner (downstream verification): consumes
approved spec acceptance criteria, runtime evidence, and approved scenario
context to mint formal TC-* golden master cases. BRD VAL-* entries are
planning seeds, not final test cases.
legacy-step-contract (parallel): defines the Step Contract shape that
this skill conforms to.
legacy-modernization-orchestrator (meta): routes to BRD-writer as the
standard legacy-discovery output after module analysis and routes to
spec-writing only after a separate post-BRD promotion / gap-analysis
decision.
From the repository root, use scripts/sync-skills.sh to create or check
runtime copies. Do not edit adapter copies directly.
No runtime-specific assumptions are baked into this canonical source.
Version History
v0.1.10 (2026-06-04): Added daily delivery BRD mode. Daily runs use
status: delivery_draft, evidence_mode: daily_delivery, and
review_policy: exception_only, consolidate findings into one review pack,
and remain barred from spec / SDD handoff until promoted through approved
baseline gates.
v0.1.9 (2026-06-03): Added internal POC BRD mode. Sparse, context-only, or
tooling-constrained inputs may now produce status: poc_draft BRDs with
visible low-confidence statements and approval/spec blockers instead of
blocking BRD authoring. Authorization/sensitivity and downstream approval
gates remain non-bypassable.
v0.1.8 (2026-06-03): Added the BRD source-of-truth firewall. BRD conclusions
may come only from SME-confirmed or code-backed evidence; source-documented
hints require review, and candidate/generated/questions-only module rows
become TBDs or SME questions.
v0.1.7 (2026-05-30): Added code-backed BRD gate. Standard BRDs now require
object-map.md, approved program analyses, and approved flow analyses before
approval; context-only BRDs require named risk acceptance and remain
non-approved drafts with visible TBD blockers.
v0.1.6 (2026-05-30): Reframed BRD writer for migration discovery. BRD is now
the primary near-term old-system artifact and remains legacy-system-only:
No-gap / Gap1 / Gap2 comparison, risk assessment, formal gap analysis, target
requirements, and SDD handoff content are explicitly post-BRD concerns.
v0.1.5 (2026-05-29): Aligned the BRD writer with BRD-first orchestration.
BRD became the required business review artifact before any later
spec-writing decision; direct module-to-spec work required an explicit
technical-spec-only bypass.
v0.1.4 (2026-05-27): SME functional-analysis alignment
Reframed brd.md around required SME sections 1-9
Made security/auth, workflow/design notes, and source document mapping
optional evidence-backed sections
Kept document-readiness checks in brd-review.md instead of BRD body
