Merge multiple IBM i program-analysis artifacts into a compact SME program-set core review, especially when an SME provides a program-flow list and wants Calculation Logic, Validation Logic, Exception Handling, and Message Inventory. Default field workflow is program-evidence first with no cross-run reuse: analyze every distinct SME-provided program for the current run, reuse only artifacts produced earlier in the same run/batch, then assemble the self-contained program-set review. Also supports approved document-repo reuse after an all-program scan has been merged: SME can clone the document repo locally, provide a program flow, and assemble a core review from approved existing program artifacts without re-scanning source. Supports source inventory cache reuse from outputs/repo-scan when fresh and explicit full transaction-flow analysis for seven trigger models when requested separately. Layer 1.5 (platform-specific) skill of the Legacy Spec Factory reverse chain.
설치
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Merge multiple IBM i program-analysis artifacts into a compact SME program-set core review, especially when an SME provides a program-flow list and wants Calculation Logic, Validation Logic, Exception Handling, and Message Inventory. Default field workflow is program-evidence first with no cross-run reuse: analyze every distinct SME-provided program for the current run, reuse only artifacts produced earlier in the same run/batch, then assemble the self-contained program-set review. Also supports approved document-repo reuse after an all-program scan has been merged: SME can clone the document repo locally, provide a program flow, and assemble a core review from approved existing program artifacts without re-scanning source. Supports source inventory cache reuse from outputs/repo-scan when fresh and explicit full transaction-flow analysis for seven trigger models when requested separately. Layer 1.5 (platform-specific) skill of the Legacy Spec Factory reverse chain.
IBM i Legacy Flow Analyzer
Skill Card
Field
Notes
Problem solved
Merges a SME-provided multi-program flow/list into one compact program-set core review without losing any requested program. Full transaction-flow analysis remains available only when explicitly requested.
Input
Approved inventory, current-run program analyses or an approved document repo clone containing all-program scan artifacts, trigger context, runtime clues, screen/report evidence, and SME notes.
Output
For SME-provided program-flow/core merge requests, program-set-sme-core-review.md only. For separately requested full transaction-flow analysis, flow-<FLOW-SLUG>.md.
Core prompt strategy
Resolve every SME-provided program to approved program-analysis evidence first: either current-run artifacts from the active scan branch or approved artifacts from a local document repo clone, aggregate compact core artifacts, preserve SME order as navigation evidence, and expose missing programs/sections in a completeness ledger.
Upstream skill
legacy-ibmi-program-analyzer and legacy-ibmi-inventory.
Downstream consumer
legacy-ibmi-module-analyzer, legacy-brd-writer, and capability/spec preparation.
Validation standard
All called programs and files map to inventory IDs, trigger model is declared, and unresolved branches are not hidden.
Known risk
Over-connecting programs into a transaction flow because names or call chains look related.
Practical example
Analyze an order-release flow from menu option through RPGLE validation, update programs, spool output, and error handling.
Purpose
For the current SME program-flow workflow, merge multiple analyzed programs
into one compact program-set SME core review. Where
legacy-ibmi-program-analyzer looks inside one program, this skill combines
the already-produced core logic from every program the SME named, without
re-reading full source and without producing a full engineering flow artifact.
Full transaction-flow analysis is still available when explicitly requested,
but it is not the default output for SME-provided program-flow intake.
The primary orientation view is a Transaction Call Map: an RDi-style
cross-program/cross-boundary call map. It is not a statement-level
flowchart. It should preserve the transaction's program and external
dependency structure while keeping full call details in tables.
The default field workflow is program-evidence first with no cross-run
reuse. For every program named by the SME in the program flow, use the source
inventory cache to locate and classify the source, run
legacy-ibmi-program-analyzer with the appropriate normal, complex, or large
strategy, and write the program-level artifacts to the delivery working branch
for this run. If the same program appears again in the same SME batch, reuse the
artifact already produced in this run; do not reuse remote-main, prior-run, or
other analysts' artifacts as the default evidence source. Only after every
SME-provided program has enough routine evidence should this skill assemble
program-set-sme-core-review.md.
Git remains the review and conflict-resolution mechanism. If two analysts
produce different program-analysis artifacts, compare them through normal branch
diff/PR review rather than silently treating older artifacts as approved input.
After a team has completed an all-program scan and merged the resulting
program-analysis artifacts into the approved document/delivery repo, a second
field workflow is supported: approved document repo reuse. In this mode the
SME clones the document repo locally, supplies a program flow/list, and the
builder reads only the requested programs' existing compact artifacts from that
local clone. This is not remote-main probing and not arbitrary prior-run cache
reuse; it is an explicit approved-artifact mode. Missing programs stay visible
instead of being invented or silently dropped. If the program is missing from
the document repo but a fresh source inventory finds it, mark it
pending_source and route only that program back to a targeted program scan.
If a fresh source inventory also does not contain it, mark it
blocked_missing_source and ask the SME to confirm the program name, library,
alias, deployment scope, or provide the missing source.
Before analyzing unresolved programs from source, this skill uses the source
repo inventory cache. By default it checks
<source-root>/outputs/repo-scan/program-list.csv and
<source-root>/outputs/repo-scan/scan-summary.yaml. If the cache exists and
scan-summary.yaml.source_revision_key matches the current clean Git source
HEAD, reuse the inventory to locate source path and size tier. If the cache is
missing, stale, or the source worktree has uncommitted source changes, rerun
legacy-ibmi-inventory repo-level scan first.
This skill supports these assembly modes:
core_review_only (default for SME-provided program-flow input) — user
provides an ordered or partial program flow/list; the analyzer produces
program-set-sme-core-review.md by first resolving every SME-provided
program to current-run working-branch program-analysis artifacts or, when
explicitly requested, approved artifacts in a local document repo clone, then
merging those compact artifacts. Do not produce flow-<FLOW-SLUG>.md in this
mode.
orchestrated — start from a trigger / entry program, discover the
involved programs, run or route missing per-program analysis first, then
assemble a full transaction flow from the generated compact artifacts. Use
only when the user explicitly asks for full flow analysis.
assemble_existing — user provides existing per-program analysis
directories, and the flow analyzer assembles them into one flow, filling only
missing artifacts when possible. Use only when the user explicitly asks for
full flow analysis.
This skill does not re-analyze individual program semantics during the
assembly step. Every program in the program set must first have program
analysis evidence from a current working-branch scan. All program tiers require
program-prefixed artifacts: <PROGRAM>-program-analysis.md,
<PROGRAM>-program-analysis-summary.yaml, <PROGRAM>-source-index.yaml,
<PROGRAM>-routine-index.md, <PROGRAM>-message-inventory.yaml,
<PROGRAM>-routine-logic-details.md, and
<PROGRAM>-routine-logic-details.yaml. Routine detail sidecars are
audit/checkpoint evidence for traceability; they do not replace the
reader-first <PROGRAM>-program-analysis.md and program-set review surfaces.
Optional sidecars such as <PROGRAM>-file-io-inventory.yaml,
<PROGRAM>-field-mutation-matrix.yaml, and <PROGRAM>-sql-inventory.yaml are
required only when program-analyzer triggers produced them or when the flow
claim needs file I/O, persisted mutation, or SQL evidence. Complex and large
programs must retain enough deep-read evidence, such as
<PROGRAM>-deep-read-plan.md, <PROGRAM>-all-routine-coverage-ledger.md, and
routine-logic-details/<PROGRAM>-deep-read-batch-*.md, to make handoff stable
under weak LLM constraints. If required core artifacts or claim-specific optional
artifacts are missing, route only that program back to
legacy-ibmi-program-analyzer instead of concatenating full Markdown analyses
or restarting the whole flow.
This skill also supports two analysis intents:
standalone_exploratory — default. Use for quick SME validation, partial
flow scans, or assembling existing program artifacts before the chain is fully
approved. Missing inventory approval, missing sidecars, or incomplete
program approval become downstream-readiness gaps / TBDs, not immediate
blockers, unless the trigger is ambiguous or the requested flow cannot be
scoped to one business transaction.
chain_ready — strict downstream mode. Requires approved inventory,
approved program analyses, required compact sidecars, coverage gates, and
trigger clarity before the flow can feed module/BRD/spec work.
This skill does not infer business rules. It captures the structure
of the transaction so that downstream legacy-spec-writer (via
legacy-ibmi-module-analyzer) can derive business rules with proper SME
involvement.
Inputs
Accept:
Analysis intent — standalone_exploratory by default, or chain_ready
when the user explicitly asks for downstream-ready output.
SME-provided program flow — an ordered or partial list of programs,
entry/exit hints, menu/job/API context, or a rough sequence from the SME.
When the SME provides a program flow for the current core-merge workflow,
default flow_scan_mode to core_review_only, preserve the supplied order as
SME navigation evidence, and output program-set-sme-core-review.md.
Reconcile contradictions against upstream Call Evidence as notes/TBDs rather
than silently rewriting the SME-provided list.
Multiple SME-provided program flows — one request may contain multiple
named flow blocks. Treat each block as a separate program set with its own
review name, program list, manifest, and {review_slug} output folder under
delivery_workspace_profile.program_set_review_parent. Reuse the same
source inventory cache, delivery working branch, and PR when they belong to
the same SME batch. If the same program appears in more than one flow block,
analyze it once in the current batch and reuse that current-run artifact for
sibling reviews. Do not merge unrelated business flows into one
program-set-sme-core-review.md.
Delivery workspace profile — optional delivery_workspace_profile
describing where new scan output is written in the delivery repo. The current
default is branch_mode: use_or_create_provided with a provided
develop-<person> branch such as develop-leo; create it from origin/main
if it does not exist, but never write directly to main.
Program artifact resolution profile — optional
program_artifact_resolution_profile describing folder patterns, artifact
filenames, and program-name normalization for artifact lookup under the
working root. This profile does not authorize cross-run reuse by itself; use
explicit approved document repo mode when the working root is a cloned repo of
approved all-program scan artifacts.
Approved document repo clone — optional local clone of the delivery /
document repo after the all-program scan artifacts have been reviewed and
merged. Use this only with
artifact_repo_mode: approved_document_repo /
--artifact-repo-mode approved_document_repo. The clone supplies already
approved per-program artifacts for SME flow assembly; it does not authorize
inventing missing programs or bypassing required compact artifacts.
Source inventory profile / source root — optional
source_inventory_profile plus source repo path. The default cache is
<source-root>/outputs/repo-scan/ with program-list.csv and
scan-summary.yaml. Teams may change filenames or cache directory in the
profile; SMEs should not edit these values every run.
Flow definition — the entry point of the chain, declared as one of
seven trigger types (see references/trigger-models.md):
Current-run program analyses for every program in the chain. Preferred
core inputs are compact artifacts from legacy-ibmi-program-analyzer:
<PROGRAM>-program-analysis.md, <PROGRAM>-program-analysis-summary.yaml,
<PROGRAM>-source-index.yaml, <PROGRAM>-routine-index.md,
<PROGRAM>-message-inventory.yaml, <PROGRAM>-routine-logic-details.md, and
<PROGRAM>-routine-logic-details.yaml. Optional sidecars
(<PROGRAM>-file-io-inventory.yaml, <PROGRAM>-field-mutation-matrix.yaml,
<PROGRAM>-sql-inventory.yaml) are required only when present/triggered by
the program tier or when the flow needs native file I/O, persisted field
mutation, or SQLRPGLE evidence.
Run resolution per program — record one row per requested or discovered
program with run_resolution: analyzed_this_run, reused_same_run,
reused_artifact_repo, pending_source, or blocked_missing_source.
reused_artifact_repo is valid only when the manifest
run_profile.artifact_repo_mode is approved_document_repo. Do not mark a
program complete from remote-main or unapproved prior-run artifacts.
Approved inventory with relationships populated for the involved objects
Optional: SME notes on trigger context, BAU rhythm, known error scenarios
Each upstream program-analysis should expose the program-chain readiness
sections and sidecars from legacy-ibmi-program-analyzer v0.2.5 or later:
<PROGRAM>-program-analysis.md, <PROGRAM>-program-analysis-summary.yaml,
<PROGRAM>-source-index.yaml, <PROGRAM>-routine-index.md,
<PROGRAM>-message-inventory.yaml, <PROGRAM>-routine-logic-details.md,
<PROGRAM>-routine-logic-details.yaml, and triggered optional sidecars
(<PROGRAM>-file-io-inventory.yaml, <PROGRAM>-field-mutation-matrix.yaml,
<PROGRAM>-sql-inventory.yaml),
Program Call Map with Call Evidence, Logic Decomposition Ledger,
Routine Logic Details with conditioned calculation blocks, routine-local
field lineage / carriers, and routine-local exception closure,
Key File & Field Logic with source identifiers plus business meanings,
file-io-inventory.yaml, field-mutation-matrix.yaml,
sql-inventory.yaml, External Calls with dynamic-call resolution status,
Validation Logic, Exception Closure Ledger, Routine / Window Data Flow,
Redundancy Candidate Notes, and Open Items / Limitations. If an older
approved program-analysis is used,
the flow must either route back for refresh or record a named SME waiver for
each missing v0.2.4 detail.
Stop and require clarification if:
In orchestrated mode, the trigger / entry point is missing or ambiguous.
In assemble_existing mode, the user-provided program set does not describe
one flow or lacks an entry / ordering hint.
In chain_ready, a required program in the chain lacks approved program
analysis artifacts → route only the missing program to
legacy-ibmi-program-analyzer. In standalone_exploratory, record
missing_program_artifact and continue with a quick-validation draft when
enough compact evidence exists.
In chain_ready, a required call edge, data flow, or error path depends on
an upstream program routine marked blocked or indexed_only with business
state impact → route back to legacy-ibmi-program-analyzer. In
standalone_exploratory, mark the affected flow claim unresolved and record
a downstream-readiness gap.
In chain_ready, inventory sme_review.decision is blocked or
relationships are incomplete → route to legacy-ibmi-inventory. In
standalone_exploratory, record inventory linkage gaps and keep the artifact
draft_exploratory.
Trigger model cannot be identified from the entry point → require SME
clarification (do not guess)
Flow appears to span multiple business modules → narrow scope; one flow
= one business transaction
Output Contract
For current SME-provided program-flow/core-merge requests, produce only:
program-set-sme-core-review.md
Do not produce flow-<FLOW-SLUG>.md for this current workflow. Produce a full
flow artifact only when the user explicitly asks for full transaction-flow
analysis with trigger/context, edges, data flow, persistence, replay, and
capability seeds.
program-set-sme-core-review.md must be reader-first. It starts with
Program Set Reading Summary, Cross-Program Processing Overview,
Calculation Logic, Validation Logic, Exception Handling, and
Message Inventory. The audit/control sections then follow:
Core Completeness Ledger, Sources, Run Profile, and
Source Inventory Cache. The ledger lists every program from the
SME-provided flow, inventory relationship, or discovered call evidence; its
run_resolution; required compact artifacts; missing core evidence; and
whether the next action is same-run reuse, source scan, or missing-source
follow-up. No program may be omitted merely because its artifact is missing.
For multiple SME-provided program flows, produce one program-set review folder
per flow:
The current deterministic builder runs once per flow block. A future wrapper
may batch these calls, but the contract remains one manifest and one SME core
review per flow.
<FLOW-SLUG> is uppercase, hyphen-separated, business-event named (e.g.,
ONUS-AUTH, BATCH-RECON, CHARGEBACK-INTAKE), and stable across the
analysis chain.
Use:
templates/flow.md as the starting structure
templates/sme-core-review.md as the starting structure for the compact
multi-program SME review view
references/output-contract.md for section definitions and required fields
references/trigger-models.md for trigger-specific analysis guidance
references/data-flow-patterns.md for cross-program data passing
references/error-propagation.md for error-chain analysis
Follow:
../../docs/id-conventions.md for stable IDs (FLOW-*, NODE-*, EDGE-*, DATA-*)
../../docs/evidence-and-knowledge-taxonomy.md for evidence strength tagging
../../docs/input-readiness-rubric.md for input readiness scoring
examples/incomplete-flow-negative/ — flow with a missing program-analysis (must route back)
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 the full
field-level rules. The summary below is normative for this skill.
Input
Required: analysis_intent set to standalone_exploratory by default
or chain_ready when downstream-ready handoff is requested; flow_scan_mode
set to orchestrated or assemble_existing; and enough program evidence to
identify at least one entry node and one flow path.
Required for chain_ready: approved 01_inventory/inventory.yaml with
relationships populated for the involved objects; approved program
analysis evidence for every program in the chain; required core compact
artifacts; triggered/claim-specific optional sidecars; and coverage gates
satisfied.
Required for orchestrated: flow definition (entry point + one of seven
trigger models). The analyzer may discover the program set from inventory
relationships and program call summaries, but any missing per-program
artifact must be routed back to legacy-ibmi-program-analyzer.
Required for assemble_existing: user-provided program analysis
directories or artifact list, preferably including
<PROGRAM>-program-analysis.md, <PROGRAM>-program-analysis-summary.yaml,
<PROGRAM>-source-index.yaml, <PROGRAM>-routine-index.md,
<PROGRAM>-message-inventory.yaml, <PROGRAM>-routine-logic-details.md,
and <PROGRAM>-routine-logic-details.yaml for each program,
plus optional <PROGRAM>-file-io-inventory.yaml,
<PROGRAM>-field-mutation-matrix.yaml, and <PROGRAM>-sql-inventory.yaml
when present/triggered or needed by a flow claim.
Missing required artifacts should be filled for only the affected program
when source is available; otherwise record TBD: missing_program_artifact.
Optional: DSPF / PRTF / *MENU definitions (UI-aware flows);
WRKJOBSCDE export (scheduler triggers); trigger-program registration
(DB triggers); SME notes on BAU rhythm and known error scenarios.
Input readiness scoring:
0-5 blocked: trigger model / entry unresolved, provided programs cannot
be scoped to one business transaction, source authorization unresolved, or
a chain_ready request is missing approved inventory / required program
analyses.
6 minimum_pass: enough compact evidence exists to produce a
standalone_exploratory quick-validation flow draft. Missing approvals,
inventory linkage, or sidecars are recorded as downstream-readiness gaps.
7-8 usable: scheduler/job notes, menu/display/report definitions, and
object dependencies are available for most chain edges.
9-10 strong: runtime logs, screen/report samples, SME notes on triggers
and manual workarounds, and known exception cases are also supplied.
Missing runtime samples does not block structural flow analysis; it leaves
runtime ordering and trigger confidence lower.
Readiness checks: trigger model identified unambiguously for
orchestrated; the flow represents one business transaction; every node
has approved program-analysis evidence or an explicit
missing_program_artifact TBD. For chain_ready, Inventory Completeness
Gate must pass and no load-bearing program artifacts may be missing.
Stop conditions: trigger / entry ambiguity; user-provided program set
spans multiple business transactions; any required chain_ready node missing
approved program-analysis evidence needed for flow claims and not repairable
in the current run;
a required call edge, data flow, or error path depends on an upstream
routine marked blocked or indexed_only with business state impact;
inventory sme_review.decision is blocked or relationships are
incomplete; trigger model cannot be identified from the entry point;
flow spans multiple business modules.
Execution
Procedure: see the Workflow section below (12 ordered steps).
Allowed inference: assembling cross-program call edges from upstream
Call Evidence and External Calls rows; classifying call types (CALL /
CALLP / CALLPRC / SBMJOB / remote); deriving branch destinations from
DSPF option tables; stitching cross-program field lineage only through
upstream Key File & Field Logic, Routine / Window Data Flow, field
mutation matrices, visible carrier fields, or SME-confirmed handoffs;
reading scheduler / trigger / API configuration exports as tier-1 evidence.
Forbidden assumptions: calls not visible in upstream Call Evidence
or External Calls rows; dynamic calls whose upstream resolution is
unresolved or needs_sme_review unless a named SME waiver is recorded;
data flow whose parameter semantics require guessing; branch destinations
not in DSPF DDS; trigger conditions without configuration export; scheduler
frequency without WRKJOBSCDE; commit boundaries without code or SME
confirmation; flow-level field lineage not backed by upstream lineage,
source identifier + business meaning pairs, or a visible carrier; persisted
file/field updates absent from upstream mutation matrices; exception
propagation not backed by upstream Validation Logic / Exception Closure
Ledger rows; business rules (these are seeds, never facts).
TBD handling: missing program-analysis or sidecar →
TBD: missing_program_artifact routing only the affected program to
legacy-ibmi-program-analyzer; ambiguous trigger →
TBD: pending_sme_judgment; unnamed business event → stop and request
the name from SME (do not autogenerate from program names). In
standalone_exploratory, unresolved artifacts become downstream-readiness
gaps unless they prevent identifying the flow itself.
Coverage propagation: consume each upstream program's compact artifacts
first: <PROGRAM>-program-analysis-summary.yaml,
<PROGRAM>-source-index.yaml, <PROGRAM>-routine-index.md,
<PROGRAM>-message-inventory.yaml, <PROGRAM>-routine-logic-details.md,
<PROGRAM>-routine-logic-details.yaml,
<PROGRAM>-file-io-inventory.yaml, <PROGRAM>-field-mutation-matrix.yaml,
and <PROGRAM>-sql-inventory.yaml. Use full
<PROGRAM>-program-analysis.md / <PROGRAM>-program-analysis-<OBJ-ID>.md
only for targeted clarification, not as the primary flow aggregation input. The compact
artifacts must expose Analysis Coverage & Scope, Routine Cards / routine
summary, Routine Logic Details sidecar status, front-loaded Validation Logic
references, Deep Read Windows, Call Evidence, Logic Decomposition Ledger,
Key File & Field Logic, file-io-inventory.yaml,
field-mutation-matrix.yaml, sql-inventory.yaml, Exception Closure
Ledger, Routine / Window Data Flow, Redundancy Candidate Notes, and Open
Items / Limitations before using program-level evidence in a flow. If
the requested flow relies on a
routine that was only indexed_only and that routine changes state,
performs external handoff, handles commit/rollback, controls error
outcome, supplies critical field lineage, or mutates persisted fields,
flow analysis is blocked until the routine is deep_read or a named
SME waiver is recorded in review metadata.
Output
Default SME program-flow artifact: program-set-sme-core-review.md.
This artifact starts with Program Set Reading Summary and Cross-Program
Processing Overview, then contains Calculation Logic, Validation Logic,
Exception Handling, and Message Inventory, followed by the audit/control
sections Sources, Core Completeness Ledger, Run Profile, and Source
Inventory Cache.
Full transaction-flow artifact: flow-<FLOW-SLUG>.md, only when the user
explicitly asks for full flow analysis.
Required sections: front-loaded Calculation Logic, Validation Logic,
Exception Handling, Message Inventory, metadata, trigger model & entry point,
transaction call map, nodes, edges, common dependencies, cross-program data
flow, flow replay path, cross-program field lineage, flow persistence matrix,
branch points, UI surfaces (or N/A — non-interactive), error propagation &
commit boundaries, exception propagation chain, capability seeds, SME review
checklist.
Required IDs: mints FLOW-*, NODE-*, EDGE-*, DATA-*,
REPLAY-*, LINEAGE-*, PERSIST-*, EXCHAIN-*, SEED-*,
TBD-*; reuses OBJ-*, EV-*, and program-level BEH-* from
upstream. Flow analysis does not mint BR-*; branch points are
represented by NODE-* / EDGE-* entries and capability questions by
SEED-*.
Handoff status: status: draft_exploratory for standalone quick
validation; status: draft → needs_sme_review → approved or
approved_with_non_blocking_tbd for chain-ready flow work.
blocked_pending_source and blocked_pending_sme halt
module-analyzer. draft_exploratory is not eligible for module/BRD/spec
handoff until rerun or promoted as chain_ready.
Validation
Mechanical: every edge traces to evidence type 1, 2, or 3 (source
statement, config export, or integration contract); every data exchange
traces to a source line; every UI surface traces to a DSPF/PRTF/*MENU
in inventory; every trigger has evidence type 2 plus SME confirmation
of business meaning; every replay step maps to NODE-*, EDGE-*,
DATA-*, PERSIST-*, or EXCHAIN-*; every field lineage is backed
by upstream field lineage or visible carrier fields; every persistence
row is backed by upstream field-mutation-matrix.yaml and, for SQLRPGLE,
sql-inventory.yaml; every node carries
upstream coverage state and any blocking coverage gaps; all required
sections populated.
AI semantic: edges match upstream program-analyses (no invented
calls); branch destinations match DSPF option tables; error
propagation matches each node's program-analysis Exception Closure
Ledger; commit boundaries are evidenced, not assumed; flow evidence is
not taken from
indexed_only or blocked routines when those routines have business
state impact unless a named SME waiver is recorded; capability seeds
are questions, not rule claims.
SME / human approval: SME confirms trigger model, business event
name, node/edge correctness, branch points, UI surfaces (interactive
flows), error propagation realism, commit boundaries, and that seeds
are reasonable questions. Required when a cross-program rule emerges
or when the flow touches money, inventory, compliance, or customer
status.
Blocking conditions: trigger model unresolved; business event name
unnamed by SME; ambiguous module boundary. For chain_ready, also block when
any node lacks approved program-analysis evidence, any edge has no evidence,
required flow evidence depends on blocked coverage or indexed_only
state-impacting routines without a named SME waiver, or SME absence conflicts
with the flow's risk class. For standalone_exploratory, these become
downstream-readiness gaps unless they prevent identifying the flow itself.
Emit a Step Validation Report (see
../legacy-step-contract/templates/step-validation-report.md) with
status pass, pass_with_warnings, or blocked when reporting upward
to the orchestrator.
Workflow
Select Analysis Intent & Flow Scan Mode
Default analysis_intent to standalone_exploratory unless the user asks
for approved/downstream-ready/chain-ready output.
Use analysis_intent: chain_ready only when the flow must feed module,
BRD, spec, or formal downstream handoff.
If the SME provides a program flow/list and asks for core logic merge,
default to flow_scan_mode: core_review_only, normalize the supplied
program list/order, and mark it as SME navigation evidence. Do not produce
flow-<FLOW-SLUG>.md unless the user explicitly asks for full transaction
flow analysis.
If the SME provides multiple named program flow blocks, split them first
and process each flow block independently. Use one review_name,
programs.txt, manifest, review skeleton, and {review_slug} output
folder per flow. The blocks may share the same develop-<person> working
branch and PR.
Load the team/project delivery profile when supplied. If no team profile is
supplied, use templates/delivery-profile.yaml as the editable starting
shape. Use program_artifact_resolution_profile for artifact folder/name
patterns, and use the workspace profile to decide where current-run
program artifacts and program-set reviews are written.
If the SME is working from a locally cloned document/delivery repo that
already contains approved all-program scan artifacts, set artifact repo
mode to approved_document_repo. In that mode, resolve the SME-provided
program list from the local clone and do not route found programs back to
source scanning. Missing or incomplete programs remain explicit ledger gaps
for follow-up. When a fresh source inventory finds a missing program, route
only that program to legacy-ibmi-program-analyzer / program-list batch
and refresh the document repo artifact. When fresh inventory does not find
it, record blocked_missing_source and request SME correction or source.
For the default SME program-flow workflow, run program-evidence first:
route every distinct SME-provided program through
legacy-ibmi-program-list-batch when multiple programs need source
analysis in a Copilot Chat-limited runtime, or directly through
legacy-ibmi-program-analyzer for a single program. Reuse only artifacts
already produced earlier in this same run/batch before program-set
assembly.
Preserve exact program/member identity. The current lending-card default
preserves leading @, so @CU118 and CU118 are distinct programs.
Other departments can override folder patterns, artifact filenames, and
alias rules.
For every SME-provided program, generate or verify current
tier-appropriate program-level artifacts before assembly. Normal,
complex, and large programs require <PROGRAM>-program-analysis.md,
<PROGRAM>-program-analysis-summary.yaml, <PROGRAM>-source-index.yaml,
<PROGRAM>-routine-index.md, <PROGRAM>-message-inventory.yaml,
<PROGRAM>-routine-logic-details.md, and
<PROGRAM>-routine-logic-details.yaml. Retain deep-read batches and
coverage ledgers when needed for stable handoff.
Do not check remote main, clone a remote-main snapshot, or mark programs
as found/not-found on remote main for the default program-flow workflow.
Existing historical artifacts may be compared later in Git review, but they
do not satisfy this run's evidence gate.
Before source scanning any unresolved program, check the source inventory
cache. Default path:
<source-root>/outputs/repo-scan/program-list.csv plus
<source-root>/outputs/repo-scan/scan-summary.yaml. Reuse it only when
the source revision in scan-summary.yaml matches the current clean Git
source HEAD. If missing, stale, or dirty, rerun repo-level
legacy-ibmi-inventory first, then use program-list.csv to locate each
program's source path and tier for targeted program analysis.
On a Windows machine where py -3 and python are both unavailable, a
missing, stale, or dirty inventory cache remains
blocked_pending_inventory_runtime until a fresh cache is supplied or the
separate legacy-ibmi-inventory repo scanner receives a native
PowerShell implementation. The program-set builder and validator fallback
do not silently treat a stale cache as fresh.
Write current-run program analysis output to the delivery repo
working branch named by the user/profile, normally develop-<person>. If
that branch does not exist, create it from origin/main; if it exists,
update it before writing. This working branch is the current run's draft
workspace.
Write program artifacts under
delivery_workspace_profile.program_tier_roots by size tier. Write
program-set-sme-core-review.md under
delivery_workspace_profile.program_set_review_parent/{REVIEW_SLUG}/.
For core_review_only in current-run mode, run the deterministic builder
after every program analysis artifact is complete and before the LLM fills
the review:
Windows:
py -3 .agents\skills\legacy-ibmi-flow-analyzer\scripts\program_set_core_review.py build --review-name "<name>" --programs-file <programs.txt> --working-root <delivery-working-branch-checkout> --source-root <source-repo> --profile <delivery-profile.yaml> --output-dir <delivery-worktree-output-dir> --working-branch <develop-person> --program-first.
If py -3 is unavailable, run the same command again with python
replacing py -3; do not use PowerShell or .cmd.
macOS/Linux:
python3 scripts/build-program-set-core-review.py --review-name "<name>" --programs-file <programs.txt> --working-root <delivery-working-branch-checkout> --source-root <source-repo> --profile <delivery-profile.yaml> --output-dir <delivery-worktree-output-dir> --working-branch <develop-person> --program-first.
For core_review_only from an approved local document repo clone, run the
deterministic builder with approved document repo mode:
Windows:
py -3 .agents\skills\legacy-ibmi-flow-analyzer\scripts\program_set_core_review.py build --review-name "<name>" --programs-file <programs.txt> --working-root <local-document-repo-clone> --profile <delivery-profile.yaml> --output-dir <local-document-repo-clone>\<program_set_review_parent>\<review_slug> --working-branch main --artifact-repo-mode approved_document_repo.
If py -3 is unavailable, run the same command again with python
replacing py -3; do not use PowerShell or .cmd.
macOS/Linux:
python3 scripts/build-program-set-core-review.py --review-name "<name>" --programs-file <programs.txt> --working-root <local-document-repo-clone> --profile <delivery-profile.yaml> --output-dir <local-document-repo-clone>/<program_set_review_parent>/<review_slug> --working-branch main --artifact-repo-mode approved_document_repo.
The builder writes program-set-core-input-manifest.yaml plus a fixed
program-set-sme-core-review.md skeleton. Treat the manifest as the
control input for all subsequent summarization. In program-evidence-first
mode, the manifest records run_resolution as analyzed_this_run,
reused_same_run, pending_source, or blocked_missing_source. In
approved document repo mode, found artifacts use
run_resolution: reused_artifact_repo and
artifact_source: approved_document_repo.
--source-root enables the default <source-root>/outputs/repo-scan
lookup; --inventory-dir may override that cache path for teams with a
different layout. --working-root is the only artifact lookup root for
the default program-flow workflow and for approved document repo reuse.
Use flow_scan_mode: orchestrated when the user explicitly provides a trigger /
entry program and wants the skill to discover, index, and assemble the
whole flow.
Use flow_scan_mode: assemble_existing when the user explicitly provides several
existing program analysis directories and asks to combine them into one
flow.
In all modes, aggregation must prefer compact artifacts:
<PROGRAM>-program-analysis-summary.yaml, <PROGRAM>-source-index.yaml,
<PROGRAM>-routine-index.md, <PROGRAM>-message-inventory.yaml,
<PROGRAM>-routine-logic-details.md,
<PROGRAM>-routine-logic-details.yaml,
<PROGRAM>-file-io-inventory.yaml,
<PROGRAM>-field-mutation-matrix.yaml, and <PROGRAM>-sql-inventory.yaml.
Do not concatenate full <PROGRAM>-program-analysis.md /
<PROGRAM>-program-analysis-*.md files across programs. Open full
Markdown only for targeted human-readable clarification when compact
artifacts are insufficient.
If the user asks to merge multiple program-analysis results or provides a
SME program flow for core SME review, generate
program-set-sme-core-review.md from templates/sme-core-review.md.
Fill Program Set Reading Summary and Cross-Program Processing Overview
first, then Calculation Logic, Validation Logic, Exception Handling, and
Message Inventory as self-contained SME reading surfaces. Put Core
Completeness Ledger, Sources, Run Profile, and Source Inventory Cache
after the reader-first core as audit/control evidence. Do not include
Nodes, Edges, Replay, Persistence, Lineage, UI Surfaces, Capability Seeds,
or SME Checklist in that compact core-review artifact.
After the four core sections are filled, run
Windows:
py -3 .agents\skills\legacy-ibmi-flow-analyzer\scripts\program_set_core_review.py validate --manifest <program-set-core-input-manifest.yaml> --review <program-set-sme-core-review.md>.
If py -3 is unavailable, run the same command again with python
replacing py -3; do not use PowerShell or .cmd.
macOS/Linux:
python3 scripts/validate-program-set-core-review.py --manifest <program-set-core-input-manifest.yaml> --review <program-set-sme-core-review.md>.
A failed validator means the artifact is structurally unsafe to hand to
the SME; fix the review instead of hiding missing programs or changing the
manifest.
For multiple flow blocks, validate each generated
program-set-sme-core-review.md independently. A failure in one flow
blocks that flow's SME handoff but does not require regenerating sibling
flows that already passed validation.
In standalone_exploratory, continue with warning rows when approvals,
inventory linkage, or sidecars are missing; in chain_ready, enforce the
blocking gates.
Identify Trigger Model & Entry Point
Determine which of the seven trigger models applies (see
references/trigger-models.md). If unsure, ask the SME — do not guess.
In assemble_existing mode, accept the user's stated entry/order hint; if
it is missing or contradictory, stop for SME clarification.
Document the trigger object:
batch job → CL program + direct CALL from another program or operator
DB trigger → file name + trigger program + event (insert/update/delete)
scheduler → WRKJOBSCDE entry + frequency (may submit a batch job via SBMJOB)
API / remote → remote-call mechanism + parameter contract
Note: Scheduler-submitted batch jobs form a single flow: the scheduler
entry is the primary trigger, SBMJOB is the submission mechanism, and the
CL/RPG program(s) are the nodes. This is one trigger model, not two.
Assign FLOW-<SLUG>-<NNN> ID.
Capture the business event name the SME uses for this flow
(e.g., "Customer authorization request", not "RPG1 call").
Enumerate Nodes (Programs in the Chain)
List every program the flow touches, in call order.
For each node:
assign NODE-<SLUG>-<NNN> (sequence-numbered)
reference the program's OBJ-* (from inventory) and approved
program-analysis artifact set
record artifact availability:
<PROGRAM>-program-analysis.md, <PROGRAM>-program-analysis-summary.yaml,
<PROGRAM>-source-index.yaml, <PROGRAM>-routine-index.md,
<PROGRAM>-message-inventory.yaml, <PROGRAM>-routine-logic-details.md,
<PROGRAM>-routine-logic-details.yaml,
<PROGRAM>-file-io-inventory.yaml,
<PROGRAM>-field-mutation-matrix.yaml, <PROGRAM>-sql-inventory.yaml,
and optional object-ID-specific human-readable
program-analysis-<OBJ-ID>.md
record the program's role in the flow (entry / orchestrator /
worker / data-access / reporter / exit)
carry forward the program's Analysis Coverage & Scope, Routine
Cards, and Deep Read Windows; record node coverage status as
deep_read, indexed_only, or blocked plus flow readiness
(approved, warning, or blocked)
record blocking coverage gaps when a required call edge, data
exchange, branch, error path, or commit boundary depends on a
routine that is blocked or is indexed_only with business state
impact
If a node has no approved program-analysis evidence, create
TBD-<SLUG>-<NNN>: missing_program_artifact routing only that program
back to legacy-ibmi-program-analyzer. In chain_ready, stop; in
standalone_exploratory, continue only with claims supported by available
compact evidence.
If a node has an approved human-readable analysis but lacks compact
sidecars, fill only the missing sidecars when local source is available;
otherwise create TBD-<SLUG>-<NNN>: missing_program_artifact and avoid
using that program for flow claims that require the missing sidecar.
If a node's required state-impacting routine is only indexed_only,
stop in chain_ready until legacy-ibmi-program-analyzer deep-reads it
or a named SME waiver is recorded in evidence/review/sign-off metadata. In
standalone_exploratory, mark the affected flow claim unresolved and keep
it out of downstream-ready conclusions.
Trace Edges (Calls Between Nodes)
For every call between two nodes, document an edge:
EDGE-<SLUG>-<NNN>: NODE-A → NODE-B
Call type (CALL, CALLP, CALLPRC, EXSR-not-applicable-cross-program,
SBMJOB, remote-call)
Call site (source program + line number) — from the caller's
program-analysis
Via routine/procedure — the internal node from the caller's
Program Call Map where the edge exits the program, when visible
Build the Transaction Call Map (see
references/output-contract.md for format). The map shows
cross-program and cross-boundary structure; detailed call conditions
remain in the edge table.
Classify Common Dependencies
If multiple nodes call the same common program, service program, API,
data queue, message queue, or wrapper, keep every inbound call in the
edge table and summarize the target in Common Dependencies.
Keep common nodes expanded as formal flow nodes when they write
business files, change customer/account/inventory/money state,
decide approve/decline outcomes, control commit/rollback, or call an
external business system.
Fold common nodes in the visual map only when they are technical
utilities such as logging, message formatting, date/time,
delay/wait, or wrapper code. Folding is visual only; evidence and
edge rows remain complete.
Do not infer a Java service boundary from shared usage alone. That is
a modernization decision for module/spec stages with SME review.
Cross-Program Data Flow
For each edge, document what data passes (see
references/data-flow-patterns.md):
parameters passed in the CALL
shared data area (*DTAARA) updates / reads
data queue (*DTAQ) messages
shared file writes/reads (program A writes, program B reads)
temporary work files / spool / IFS files
signal flags (return codes, indicators, error fields)
For each DATA-*, capture carrier, producer, consumer, mechanism,
payload/key fields, direction and timing, state impact, and
evidence. Preserve critical fields as FIELD_NAME (business meaning)
or VARIABLE_NAME (business meaning) [direction] whenever the upstream
program-analysis resolved both identity and meaning. A carrier can be an
EDGE-*, PF/LF, data area, data
queue, message queue, spool, IFS file, DSPF field set, or manual
handoff.
Track object / record / critical-field granularity. Critical fields
include money, account/customer/card/inventory identifiers,
approval/decline/status/posting flags, return/error codes, audit
IDs, and external payload fields. Do not trace every temporary work
variable.
Add short critical trails for business-important records or
messages, such as request -> DTAQ -> program -> log PF -> batch posting.
Assign DATA-<SLUG>-<NNN> for each distinct data exchange.
Tag evidence; cross-reference to source line numbers via EV-*.
Build Flow Replay Path, Field Lineage & Persistence Matrix
Build a Flow Replay Path from trigger to terminal outcome:
trigger, entry parameters, node execution, data handoff, key
decisions, persisted mutations, commit/rollback/output, and final
response or operator-visible state.
Each replay step must reference existing NODE-*, EDGE-*,
DATA-*, PERSIST-*, EXCHAIN-*, or UI-* rows. Do not invent
a replay step that is not supported by an upstream program-analysis
or flow evidence.
Build Cross-Program Field Lineage for critical fields that
cross program boundaries. Stitch program-local lineages through
CALL parameters, shared files, data areas, queues, screens, IFS,
spool, or SME-confirmed manual handoffs. Do not collapse a resolved
source identifier and business meaning into a plain field label.
For program-analyzer v0.2.5 inputs, prefer Routine Logic Details'
conditioned calculation blocks and routine-local field lineage / carrier
rows when they provide the guard-scoped source carrier, intermediate
variable, output/persisted carrier, or lineage/mutation reference for a
cross-program value.
Build the Flow Persistence Matrix by aggregating each
program-analysis file-io-inventory.yaml,
field-mutation-matrix.yaml, and sql-inventory.yaml summary into
transaction-level outcomes:
node and routine that performs the mutation
file and field persisted, updated, deleted, or skipped
upstream field / parameter / carrier that drives it
downstream readers or consumers
commit / rollback / retry impact
evidence and TBDs for missing DDS, mutation source, SQL host-variable
mapping, or rollback
behavior
For read-only flows, explicitly mark the persistence matrix
N/A — read-only flow and cite the upstream analyses proving no
persisted mutations.
Branch Points & Decision Nodes
Identify points where the flow forks (subfile option, F-key,
conditional CALL, trigger event).
For each branch point:
the deciding artifact (which field / which key / which condition)
the alternatives and which target node each leads to
whether unhandled options/keys exist (silently dropped vs. error)
UI Surfaces (interactive flows only)
List every DSPF / PRTF / MENU the user sees during the flow.
For each:
object name + OBJ-*
which node displays it
key fields shown / entered
F-keys handled
branch destination per F-key / option
For non-interactive flows (batch, trigger, scheduler, API), this
section may be N/A — non-interactive flow.
Error Propagation & Exception Chain
For each node, summarise (from its program-analysis Error Handling
section) what happens when each error condition occurs.
Trace propagation: does the error abort the whole flow, roll back to
a checkpoint, log-and-continue, or branch to an error handler?
Build an Exception Propagation Chain from every upstream Error Code
Inventory / Exception Closure Ledger row and Routine Logic Details'
routine-local exception closure row that affects this flow. Each
row must show source node, observed message ID / error code / return
code, error type, output carrier, evidence status, propagation carrier,
caller reaction, skipped/allowed downstream edges, persistence impact,
operator/user visibility, and final flow outcome.
Identify commit boundaries: where does the flow consider work
"committed" vs "rolled back"?
Identify unhandled error windows: nodes where an unhandled
error would crash the entire flow without recovery (create TBD).
See references/error-propagation.md.
Business Capability Seeds
Extract SEED-* candidates (business rule seeds) that this flow
plausibly enforces — without inventing rules.
Each seed is a question for SME review (e.g., "Does the rule
'credit limit must not be exceeded' live in this flow?").
Phrase the candidate and SME question in business terms first:
business event, business object, decision, outcome, control, or
exception. Put program names, node IDs, field names, and file names in
Evidence Basis, not in the candidate statement.
Evidence Basis should reference replay path, cross-program field
lineage, persistence matrix, or exception chain rows when those
rows are the real support for the seed.
The spec-writer skill will resolve seeds into approved rules; the
flow analyzer never approves rules itself.
Note: flow analysis does not mint BR-* IDs. Branch points are
represented by NODE-* / EDGE-* entries; capability seeds use
SEED-* IDs.
Prepare for SME Review
Consolidate all TBDs grouped by blocking status (pending_source /
missing_program_artifact / pending_sme_judgment / non_blocking).
If analysis_intent: standalone_exploratory:
Mark flow as draft_exploratory.
Complete the quick-validation artifact with all visible Calculation
Logic, Validation Logic, Exception Handling, Message Inventory, Nodes,
Edges, and gaps.
Record missing approvals, inventory linkage, sidecars, and indexed-only
state-impacting routines as downstream-readiness gaps.
Do not route to module-analyzer until rerun/promoted as chain_ready.
If analysis_intent: chain_ready and any blocking TBDs exist:
Mark flow as blocked_pending_source (missing program-analysis, missing DSPF, etc.)
or blocked_pending_sme (ambiguous trigger, unclear error handling, missing business context)
Do not complete the full required analysis; provide partial flow stub
Route to legacy-ibmi-program-analyzer or SME assignment process
Do not proceed to module-analyzer
If analysis_intent: chain_ready and no blocking TBDs exist:
Confirm every node's program-analysis is approved.
Generate the flow review checklist (see templates/trigger-checklist.md).
Mark flow as draft and route to SME.
Gate: the flow is ready for module-analyzer when:
all required sections populated
every node has an approved program-analysis
no blocking TBDs (or SME has explicitly waived them)
SME has signed off on the trigger model + business event name +
capability seeds
3d Flow Analysis Done when every in-scope SME program set has a completed
program-set-sme-core-review.md, or every explicitly requested full
transaction flow has an approved flow-*.md
3c Flow Analysis In Progress when one or more in-scope program sets or full
flows are still draft, blocked, or missing
Last artifact path pattern:03_flows/<MODULE-SLUG>/program-set-sme-core-review.md for the current SME
program-flow core-merge workflow, or
03_flows/<MODULE-SLUG>/flow-<FLOW-SLUG>.md for explicitly requested full flow
analysis.
Writes per run:
Overwrite capabilities[<CAP-* from current_focus>] with stage id, the
path of the program-set core review or full flow you just saved,
last_skill: legacy-ibmi-flow-analyzer, and blocking IDs (tbds,
sme_pending for missing program artifacts or, for full flow analysis,
trigger context / commit boundary questions).
Append one history[] entry with note naming the flow
(e.g. "analyzed flow-submit-order").
Never touch current_focus, other capabilities' entries, or past
history[] rows. A re-run on the same flow is allowed; a re-run that
would lower stage_id requires the orchestrator's Rollback Protocol.
Anti-Hallucination Rules
Code is ground truth. See ../../docs/code-as-ground-truth.md. Every
edge in this flow must trace to evidence from authoritative sources
(see references/output-contract.md Evidence Taxonomy). Authoritative
evidence includes:
Integration contracts (MQ queue configs, API gateway definitions, DDM registrations)
SME confirmation (documented procedures, business event naming, SLA agreement)
Shop documentation, prior architecture diagrams, and SME recollection are
navigation aids; they do not become evidence until tier-1 authoritative
confirmation exists. If SME claims a flow path the code does not contain,
both observations are recorded and a TBD blocks the flow until SME reconciles them.
Do NOT invent:
Calls not visible in any program's program-analysis Call Evidence or
External Calls sections
Data flow that requires guessing parameter semantics (use only what
source explicitly shows or SME confirms)
Cross-program field lineage that cannot be stitched through
upstream program-analysis lineage, source identifier + business meaning
pairs, carrier fields, or SME-approved manual handoff
Persistence outcomes not present in upstream file-io-inventory.yaml,
field-mutation-matrix.yaml, sql-inventory.yaml, or SME-confirmed
durable-output evidence
Exception chains not present in upstream Validation Logic /
Exception Closure Ledger rows, return-code checks, message IDs, or
SME-confirmed operational recovery notes
Branch destinations for F-keys / options not visible in DSPF DDS
Trigger conditions for DB triggers without seeing trigger configuration export
Scheduler frequency or submitted command without WRKJOBSCDE evidence
Commit boundaries if neither code nor SME has confirmed them
Business capabilities or rules — these are seeds (questions), not
facts; spec-writer resolves them
Instead:
If a node's program-analysis is missing → TBD: pending_source →
route to legacy-ibmi-program-analyzer
If configuration evidence is missing → TBD: pending_source → request
WRKJOBSCDE export, DDS listing, or trigger configuration
If a trigger model is ambiguous → TBD: pending_sme_judgment
If the SME cannot name the business event → stop and request a name
(do not autogenerate from program names)
If error propagation is unclear → tag needs_sme_review; do not assume
commit / rollback behavior
Evidence minimum (see Evidence Taxonomy):
Every edge must trace to evidence type 1, 2, or 3 (source statement, config export, or integration contract)
Every edge derived from a program call must cite upstream Call Evidence
and carry its resolution status (confirmed_from_code, observed_in_runtime,
sme_confirmed, needs_sme_review, or unresolved)
Every data exchange must trace to source (parameter, data area, queue,
file, screen field, message, IFS file, or SME note)
Every UI surface must trace to a DSPF / PRTF / MENU object in inventory
Every trigger must have evidence type 2 (config export) + evidence type 4 (SME confirmation of business meaning)
SME Review Questions
When full transaction-flow analysis is explicitly requested, the generated
flow-<FLOW-SLUG>.md must include a review checklist covering:
Trigger model correctly identified (batch / menu / subfile /
F-key / trigger / scheduler / API)
Business event name accurately reflects what the business calls this
transaction
All nodes (programs) in scope are correct; no missing programs and
no extra programs
All edges (calls) reflect actual production behavior
Cross-program data flow captures carriers, producers, consumers,
timing, state impacts, and no undocumented shared files or data areas
Flow Replay Path can be followed from trigger to final outcome
Cross-program field lineage preserves critical source, carrier,
mutation, and output fields
Required flow/program-set prompts and artifact discovery to prefer
<PROGRAM>-<artifact> filenames while retaining legacy bare-name fallback
for existing runs.
v0.2.11 (2026-07-10): Installed-skill Windows router
Added a skill-local Windows launcher so synced .agents, .claude,
.codex, and .opencode installs do not depend on a repository-root
launcher.
Prohibited py -3 ... || python ... fallback chains under Windows
PowerShell 5.1.
Standardized Cline commands as direct py -3 <script.py> ... calls with a
manual python <script.py> ... fallback only when py -3 is unavailable.
Removed PowerShell and .cmd launchers from the documented Cline
execution path.
v0.2.10 (2026-07-10): Python-first Windows PowerShell fallback
Added native Windows PowerShell 5.1 program-set builder and validator for
Cline environments without Python or PyYAML.
Standardized Windows routing as py -3, then python, then native
PowerShell while preserving manifest and validator failure semantics.
v0.2.9 (2026-07-08): Approved document repo reuse for SME flow assembly
Added explicit approved document repo mode for the post all-program-scan
workflow: SME clones the document/delivery repo locally, provides a program
flow/list, and assembles program-set-sme-core-review.md from approved
existing per-program artifacts
Manifest records found programs as run_resolution: reused_artifact_repo
and artifact_source: approved_document_repo
Validator accepts reused_artifact_repo only when
run_profile.artifact_repo_mode: approved_document_repo
v0.2.8 (2026-06-26): No cross-run reuse for SME program-flow core review
Default program-flow workflow now analyzes every distinct SME-provided
program for the current run before program-set assembly
Reuses only artifacts produced earlier in the same run/batch for repeated
programs
Builder manifest records run_resolution and run_profile.cross_run_reuse: false
program-set-sme-core-review.md must be self-contained for Calculation
Logic, Validation Logic, Exception Handling, and Message Inventory
Added resume guidance so a new session can continue from run state,
manifest, current-run artifact folders, and validator results
v0.2.7 (2026-06-18): Multiple SME program-flow block guidance
Clarified that one SME request may contain multiple named program-flow
blocks, but each block produces its own {review_slug} folder,
program-set-core-input-manifest.yaml, and
program-set-sme-core-review.md
Documented that multiple flows can share the same source inventory cache,
delivery working branch, and PR while still validating each review
independently
v0.2.6 (2026-06-18): Source inventory cache gate for program-set runs
Added source_inventory_profile support with default
<source-root>/outputs/repo-scan/program-list.csv and
<source-root>/outputs/repo-scan/scan-summary.yaml
Builder records source_inventory freshness/action in
program-set-core-input-manifest.yaml and renders a Source Inventory
Cache section in program-set-sme-core-review.md
Inventory cache is reused only when the recorded source revision matches
the current clean Git source HEAD; missing, stale, or dirty source states
route to repo-level legacy-ibmi-inventory before targeted program scan
v0.2.5 (2026-06-18): Program-set builder and validator
Added scripts/program_set_core_review.py under the canonical flow
analyzer skill plus root wrappers:
scripts/build-program-set-core-review.py and
scripts/validate-program-set-core-review.py
Builder creates program-set-core-input-manifest.yaml and a deterministic
program-set-sme-core-review.md skeleton from the current-run artifact
root
Builder reads program artifacts from the delivery working branch via
--working-root
Validator checks required core sections, per-program coverage in Sources
and Core Completeness Ledger, legal run_resolution values, and blocks
accidental full-flow sections in compact SME reviews
v0.2.4 (2026-06-17): SME program-flow intake and configurable profiles
Added configurable delivery profile values for delivery repo, module roots,
program folder patterns, workspace roots, and source inventory cache instead
of assuming every department uses the same repository layout; the
lending-card default preserves leading @ as part of program identity
Earlier central-reuse experimentation is superseded by v0.2.8's no
cross-run reuse default for program-flow core review
Required compact SME core reviews to include a Core Completeness Ledger so
every requested/discovered program is accounted for even when an artifact is
missing
Scoped SME-provided program-flow intake to
program-set-sme-core-review.md by default; full flow-<FLOW-SLUG>.md
generation is reserved for explicit full transaction-flow analysis
Added templates/delivery-profile.yaml as the portable team configuration
shape for program resolution, develop-* delivery working branches,
tier-specific program artifact roots, and cross-tier program-set review
output
Required flow aggregation to prefer file-io-inventory.yaml,
field-mutation-matrix.yaml, and sql-inventory.yaml alongside the
existing compact program artifacts
Updated node artifact availability, persistence matrix evidence, and
missing-artifact routing for File I/O dense and SQLRPGLE programs
Required flow analysis to consume Call Evidence, File I/O Purpose,
dynamic-call resolution status, Validation Logic, Routine / Window
Data Flow, and Open Items / Limitations
Required flow field, lineage, persistence, and exception tables to
preserve source identifiers with business meanings where upstream
program-analysis resolved them
Required flow analysis to consume Routine Logic Details, routine-local
field lineage / carrier rows, and routine-local exception closure rows
when building Cross-Program Field Lineage and Exception Propagation Chain
Updated older-program-analysis waiver language from v0.2.1 detail gaps to
v0.2.4 detail gaps
