@chenmk/superflow 0.2.4 → 0.3.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (61) hide show
  1. package/README.md +4 -1
  2. package/README.zh-CN.md +1 -0
  3. package/assets/manifest.json +3 -1
  4. package/assets/prompts/superflow-clarify.md +6 -0
  5. package/assets/prompts/superflow-pipeline.md +8 -0
  6. package/assets/scripts/superflow-archive-command-hook.sh +80 -0
  7. package/assets/scripts/superflow-delivery-check.sh +23 -4
  8. package/assets/skills/superflow-archive/SKILL.md +5 -0
  9. package/assets/skills/superflow-clarify/SKILL.md +26 -7
  10. package/assets/skills/superflow-design/SKILL.md +46 -5
  11. package/assets/skills/superflow-docs/SKILL.md +27 -0
  12. package/assets/skills/superflow-implement/SKILL.md +17 -0
  13. package/assets/skills/superflow-pipeline/SKILL.md +47 -0
  14. package/assets/skills/superflow-pipeline/references/api-design-template.md +11 -0
  15. package/assets/skills/superflow-pipeline/references/batch-prompt-template.md +44 -0
  16. package/assets/skills/superflow-pipeline/references/implementation-prompt-template.md +48 -0
  17. package/assets/skills/superflow-pipeline/references/money-precision-algorithms.md +85 -0
  18. package/assets/skills/superflow-pipeline/references/quality-gate.md +3 -0
  19. package/assets/skills/superflow-pipeline/references/reviewer-checklist.md +17 -1
  20. package/assets/skills/superflow-pipeline/references/superpower-technical-design-template.md +90 -0
  21. package/assets/skills/superflow-pipeline/references/test-execution-template.md +26 -0
  22. package/assets/skills/superflow-pipeline/scripts/superflow-guard.sh +231 -0
  23. package/assets/skills/superflow-verify/SKILL.md +23 -1
  24. package/assets/skills-en/superflow-archive/SKILL.md +6 -0
  25. package/assets/skills-en/superflow-clarify/SKILL.md +26 -7
  26. package/assets/skills-en/superflow-design/SKILL.md +45 -4
  27. package/assets/skills-en/superflow-docs/SKILL.md +27 -0
  28. package/assets/skills-en/superflow-implement/SKILL.md +10 -0
  29. package/assets/skills-en/superflow-pipeline/SKILL.md +38 -0
  30. package/assets/skills-en/superflow-pipeline/references/api-design-template.md +13 -0
  31. package/assets/skills-en/superflow-pipeline/references/batch-prompt-template.md +28 -0
  32. package/assets/skills-en/superflow-pipeline/references/implementation-prompt-template.md +33 -0
  33. package/assets/skills-en/superflow-pipeline/references/money-precision-algorithms.md +96 -0
  34. package/assets/skills-en/superflow-pipeline/references/quality-gate.md +26 -0
  35. package/assets/skills-en/superflow-pipeline/references/reviewer-checklist.md +15 -0
  36. package/assets/skills-en/superflow-pipeline/references/superpower-technical-design-template.md +95 -0
  37. package/assets/skills-en/superflow-pipeline/references/test-execution-template.md +30 -0
  38. package/assets/skills-en/superflow-pipeline/scripts/superflow-guard.sh +231 -0
  39. package/assets/skills-en/superflow-verify/SKILL.md +24 -1
  40. package/dist/app/commands/archive.js +7 -3
  41. package/dist/app/commands/archive.js.map +1 -1
  42. package/dist/app/commands/change-guard.js +27 -0
  43. package/dist/app/commands/change-guard.js.map +1 -0
  44. package/dist/app/commands/design.js +2 -20
  45. package/dist/app/commands/design.js.map +1 -1
  46. package/dist/app/commands/init.js +1 -5
  47. package/dist/app/commands/init.js.map +1 -1
  48. package/dist/app/commands/update.js +2 -5
  49. package/dist/app/commands/update.js.map +1 -1
  50. package/dist/app/commands/verify.js +7 -3
  51. package/dist/app/commands/verify.js.map +1 -1
  52. package/dist/domains/config/context.js +1 -5
  53. package/dist/domains/config/context.js.map +1 -1
  54. package/dist/domains/deps.js +1 -5
  55. package/dist/domains/deps.js.map +1 -1
  56. package/dist/domains/hook.js +1 -0
  57. package/dist/domains/hook.js.map +1 -1
  58. package/dist/platform/assets.js +9 -0
  59. package/dist/platform/assets.js.map +1 -0
  60. package/package.json +3 -2
  61. package/scripts/release-notes-check.js +102 -0
@@ -14,6 +14,10 @@ do not mark archived until the user confirms final closeout.
14
14
  - `phase: archive`.
15
15
  - `verify_result: pass`.
16
16
  - `verification_report` points to an existing report.
17
+ - Every checkbox in `tasks.md` is checked.
18
+ - The verification report contains `Verification Result: PASS` and
19
+ `Archive Readiness: PASS`.
20
+ - `branch_status: handled` and `verified_at` are recorded.
17
21
  - `superflow-state.sh recover <change-dir>` gives a coherent archive recovery action.
18
22
 
19
23
  ## Procedure
@@ -42,6 +46,8 @@ do not mark archived until the user confirms final closeout.
42
46
  the user explicitly accepts the boundary.
43
47
  - Do not use archive to hide missing API, DB, hook, SQL, test-report, or real
44
48
  integration evidence.
49
+ - Never call `openspec archive` directly to bypass the lifecycle gate. The
50
+ installed archive-command hook blocks that path when native hooks are active.
45
51
  - In a standard `openspec/changes/<name>` layout, archive must use OpenSpec
46
52
  archive semantics so delta specs are merged and the change is moved to
47
53
  `openspec/changes/archive/`. State-only archive is only a fallback for
@@ -50,6 +50,15 @@ any full SDD document or implementation prompt is generated.
50
50
  external input. If semantics are unclear, mark the feature blocked instead of
51
51
  inventing a fallback.
52
52
  - Process one feature at a time, in product-document order.
53
+ - When the pipeline selected embedded deep-clarification mode, investigate facts
54
+ from the repository, source documents, configuration, tests, logs, and tools
55
+ before asking. Ask the user only about an unresolved decision that needs owner
56
+ intent; do not ask for facts that can be found locally.
57
+ - In embedded deep-clarification mode, ask exactly one decision question at a
58
+ time. Include the recommended answer and its important consequence, wait for
59
+ the answer, then record the result before asking the next question. Do not
60
+ create SDD artifacts or implementation prompts while a current-feature
61
+ decision remains pending.
53
62
  - For long PRDs, Lark/Feishu exports, screenshots, or mixed product docs, do
54
63
  not read the whole document and generate all tasks in one pass. First create
55
64
  a source index, then process one bounded section or feature at a time. Each
@@ -85,6 +94,10 @@ Create or update these files in the active OpenSpec change directory:
85
94
  - Business semantic evidence when status/enum/sync/default behavior is involved:
86
95
  source-of-truth field, enum mapping, consumer interpretation, business basis,
87
96
  explicitly rejected fallback/default choices, and unresolved owner questions.
97
+ - Deep-clarification decision evidence when this mode is active: each question,
98
+ recommendation, user decision, consequence, and the resulting
99
+ `confirmed`/`candidate`/`pending`/`rejected` entry in
100
+ `.sdd/handoff/brainstorm-summary.md`.
88
101
 
89
102
  Use the detailed workflow in:
90
103
 
@@ -127,14 +140,20 @@ Use the detailed workflow in:
127
140
  If a proposed default, fallback, old-value retention, or null conversion is
128
141
  not explicitly approved by the requirement or owner, mark it rejected and do
129
142
  not include it in the design.
130
- 11. Ask only current-feature clarification questions.
131
- 12. Wait for programmer confirmation.
132
- 13. Draft a rough API contract: path, action, core request, core response. Field types can be approximate.
133
- 14. Wait for programmer confirmation of the rough API.
134
- 15. Produce a precise API section with field names, types, requiredness, defaults, enums, errors, examples, and curl.
135
- 16. Mark the feature as `frozen` only when understanding, rough API, precise API,
143
+ 11. If embedded deep-clarification mode is active, resolve one decision at a
144
+ time. For each question, state the recommended answer and consequence, wait
145
+ for the programmer response, then update
146
+ `.sdd/handoff/brainstorm-summary.md` under `confirmed`, `candidate`,
147
+ `pending`, or `rejected`. Stop asking once only repository-verifiable facts
148
+ remain.
149
+ 12. Ask only any remaining current-feature clarification questions.
150
+ 13. Wait for programmer confirmation.
151
+ 14. Draft a rough API contract: path, action, core request, core response. Field types can be approximate.
152
+ 15. Wait for programmer confirmation of the rough API.
153
+ 16. Produce a precise API section with field names, types, requiredness, defaults, enums, errors, examples, and curl.
154
+ 17. Mark the feature as `frozen` only when understanding, rough API, precise API,
136
155
  and validation data needs are confirmed or explicitly blocked.
137
- 17. Only after the current feature is frozen or blocked, move to the next
156
+ 18. Only after the current feature is frozen or blocked, move to the next
138
157
  indexed feature. Do not batch-confirm multiple features from memory or a
139
158
  compressed chat summary.
140
159
 
@@ -54,15 +54,56 @@ Superpowers own source-level HOW without weakening OpenSpec/SDD contracts.
54
54
  adapter/protocol translator/device gateway into a business-entry
55
55
  orchestrator, stop and return to `$superflow-docs` unless the OpenSpec/SDD
56
56
  contract explicitly grants that ownership with approval evidence.
57
- 6. For field/status/enum/sync changes, include `Field And Status Reverse
57
+ 6. For third-party platforms/tools, SDKs, MQ/Kafka, callbacks, payment
58
+ gateways, cloud services, or other external integrations, include
59
+ `External Integration Configuration And Deployment Contract`. Inventory
60
+ every endpoint, app/tenant/project ID, Topic, Tag, Consumer Group,
61
+ namespace, webhook, ACL/role, feature switch, timeout, and credential
62
+ reference. Prove the source and injection/provisioning method for local,
63
+ test, and production environments, plus owner, creation timing, readiness
64
+ check, rollback, and unresolved blockers. Runtime values and server-side
65
+ resources must not exist only as hard-coded annotations/constants. Test
66
+ auto-creation or pre-existing resources must not be treated as proof that
67
+ production will provision them. Missing production provisioning evidence
68
+ blocks design completion.
69
+ 7. For concurrent requests, batch issue/activation/renewal, duplicate callbacks,
70
+ duplicate consumption, or repeated external delivery, include
71
+ `Concurrency And Idempotency Ownership`. Use a stable business idempotency
72
+ key and an application-layer atomic claim in a short transaction; persist
73
+ `PENDING` before the external call, release database locks before calling
74
+ externally, model success/failure/uncertain states, and reuse the original
75
+ business code on retry. A unique index is not the default and may only be an
76
+ optional fallback with proven natural uniqueness, historical cleanup,
77
+ NULL/soft-delete semantics, and a conflict contract. Check-then-insert,
78
+ process-local locks, and random IDs alone are blocked.
79
+ 8. For field/status/enum/sync changes, include `Field And Status Reverse
58
80
  Impact` and prove writers, readers, filters, derived sync paths, consumers,
59
81
  and tests. Direct setter-only design is blocked.
60
- 7. Record state:
82
+ 9. For amount, fee, discount, deduction, refund, sharing, payment, invoice,
83
+ balance, electricity fee, service fee, package settlement, proration,
84
+ allocation, reconciliation, or financial display changes,
85
+ include `Money Precision Boundary`. Prove calculation-state fields,
86
+ settlement/display-state fields, rounding boundary, scale, rounding mode,
87
+ allocation/reconciliation rule, and tests covering half-cent or residual
88
+ cases. If the design rounds money before later slicing, proration,
89
+ aggregation, discounting, or allocation, stop and redesign the calculation
90
+ boundary. When an additive identity has an authoritative total, the design
91
+ must calculate only N-1 components independently and derive the final
92
+ complement as `authoritative total - sum(other components)`. Independently
93
+ calculating and rounding all components before rebuilding the total is
94
+ blocked. A claimed exception must name the actual source of truth and owner
95
+ evidence. Read
96
+ `../superflow-pipeline/references/money-precision-algorithms.md` and also
97
+ freeze the exact representation, currency/minor-unit boundary, rounding
98
+ level and policy source, deterministic residual strategy with stable
99
+ tie-breaker, and positive/zero/negative evidence. For FX changes, freeze the
100
+ directional rate metadata and one canonical conversion path.
101
+ 10. Record state:
61
102
  ```bash
62
103
  ../superflow-pipeline/scripts/superflow-state.sh set <change-dir> technical_design docs/superpowers/specs/YYYY-MM-DD-<change-id>-technical-design.md
63
104
  ../superflow-pipeline/scripts/superflow-state.sh set <change-dir> design_doc design.md
64
105
  ```
65
- 8. Update `design.md` with `Superpowers Technical Design Handoff`, update
106
+ 11. Update `design.md` with `Superpowers Technical Design Handoff`, update
66
107
  `sdd-quality-gate.md` with the technical design path and hash, then run
67
108
  `../superflow-pipeline/scripts/superflow-handoff.sh <change-dir> --refresh`.
68
109
  If the guard reports a stale or missing hash, record the printed hash in
@@ -70,7 +111,7 @@ Superpowers own source-level HOW without weakening OpenSpec/SDD contracts.
70
111
  `--refresh` again. Do not set `design_doc` to the Superpowers document;
71
112
  `design_doc` is the OpenSpec/SDD contract design, while `technical_design`
72
113
  is the Superpowers HOW document.
73
- 9. Run:
114
+ 12. Run:
74
115
  ```bash
75
116
  ../superflow-pipeline/scripts/superflow-yaml-validate.sh <change-dir>
76
117
  ../superflow-pipeline/scripts/superflow-guard.sh <change-dir> design --apply
@@ -76,6 +76,28 @@ For complex requirements, verify these before writing docs:
76
76
  If the docs only prove that a value exists, is non-null, or lets the request
77
77
  succeed, stop and return to `$superflow-clarify`. If fallback/default/keep-old-value
78
78
  behavior is proposed without explicit requirement or owner approval, stop.
79
+ - For third-party platforms/tools, SDKs, MQ/Kafka, callbacks, payment gateways,
80
+ cloud services, or other external integrations, `api.md` and
81
+ `sdd-quality-gate.md` must include `External Integration Configuration And
82
+ Deployment Contract`. Inventory endpoints, app/tenant/project IDs,
83
+ Topic/Tag/Consumer Group, namespace, webhook, ACL/role, feature switches,
84
+ timeout, and credential references across local, test, and production
85
+ environments. Record injection/provisioning method, runtime owner,
86
+ provisioning owner/time, readiness evidence, rollback, secret handling, and
87
+ blockers. Environment-dependent values must not be hard-coded; test
88
+ auto-creation or existing test resources do not prove production readiness.
89
+ Missing production provisioning evidence blocks document freeze.
90
+ - For amount, fee, discount, deduction, refund, sharing, payment, invoice,
91
+ balance, electricity/service fee, package settlement, proration, allocation,
92
+ reconciliation, or financial display changes, `sdd-quality-gate.md` must
93
+ declare `Money Precision Boundary` as blocking. `tests.md` and the
94
+ `test-report.md` skeleton must reserve half-cent, residual, or multi-detail
95
+ cases and reconciliation evidence. Do not freeze docs that allow intermediate
96
+ money to be rounded before later slicing, aggregation, discounting, or
97
+ allocation. For additive identities, docs must name the authoritative total,
98
+ independently calculated N-1 components, and the final complement formula:
99
+ `authoritative total - sum(other components)`. Do not allow every component
100
+ to be calculated and rounded independently before rebuilding the total.
79
101
  - For low-freedom implementation handoff, docs include the five hard gates:
80
102
  field semantic contract, write-through persistence closure, real-entry call
81
103
  chain, no-fallback/no-guessing boundary, and pre-coding agent self-check. If
@@ -293,6 +315,11 @@ Use references only as needed:
293
315
  `RED failure evidence`, `GREEN pass evidence`, `interface automation evidence`, `DB evidence`,
294
316
  `log evidence`, `manual/blocking cases`, and `Partially verified boundary`.
295
317
  Each section must reference concrete `tests.md` case IDs.
318
+ - For money-related changes, `test-report.md` must also include `Money Precision
319
+ Boundary` evidence: original calculation inputs and precision, actual rounding
320
+ boundary and mode, half-cent/residual/multi-detail cases, and reconciliation
321
+ of original, discount, actual/refund, and allocated totals. Evidence must show
322
+ which total was authoritative and which complement was derived by subtraction.
296
323
  - `sdd-quality-gate.md` must include blocking checks for:
297
324
  executable tests frozen, RED/GREEN expectations present, interface automation
298
325
  commands present, DB/log assertions present, and mock/skipped cases excluded
@@ -60,6 +60,14 @@ Every implementation prompt must include these sections:
60
60
  DTO, converter, SQL, scheduled job, MQ, callback, and sibling repo references.
61
61
  - Field semantic contract: source field, source meaning, target field, target
62
62
  meaning, equivalence, evidence anchor, forbidden usage, and owner for unknowns.
63
+ - Money precision boundary when monetary behavior changes: calculation-state
64
+ source and intermediate precision, settlement/display boundary, scale and
65
+ rounding mode, forbidden early rounding, deterministic residual allocation,
66
+ authoritative total, complement derivation as `authoritative total -
67
+ sum(other components)`, and half-cent or multi-detail reconciliation evidence.
68
+ It also inherits `money-precision-algorithms.md`: exact representation,
69
+ currency/provider minor units, rounding policy source, stable residual
70
+ tie-breaker, signed-value evidence, and conditional FX metadata.
63
71
  - Write-loop verification: business action, setter/assignment, converter, mapper
64
72
  insert/update, DB column, downstream reader, consumer entry, SQL check, test
65
73
  case.
@@ -99,6 +107,8 @@ Block prompt generation or execution when any of these are missing:
99
107
  - handoff files and matching hash;
100
108
  - technical_design for full workflow;
101
109
  - field/status reverse impact matrix when risky fields or states are touched;
110
+ - money precision boundary when monetary calculation, settlement, allocation,
111
+ reconciliation, or financial display is touched;
102
112
  - architecture boundary and call-direction matrix for cross-service or external
103
113
  flows;
104
114
  - executable tests.md commands for L3/L4 or real-entry requirements;
@@ -37,6 +37,11 @@ hotfix:
37
37
 
38
38
  ## Routing
39
39
 
40
+ - When the user explicitly asks for rigorous questioning, or a bounded feature
41
+ has unresolved owner decisions, mutually exclusive approaches, an unclear
42
+ source of truth, unclear acceptance behavior, or cross-system responsibility,
43
+ use `superflow-clarify` in embedded deep-clarification mode. This is a mode
44
+ inside the normal clarify phase, not a separate command the user must remember.
40
45
  - New or unclear requirement: use superflow-clarify.
41
46
  - Confirmed requirement needing full docs: use openspec-propose, then
42
47
  superflow-docs.
@@ -49,6 +54,29 @@ hotfix:
49
54
  - Urgent small behavior repair: use superflow-hotfix, unless API, DB, cross-repo,
50
55
  status, SQL, SDK, or real-entry gates are involved.
51
56
 
57
+ ## Embedded Deep Clarification
58
+
59
+ When the router selects embedded deep clarification, follow these rules inside
60
+ `superflow-clarify`:
61
+
62
+ - Do not activate it for a clear, bounded hotfix or tweak whose intended behavior
63
+ and acceptance result are already evidenced. Do not turn routine work into an
64
+ interview.
65
+ - Establish facts from the requirement, repository, configuration, tests, logs,
66
+ or available tools before asking the user. Ask only for a decision the user or
67
+ owner must make.
68
+ - Ask exactly one decision question at a time. State the recommended option and
69
+ its important consequence, then wait for the answer before the next question.
70
+ - After each answer, update `.sdd/handoff/brainstorm-summary.md` with
71
+ `confirmed`, `candidate`, `pending`, and `rejected` entries. Do not rely on
72
+ chat memory for a resolved decision.
73
+ - Do not create or change proposal, spec, design, tasks, tests, or implementation
74
+ artifacts until all decisions that affect the current feature are confirmed or
75
+ explicitly marked blocked.
76
+ - Exit deep clarification once only repository-verifiable facts remain and the
77
+ user confirms the current feature understanding. Continue through the existing
78
+ clarify workflow without invoking an external skill dependency.
79
+
52
80
  ## Required Artifacts
53
81
 
54
82
  For a full change, the docs phase must create or refresh:
@@ -120,6 +148,15 @@ The workflow must preserve these gates:
120
148
  removal rules.
121
149
  - Field/status reverse impact analysis for IDs, statuses, enums, payment,
122
150
  online/offline, sync markers, external state, or persisted behavior.
151
+ - Money Precision Boundary for monetary calculation, settlement, allocation,
152
+ reconciliation, or financial display. Preserve calculation-state precision
153
+ until the confirmed final boundary, forbid early rounding, define scale and
154
+ rounding mode plus deterministic residual handling, and require half-cent or
155
+ multi-detail reconciliation evidence across design, prompts, review, tests,
156
+ and verify. When an additive identity has an authoritative total, calculate
157
+ only N-1 components independently and derive the final complement as
158
+ `authoritative total - sum(other components)`; never round every component
159
+ independently and rebuild the authoritative total.
123
160
  - Architecture boundary and call-direction analysis for multi-repo, service,
124
161
  SDK, MQ, scheduler, callback, gateway, adapter, device, or third-party flows.
125
162
  - SQL risk review before implementation for DB changes.
@@ -169,3 +206,4 @@ Use the reference files in this skill directory as templates:
169
206
  - references/dirty-worktree.md
170
207
  - references/subagent-progress.md
171
208
  - references/superpower-technical-design-template.md
209
+ - references/money-precision-algorithms.md
@@ -11,6 +11,14 @@ Use api.md as the canonical contract before implementation.
11
11
  - Frontend mapping: screen field, API field, DTO/VO field, DB field, test case.
12
12
  - External dependency contract: SDK version, topic/path, parameter source,
13
13
  sample request/response, timeout and retry behavior.
14
+ - External Integration Configuration And Deployment Contract for third-party
15
+ platforms/tools, SDKs, MQ/Kafka, callbacks, payment providers, cloud
16
+ services, or other external integrations. Inventory endpoint,
17
+ app/tenant/project IDs, Topic/Tag/Consumer Group, namespace, webhook,
18
+ ACL/role, feature switches, timeout, and credential references for local,
19
+ test, and production; record injection/provisioning method, runtime owner,
20
+ provisioning owner/time, readiness evidence, rollback, secret handling, and
21
+ blockers.
14
22
  - File contract when upload/download/export is involved: Content-Type,
15
23
  Content-Disposition, filename rule, body format, frontend detection rule.
16
24
  - Compatibility and migration notes.
@@ -20,3 +28,8 @@ Use api.md as the canonical contract before implementation.
20
28
 
21
29
  Do not allow design, tasks, tests, or prompts to proceed if API changes are only
22
30
  summarized as core fields, same as existing endpoint, or to be added later.
31
+ Environment-dependent external configuration and server-side resources must
32
+ not exist only as hard-coded annotations, constants, or business code. Test
33
+ auto-creation, existing historical resources, or successful local startup do
34
+ not prove production readiness; missing production provisioning evidence
35
+ blocks release.
@@ -24,6 +24,32 @@ review-checklist.md, sdd-quality-gate.md, test-report.md, and technical design.
24
24
  - Do not change forbidden files or root aggregate docs from a worker batch.
25
25
  - Use TDD: record RED failure before GREEN success.
26
26
  - Do not mark mock-only or unit-only evidence as real integration.
27
+ - For third-party platforms/tools, SDKs, MQ/Kafka, callbacks, payment
28
+ providers, cloud services, or other external integrations, inherit
29
+ `External Integration Configuration And Deployment Contract`. Cover
30
+ local/test/production sources and provisioning, injection/creation,
31
+ runtime/provisioning owners and timing, readiness evidence, rollback, secret
32
+ handling, and blockers. Do not hard-code environment-dependent values or
33
+ infer production readiness from test auto-creation.
34
+ - For concurrent requests, batch issue/activation/renewal, duplicate callbacks
35
+ or consumption, and repeated external delivery, inherit `Concurrency And
36
+ Idempotency Ownership`. Define a stable business idempotency key,
37
+ application-layer atomic claim owner, short transaction boundary,
38
+ PENDING/SUCCESS/FAILED state machine, retry reuse of the original business
39
+ code, external-call boundary, and uncertain-result reconciliation. Release
40
+ database locks before external calls. A unique index is not the default and
41
+ may only be an optional fallback with explicit natural uniqueness,
42
+ historical cleanup, NULL/soft-delete behavior, and conflict handling.
43
+ - For monetary changes, inherit `Money Precision Boundary`, retain
44
+ calculation-state precision until the confirmed settlement/display boundary,
45
+ record scale and rounding mode, forbid early rounding before slicing or
46
+ allocation, and define deterministic residual handling. For additive
47
+ identities, calculate only N-1 components independently and derive the final
48
+ complement as `authoritative total - sum(other components)`; do not calculate
49
+ and round every component independently.
50
+ - Inherit `money-precision-algorithms.md`: exact decimal construction, currency
51
+ and provider minor units, rounding policy source, deterministic residual
52
+ tie-breaker, signed-value tests, and conditional FX conversion metadata.
27
53
 
28
54
  ## Implementation Steps
29
55
 
@@ -44,5 +70,7 @@ log check, and hook commands.
44
70
  - Real-entry evidence
45
71
  - DB/log evidence
46
72
  - Hook evidence
73
+ - Money precision evidence: half-cent/residual/multi-detail cases and
74
+ reconciliation of original, discount, actual, and allocated totals
47
75
  - Deviations
48
76
  - Blockers
@@ -23,6 +23,35 @@ Superpowers technical design is canonical for source-level HOW and execution.
23
23
  - Field semantic contract checked.
24
24
  - Write-loop checked.
25
25
  - No-fallback boundary confirmed.
26
+ - External Integration Configuration And Deployment Contract copied from the
27
+ API, quality gate, and technical design for third-party platforms/tools,
28
+ SDKs, MQ/Kafka, callbacks, payment providers, cloud services, or other
29
+ external integrations. Cover local/test/production sources and provisioning,
30
+ injection/creation, runtime/provisioning owners and timing, readiness
31
+ evidence, rollback, secret handling, and blockers. Do not hard-code
32
+ environment-dependent external values or treat test auto-creation as
33
+ production readiness evidence.
34
+ - Concurrency And Idempotency Ownership copied from the technical design for
35
+ concurrent requests, batch issue/activation/renewal, duplicate callbacks or
36
+ consumption, and repeated external delivery. Define a stable business
37
+ idempotency key, application-layer atomic claim owner, short transaction
38
+ boundary, PENDING/SUCCESS/FAILED state machine, retry reuse of the original
39
+ business code, external-call boundary, and uncertain-result reconciliation.
40
+ Release database locks before external calls. A unique index is not the
41
+ default; it is only an optional fallback with explicit natural uniqueness,
42
+ historical cleanup, NULL/soft-delete behavior, and conflict handling.
43
+ - Money Precision Boundary copied from the technical design when monetary
44
+ behavior changes. Keep calculation-state precision until the confirmed
45
+ settlement/display boundary, record scale and rounding mode, forbid early
46
+ rounding, and define deterministic residual allocation. For additive
47
+ identities, name the authoritative total, calculate only N-1 components
48
+ independently, and derive the complement as `authoritative total - sum(other
49
+ components)`. Do not calculate and round every component independently.
50
+ - Read `money-precision-algorithms.md`; inherit exact decimal construction,
51
+ currency and provider minor units, rounding level/mode/policy source,
52
+ residual strategy and stable tie-breaker, pre/post-round audit, and
53
+ positive/zero/negative tests. FX changes also inherit base/quote metadata,
54
+ rate source/time, canonical path, and target settlement rounding.
26
55
  - RED test executed.
27
56
  - Allowed and forbidden files confirmed.
28
57
  - Blockers listed.
@@ -32,3 +61,7 @@ Superpowers technical design is canonical for source-level HOW and execution.
32
61
  Do not claim completion until build/test, real-entry execution, DB/log evidence,
33
62
  test-report update, and SuperBridge Flow hook scripts pass or are marked blocked
34
63
  with concrete reasons.
64
+
65
+ For monetary changes, completion also requires half-cent, residual, or
66
+ multi-detail allocation tests and reconciliation evidence proving that original,
67
+ discount, actual, and allocated totals satisfy the contract identities.
@@ -0,0 +1,96 @@
1
+ # Money Precision Algorithm Baseline
2
+
3
+ Money correctness is a contract across representation, currency units,
4
+ calculation, rounding, allocation, FX, persistence, and audit. Apply this
5
+ baseline to monetary design, implementation prompts, review, and verification.
6
+ An evidenced legal, tax, settlement, or provider contract takes precedence.
7
+
8
+ ## Exact representation and units
9
+
10
+ - Use an exact decimal type such as Java `BigDecimal`, JSR 354
11
+ `MonetaryAmount`, or integer minor units coupled with an explicit currency.
12
+ Never represent or calculate money with binary `float` or `double`.
13
+ - Construct decimals from strings, exact integers, or exact database values,
14
+ never from a `double`.
15
+ - Every amount contract declares currency, unit, and scale. Never assume two
16
+ decimal places. Distinguish ISO 4217 minor units, internal calculation scale,
17
+ settlement scale, and provider-specific minor units.
18
+ - Convert to integer minor units only at an adapter boundary, with provider
19
+ rule, range, and overflow checks.
20
+ - Persist money as `DECIMAL` or `NUMERIC`. Code, DDL, mapper, and API
21
+ precision/scale must agree. Implicit database conversion is not a business
22
+ rounding policy.
23
+
24
+ ## Rounding policy matrix
25
+
26
+ Every rounding point declares its business level, input precision, output
27
+ scale, rounding mode, currency or cash increment, and policy source. Do not
28
+ apply global `HALF_UP`, or mix line-level and aggregate-level rounding inside
29
+ one identity without an explicit reconciliation rule.
30
+
31
+ - Division declares precision and rounding mode. Use `UNNECESSARY` as an
32
+ assertion where an exact result is required.
33
+ - Retain rate, quantity, product, ratio, and FX precision until a contracted
34
+ settlement boundary.
35
+ - Negative values change the meaning of modes such as `CEILING` and `FLOOR`.
36
+ Test refunds and reversals explicitly instead of inferring from positives.
37
+ - Freeze either line-level or invoice-level tax rounding as the authoritative
38
+ path. Route any permitted alternative through an explicit adjustment rule.
39
+
40
+ ## Authoritative totals and complement derivation
41
+
42
+ For `total = component 1 + ... + component N`, identify the authoritative
43
+ total, calculate only N-1 components independently, and derive the final
44
+ component as `authoritative total - sum(other components)`. Never calculate and
45
+ round all components independently and then rebuild the authoritative total.
46
+
47
+ ## Allocation and residual units
48
+
49
+ Calculate high-precision ideal shares, then quantize at the settlement unit:
50
+
51
+ 1. Express the authoritative target in integer settlement units.
52
+ 2. Calculate ideal shares and base allocations.
53
+ 3. Compute `residual units = target units - sum(base allocation units)`.
54
+ 4. Assign one unit at a time until the residual reaches zero.
55
+ 5. Use a declared strategy and stable tie-breaker such as a business key.
56
+
57
+ Prefer the largest-remainder method for neutral proportional allocation. Apply
58
+ the same absolute ordering symmetrically to refunds and negative totals.
59
+ Largest-value or last-row policies are allowed only when the contract requires
60
+ them. Never depend on database natural order. Prove conservation, stable order,
61
+ and idempotence.
62
+
63
+ ## FX conversion
64
+
65
+ - Treat an FX rate as directional data with base/quote currencies, source,
66
+ timestamp, validity, and precision.
67
+ - Never truncate or round the rate early. Do not synthesize an inverse when a
68
+ canonical directional rate exists.
69
+ - Freeze one conversion path. For triangulation, use the contracted reference
70
+ currency and operation order, then round only at the target settlement unit.
71
+ - Round-trip conversion is a tolerance check, not an exact identity.
72
+
73
+ ## Audit and verification
74
+
75
+ Record raw inputs, formula/version, currency/unit, pre-round value, policy,
76
+ post-round value, residual, and residual recipient. Record FX metadata when
77
+ applicable. Test positive, zero, negative/refund, half-unit, non-terminating
78
+ division, multi-detail, tied remainders, overflow, and repeated execution.
79
+
80
+ Required invariants: authoritative total equals component sum; allocations
81
+ equal the settlement target; identical inputs, policy version, and stable order
82
+ produce identical outputs; persistence, API, and provider conversions do not
83
+ silently change money.
84
+
85
+ ## Primary references
86
+
87
+ - ISO 4217: <https://www.iso.org/iso-4217-currency-codes.html>
88
+ - JSR 354: <https://www.jcp.org/en/jsr/detail?id=354>
89
+ - Java `BigDecimal`: <https://docs.oracle.com/en/java/javase/25/docs/api/java.base/java/math/BigDecimal.html>
90
+ - MySQL `DECIMAL`: <https://dev.mysql.com/doc/refman/en/fixed-point-types.html>
91
+ - PostgreSQL `numeric`: <https://www.postgresql.org/docs/15/datatype-numeric.html>
92
+ - Stripe minor units: <https://docs.stripe.com/api/payment_intents/object>
93
+ - Adyen minor units: <https://docs.adyen.com/development-resources/currency-codes/>
94
+ - EU conversion rules: <https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A31997R1103>
95
+ - HMRC VAT rounding: <https://www.gov.uk/hmrc-internal-manuals/vat-trader-records/vatrec12030>
96
+ - SAP allocation residuals: <https://help.sap.com/docs/PRODUCT_ID/56471df1959f4cfd9e3bf7a6d2d5be42/56a2dfe71e8b2ea1e10000000a42189c.html>
@@ -16,6 +16,32 @@
16
16
  - Field/status reverse impact matrix exists when risky fields or states change.
17
17
  - Architecture boundary and call-direction matrix exists for cross-service or
18
18
  external flows.
19
+ - External Integration Configuration And Deployment Contract exists for
20
+ third-party platforms/tools, SDKs, MQ/Kafka, callbacks, payment providers,
21
+ cloud services, or other external integrations. It inventories environment
22
+ sources and provisioning for endpoints, app/tenant/project IDs,
23
+ Topic/Tag/Consumer Group, namespace, webhook, ACL/role, switches, timeouts,
24
+ and credential references; names runtime/provisioning owners, timing,
25
+ readiness evidence, rollback, secret handling, and blockers; and forbids
26
+ environment-dependent hard-coded values. Test auto-creation, existing test
27
+ resources, or local startup do not prove production readiness.
28
+ - Concurrency And Idempotency Ownership exists for concurrent requests, batch
29
+ issue/activation/renewal, duplicate callbacks/consumption, or repeated
30
+ external delivery. It defines the business idempotency key,
31
+ application-layer atomic claim owner, short transaction boundary,
32
+ PENDING/SUCCESS/FAILED state machine, retry code reuse, external-call
33
+ boundary, and uncertain-result reconciliation. A unique index is not the
34
+ default; it is only an optional fallback with explicit natural uniqueness,
35
+ historical cleanup, NULL/soft-delete behavior, and conflict handling.
36
+ - Money precision boundary exists when amount, fee, discount, deduction,
37
+ refund, sharing, payment, invoice, balance, electricity fee, service fee,
38
+ package settlement, proration, allocation, reconciliation, or financial
39
+ display changes. It must separate calculation state from settlement/display
40
+ state, forbid early rounding, and include half-cent, residual, or
41
+ multi-detail allocation evidence. For additive identities, it must name the
42
+ authoritative total and derive the final complement as `authoritative total -
43
+ sum(other components)` instead of calculating and rounding every component
44
+ independently.
19
45
  - SQL risk review exists for DB changes.
20
46
  - No fallback or guesswork is allowed without explicit owner approval.
21
47
 
@@ -5,6 +5,21 @@
5
5
  - Is handoff_hash current and repeated in prompts/report?
6
6
  - Did Superpowers technical design cover source-level HOW?
7
7
  - Were field/status and architecture-boundary risks checked?
8
+ - For monetary changes, were calculation and settlement states separated, were
9
+ scale and rounding mode explicit, and was early rounding rejected?
10
+ - Do half-cent, residual, and multi-detail tests prove deterministic allocation
11
+ and reconciliation of original, discount, actual/refund, and allocated totals?
12
+ - For additive identities, is the authoritative total explicit, with only N-1
13
+ components calculated independently and the final complement derived as
14
+ `authoritative total - sum(other components)`?
15
+ - Reject designs that calculate and round every component independently before
16
+ rebuilding an authoritative total.
17
+ - Are exact decimal construction, currency and provider minor units, rounding
18
+ level/policy source, and persistence precision explicit?
19
+ - Does allocation use a declared residual strategy and stable tie-breaker, with
20
+ positive, zero, negative/refund, tied-remainder, and idempotence evidence?
21
+ - For FX, are base/quote, source/time, canonical path, and target settlement
22
+ rounding explicit, without early rate rounding or synthesized inverses?
8
23
  - Are SQL changes reviewed and tied to total version SQL when needed?
9
24
  - Are RED/GREEN evidence and real-entry evidence present?
10
25
  - Are mock-only results clearly labeled?