Configuration — agents and methods
Module overview · Common methods
P A skill is a complete procedure that can use LLM judgment and optional executable helpers. A tool is runnable source and/or a binary inside that procedure. Connector calls, state transitions, agent calls and permission checks are runtime access, not extra learned skills or imaginary binaries. Only the methods and segments selected for the actual case run.
Agents
configuration.root — Configuration root
P Function. Choose only needed stages and template work items, coordinate parallel catalogue/offer and wiring branches, and return an accepted product result to Support root.
| Contract | Content |
|---|---|
| Input brief | Accepted handover or scoped request, current product/delta source set and destination entitlement |
| Output brief | Stage/dependency plan, scoped child confirmations, decision packets, joint/release obligations and result report |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Only Support controlling root owns original-case closure. Unknown code behaviour can be a Source consult; changing it needs accepted Hd. |
| Memory / ownership | configuration platform/experience |
| Role instructions | Keep origin, intent, route, parent-confirmed plans, gates, accepted handovers, waits and obligations coherent. Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | case.read / papers / task state experience.query (strategy-selected) runtime.validate-contract |
| May consult | support/experience — Has a product of this shape failed before, and how? |
| Eval set | configuration.joint — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.root |
Methods used. common.skill.experience.
Selected segments. common.skill.experience: product-change strategy; current case/tenant and version filters
Source basis. Components, Agents, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.normalizer — Specification normalizer
P Function. Reconcile source meaning into a cited product or delta; coordinate independent tariff reading and measured estate facts without resolving silent business choices.
| Contract | Content |
|---|---|
| Input brief | Raw source inventory, requested delta and current baseline where relevant |
| Output brief | Validated product/delta, source mapping, template work items, batched unresolved questions and H1 review view |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. The normalizer can map live codes; the tariff analyst remains DB-blind. Missing numeric rates and contradictions are questions. |
| Memory / ownership | configuration/normalization |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | ingestion.read-source-revision runtime.render-approved-product-model scoped catalogue reads and semantic reconstruction |
| May consult | support/experience — Has a product of this shape failed before, and how? The known failure becomes an assertion in the stage plan. |
| Eval set | configuration.normalization — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.normalizer |
Methods used. common.skill.experience, configuration.skill.normalization, common.db-prober.
Selected segments. common.skill.experience: product-change strategy; current case/tenant and version filters
Executable helper scope. configuration.tool.product-validator. These are used through the selected method, not an extra workflow.
Source basis. Stage Normalization, Components, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.source_analyst — Independent source analyst
P Function. Read the insurer's material with no access to the estate and no knowledge of the intended model: what the documents actually say about rates, loadings and taxes; covers and their definitions; factors and their value lists; packages and selection rules — the axes nobody announced and the omissions nobody mentioned included. One pass over the whole source set, so a cover named in the tariff and missing from the cover sheet is caught by the same reader.
| Contract | Content |
|---|---|
| Input brief | Extracted source set with cell/page/range coordinates, legends and message ids — and no current-product context The reading rubrics this case selects: tariff, loadings and taxes · covers and definitions · factors and value lists · packages and selection rules |
| Output brief | Per-rubric independent reading with reconciliation counts, suspect cells, not-stated items and cross-rubric mismatches Pre-emptive notes where the documents are silent in a way the customer may not intend |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. No database connector, current-product intent or other spec context: independence is provisioning, not instruction. Mapping anything to a live code is the normalizer's step, never this profile's. One profile, several rubrics, one pass over the source set. It reads what the documents say; it never authors the specification, resolves a contradiction or decides a business question. |
| Memory / ownership | configuration/normalization (document-reading rules only) |
| Runtime / connector access | ingestion.read-source-revision |
| Eval set | configuration.normalization — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.source_analyst |
Methods used. configuration.skill.normalization.
Selected segments. configuration.skill.normalization: Only independent tariff reading from source material; no catalogue/DB inputs or whole-model helper.
Source basis. Stage Normalization, Components, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.tariff_analyst resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.source_analyst.
configuration.abacus.builder — Plan and build the rating
P Function. Interpret the accepted tariff, clarify missing rate meaning and coordinate the selected rating versions, examples and sign-out. It plans and builds: the profile that decides the rating shape is the profile that emits the effect packet, which the target's deterministic effect applier checks, applies and journals.
| Contract | Content |
|---|---|
| Input brief | Source-cited tariff/model, units and rate bases; worked-example oracle; template/caller/date selectors; current engine and shared-file consumers Current case brief, unresolved decisions, target scope and prerequisite result revisions |
| Output brief | Tariff plan and source decisions; selected PPA version/file changes; per-cover and loading examples with currency; selector/version proof; rating summary and downstream seed dependencies Parent-confirmed preparation plan; H2-ready effect packet; independent verdict and H3 sign-out or specific rework request |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Select helpers and tools per work item; tools shown are eligible bindings, not an unconditional sequence. Independent validator checks stage proof; root obtains the applicable decisions. Configuration declares no separate implementer. The target system is the domain, so there is no placement decision to separate, and the apply path is packet → deterministic effect applier on that target → journal (D183, D185). |
| Memory / ownership | configuration/abacus |
| Role instructions | Resolve axes, rate dimensions, cover/loading/tax scope and order; distinguish representable configuration from an evaluator change. Build one bounded plan per required template and join the results. Keep clarification, plan synthesis and the permitted implementation in the same domain task. Use the communication procedure for questions through the root; do not add a mandatory pre-stage agent. Propose the plan to the immediate parent before dependent work. H2 and effect permission remain separate. Select helper tasks only for a concrete template, evidence gap or independently checkable result. Submit fresh proof to the domain validator; repair the failed item and invalidate dependent results when the source or plan changes. |
| Runtime / connector access | connector.rating.compile-version runtime.render-approved-parameters connector.abacus.getRates |
| Context and task continuity | One domain task carries plan, clarification and execution coordination for this selected stage. Parallel template leaf tasks receive their full relevant source slice and report into the same plan; they do not own the global customer request. |
| May consult | support/investigators/pricing — The tariff request map ABC_CFG_PRICING_FACTORS for an INSIS-priced product (D121). source/serdica-backend — Unsupported filter or evaluator logic: a consult on the engine; an accepted code change is Hd. (a change is Hd) |
| Eval set | configuration.rating — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.abacus.builder, configuration.abacus.stage |
Methods used. configuration.skill.rating-build, configuration.skill.rating-verification, common.skill.communication, common.db-prober, common.effect-applier.
Selected segments. common.skill.communication: Clarification segment: source ambiguities and dependency questions via the root; no channel send authority.
Source basis. Components, Stage Abacus, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.abacus.stage resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.abacus.builder.
Former reference configuration.abacus.planner resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.abacus.builder.
configuration.ipal.builder — Plan and build the catalogue and factors
P Function. Translate product meaning into the catalogue, factors and process registration, keeping shared keys and downstream consumers consistent. It plans and builds: the profile that decides the catalogue shape is the profile that emits the effect packet, which the target's deterministic effect applier checks, applies and journals.
| Contract | Content |
|---|---|
| Input brief | Accepted product/delta and source coordinates; S2 rating summary or verified baseline; product/cover/object keys; factor and process-fit evidence Current case brief, unresolved decisions, target scope and prerequisite result revisions |
| Output brief | Catalogue/factor/process plan; scoped semantic delta and returned ids; enabled/consumed controls; downstream mapping and sign-out evidence with unresolved external dependencies Parent-confirmed preparation plan; H2-ready effect packet; independent verdict and H3 sign-out or specific rework request |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Select helpers and tools per work item; tools shown are eligible bindings, not an unconditional sequence. Independent validator checks stage proof; root obtains the applicable decisions. Configuration declares no separate implementer. The target system is the domain, so there is no placement decision to separate, and the apply path is packet → deterministic effect applier on that target → journal (D183, D185). |
| Memory / ownership | configuration/ipal |
| Role instructions | Map PR_BASIC products/covers/definitions, PR_PRICING_FACTORS and dependencies; establish consumed validations/mappings and process fit. Clarify ambiguous cover fan-out, currency, donor semantics or missing consumers before preparing changes. Keep clarification, plan synthesis and the permitted implementation in the same domain task. Use the communication procedure for questions through the root; do not add a mandatory pre-stage agent. Propose the plan to the immediate parent before dependent work. H2 and effect permission remain separate. Select helper tasks only for a concrete template, evidence gap or independently checkable result. Submit fresh proof to the domain validator; repair the failed item and invalidate dependent results when the source or plan changes. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.process.read connector.browser.scoped-action |
| Context and task continuity | One domain task carries plan, clarification and execution coordination for this selected stage. Parallel template leaf tasks receive their full relevant source slice and report into the same plan; they do not own the global customer request. |
| May consult | source/serdica-backend — Does a deployed process fit this product's shape? No is a deploy boundary: Hd to Customer-specific development for the process (D86). (a change is Hd) |
| Eval set | configuration.catalogue — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.ipal.builder, configuration.ipal.stage |
Methods used. configuration.skill.catalogue-build, common.skill.communication, common.db-prober, common.effect-applier.
Selected segments. common.skill.communication: Clarification segment: source ambiguities and dependency questions via the root; no channel send authority.
Source basis. Components, Stage IPAL, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.ipal.stage resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.ipal.builder.
Former reference configuration.ipal.planner resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.ipal.builder.
configuration.offer.builder — Plan and build the packages and selection rules
P Function. Turn the insurer's package and cover-selection wishes into explicit offers, dependencies, limits and rating-template selection. It plans and builds: the profile that decides the packaging shape is the profile that emits the effect packet, which the target's deterministic effect applier checks, applies and journals.
| Contract | Content |
|---|---|
| Input brief | Accepted package/selection wishes; S3 catalogue or verified baseline; insured-object segments; limits and rated-template information Current case brief, unresolved decisions, target scope and prerequisite result revisions |
| Output brief | Insurance package plan; LB/LBRR cover/extent/limit matrix and selector links; positive and forbidden-selection cases; signed package result Parent-confirmed preparation plan; H2-ready effect packet; independent verdict and H3 sign-out or specific rework request |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Select helpers and tools per work item; tools shown are eligible bindings, not an unconditional sequence. Independent validator checks stage proof; root obtains the applicable decisions. Configuration declares no separate implementer. The target system is the domain, so there is no placement decision to separate, and the apply path is packet → deterministic effect applier on that target → journal (D183, D185). |
| Memory / ownership | configuration/offer |
| Role instructions | Resolve mandatory/recommended/optional cover extents, dependencies/conflicts and relative limits. Distinguish object segmentation from seller roles and test the intended offer-template selector against fallback. Keep clarification, plan synthesis and the permitted implementation in the same domain task. Use the communication procedure for questions through the root; do not add a mandatory pre-stage agent. Propose the plan to the immediate parent before dependent work. H2 and effect permission remain separate. Select helper tasks only for a concrete template, evidence gap or independently checkable result. Submit fresh proof to the domain validator; repair the failed item and invalidate dependent results when the source or plan changes. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.abacus.getRates runtime.run-declared-assertions |
| Context and task continuity | One domain task carries plan, clarification and execution coordination for this selected stage. Parallel template leaf tasks receive their full relevant source slice and report into the same plan; they do not own the global customer request. |
| Eval set | configuration.offer — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.offer.builder, configuration.offer.stage |
Methods used. configuration.skill.offer-build, common.skill.communication, common.db-prober, common.effect-applier.
Selected segments. common.skill.communication: Clarification segment: source ambiguities and dependency questions via the root; no channel send authority.
Source basis. Components, Stage Offer, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.offer.stage resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.offer.builder.
Former reference configuration.offer.planner resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.offer.builder.
configuration.serdica.builder — Plan and build the product's wiring
P Function. Make the configured product usable by the intended actors through its routes, access, labels, numbering and document wiring. It plans and builds: the profile that decides the wiring shape is the profile that emits the effect packet, which the target's deterministic effect applier checks, applies and journals.
| Contract | Content |
|---|---|
| Input brief | Accepted product/template and seller/service audience; maintained languages; numbering/document requirements; current routes/endpoints and prerequisite ids Current case brief, unresolved decisions, target scope and prerequisite result revisions |
| Output brief | Wiring plan; route/endpoint/message/numbering/print delta; deployment dependencies; actor/language/document assertions and signed result Parent-confirmed preparation plan; H2-ready effect packet; independent verdict and H3 sign-out or specific rework request |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Select helpers and tools per work item; tools shown are eligible bindings, not an unconditional sequence. Independent validator checks stage proof; root obtains the applicable decisions. Configuration declares no separate implementer. The target system is the domain, so there is no placement decision to separate, and the apply path is packet → deterministic effect applier on that target → journal (D183, D185). |
| Memory / ownership | configuration/serdica |
| Role instructions | Resolve navigation versus backend authorization, exact message keys/languages, numbering principal/office/year and wanted document/copy parameters. Separate configuration from source/evaluator or external-owner dependencies. Keep clarification, plan synthesis and the permitted implementation in the same domain task. Use the communication procedure for questions through the root; do not add a mandatory pre-stage agent. Propose the plan to the immediate parent before dependent work. H2 and effect permission remain separate. Select helper tasks only for a concrete template, evidence gap or independently checkable result. Submit fresh proof to the domain validator; repair the failed item and invalidate dependent results when the source or plan changes. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.browser.scoped-action connector.process.read |
| Context and task continuity | One domain task carries plan, clarification and execution coordination for this selected stage. Parallel template leaf tasks receive their full relevant source slice and report into the same plan; they do not own the global customer request. |
| May consult | source/serdica-ui — Will this render, and to which roles? A missing component is Hd, never a forced near-fit. (a change is Hd) |
| Eval set | configuration.wiring — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.serdica.builder, configuration.serdica.stage |
Methods used. configuration.skill.wiring-build, common.skill.communication, common.db-prober, common.effect-applier.
Selected segments. common.skill.communication: Clarification segment: source ambiguities and dependency questions via the root; no channel send authority.
Source basis. Components, Stage Serdica, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.serdica.stage resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.serdica.builder.
Former reference configuration.serdica.planner resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.serdica.builder.
configuration.integration.validator — Validate the joined product and target result
P Function. Join the selected domain proofs against the original requested product and verify the claimed working-preview or delivered-target result.
| Contract | Content |
|---|---|
| Input brief | Original objective and accepted source/specification revision with citations Selected domain sign-outs and their independent verdicts; current target evidence, actor/template/date assertions, skipped-stage baselines and open dependencies |
| Output brief | Joint or target verdict with covered assertions, separating counterexamples, missing evidence and precise proof scope |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Domain-specific validator remains separate from Data/source review. Preview acceptance is not target delivery; effectful tests require authority. |
| Memory / ownership | configuration (read/propose) |
| Role instructions | Trace the requested consumer journey across the selected stage boundaries; do not repeat each domain's entire test suite without a join risk. Verify current audience, offer/template/version, factors, output and requested persistence/transfer/document surfaces. A native INSIS request does not imply an IPAL journey. Keep working preview, target delivery and customer acceptance separate; return missing external proof as an open obligation. |
| Runtime / connector access | runtime.run-declared-assertions connector.abacus.getRates connector.browser.scoped-action scoped catalogue reads and semantic reconstruction |
| Context and task continuity | Receive the original objective, approved source/specification revision and citations, current baselines, predecessor result references, acceptance assertions, dependencies and known uncertainties. Exclude the author's private reasoning/transcript, not these necessary facts. Request missing evidence from the parent before judging. |
| Eval set | configuration.joint — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.integration.validator, configuration.validator |
Methods used. common.skill.validation, configuration.skill.product-proof, configuration.skill.rating-verification.
Source basis. Components, Configuration Module, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.validator resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.integration.validator.
Former reference configuration.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.integration.validator.
Former reference configuration.integration.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.integration.validator.
configuration.deployer — Configuration release coordinator
P Function. Re-emit exact target artifacts and coordinate authorised configuration delivery, smoke proof and journal-derived recovery.
| Contract | Content |
|---|---|
| Input brief | Accepted sign-outs, requested target set, fresh baseline, H4 packet and applicable execution grants Accepted source/configuration artefact hashes and semantic change set; fresh target inventory, baseline and drift; ordered dependencies and external-owner obligations |
| Output brief | Target execution report, smoke evidence, consumed version map, recovery or remaining obligations Versioned deployment packet: ordered API requests, parameterised SQL/migrations and source/build artefacts only where required; preconditions, exact scope/counts, assertions and recovery commands H4 packet decision, applicable execution grants, effect journal, target verification refs and unresolved recovery/delivery obligations |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Independent configuration validator proves result; no copying target ids or reusing a grant after packet revision. |
| Memory / ownership | configuration (release and recovery rules) |
| Role instructions | Prepare and follow exact operation packets, park external-owner steps and reconcile unknown outcomes before retry. Prepare release scripts and exact API/SQL bindings after the working result is accepted. Reconcile against each target; do not copy working-environment surrogate ids or erase target-only valid state. Reconfirm if target baseline or packet revision changes. Execute only through the shared auditor/gates/executor; reconcile unknown effects before retry. Deployment preparation is after authoring, but still part of requested delivery. Do not close that obligation while scripts, confirmations, application, proof or due restoration are missing. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction runtime.request-effect connector.what-landed runtime.validate-contract |
| Context and task continuity | One deployment responsibility retains the chosen target set, packet revisions, decisions and effect journal through preparation, dispatch, verification and recovery. New target/scope needs a new current packet and authority, not a fresh agent for every step. |
| Eval set | configuration.release — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.deployer |
Methods used. configuration.skill.release, source.skill.release, common.effect-applier.
Source basis. Components, Configuration Module, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.abacus.validator — Validate tariff and selectors
P Function. Re-derive rate bases and expected components from approved source examples; test intended template/caller/date and fallback, effective-date boundaries and unaffected versions. Compilation or a DEPLOYED row alone is insufficient.
| Contract | Content |
|---|---|
| Input brief | Source-cited tariff/model, units and rate bases; worked-example oracle; template/caller/date selectors; current engine and shared-file consumers Current proposed delta and observations with provenance; original acceptance/failure assertions; current scope, known gaps and dependencies |
| Output brief | Domain verdict per assertion: pass, counterexample or missing evidence; exact source/target revisions and proof limits |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Domain-specific validator remains separate from Data/source review. Preview acceptance is not target delivery; effectful tests require authority. |
| Memory / ownership | configuration/abacus (read/propose; planner owns canonical publication) |
| Role instructions | Re-derive rate bases and expected components from approved source examples; test intended template/caller/date and fallback, effective-date boundaries and unaffected versions. Compilation or a DEPLOYED row alone is insufficient. Return the verdict to the requesting parent with evidence; do not edit the author's artefacts, approve the result or acquire target authority. |
| Runtime / connector access | connector.abacus.getRates scoped PPA version/file/selector reads runtime.run-declared-assertions |
| Context and task continuity | Receive the original objective, approved source/specification revision and citations, current baselines, predecessor result references, acceptance assertions, dependencies and known uncertainties. Exclude the author's private reasoning/transcript, not these necessary facts. Request missing evidence from the parent before judging. |
| Eval set | configuration.rating — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.abacus.validator |
Methods used. common.skill.validation, configuration.skill.rating-verification.
Source basis. Stage Abacus, Configuration case-specific work and falsifiers.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.abacus.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.abacus.validator.
configuration.ipal.validator — Validate catalogue and consumers
P Function. Check meaning and returned keys across cover/object definitions, factor dependencies, validations and actual downstream consumers. Include historical/unaffected keys and the operator flow when its wiring is ready; row presence or a populated dropdown is insufficient.
| Contract | Content |
|---|---|
| Input brief | Accepted product/delta and source coordinates; S2 rating summary or verified baseline; product/cover/object keys; factor and process-fit evidence Current proposed delta and observations with provenance; original acceptance/failure assertions; current scope, known gaps and dependencies |
| Output brief | Domain verdict per assertion: pass, counterexample or missing evidence; exact source/target revisions and proof limits |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Domain-specific validator remains separate from Data/source review. Preview acceptance is not target delivery; effectful tests require authority. |
| Memory / ownership | configuration/ipal (read/propose; planner owns canonical publication) |
| Role instructions | Check meaning and returned keys across cover/object definitions, factor dependencies, validations and actual downstream consumers. Include historical/unaffected keys and the operator flow when its wiring is ready; row presence or a populated dropdown is insufficient. Return the verdict to the requesting parent with evidence; do not edit the author's artefacts, approve the result or acquire target authority. |
| Runtime / connector access | scoped catalogue/factor/mapping reads connector.process.read runtime.run-declared-assertions |
| Context and task continuity | Receive the original objective, approved source/specification revision and citations, current baselines, predecessor result references, acceptance assertions, dependencies and known uncertainties. Exclude the author's private reasoning/transcript, not these necessary facts. Request missing evidence from the parent before judging. |
| Eval set | configuration.catalogue — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.ipal.validator |
Methods used. common.skill.validation, configuration.skill.catalogue-build, configuration.skill.product-proof.
Source basis. Stage IPAL, Configuration case-specific work and falsifiers.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.ipal.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.ipal.validator.
configuration.offer.validator — Validate package rules and selection
P Function. Independently exercise LB/LBRR dependencies, conflicts and restrictions; assert optionality, cover selection and TemplateOfferCode/OfferCode template resolution. Available cover is not selected cover; pricing one happy path is insufficient.
| Contract | Content |
|---|---|
| Input brief | Accepted package/selection wishes; S3 catalogue or verified baseline; insured-object segments; limits and rated-template information Current proposed delta and observations with provenance; original acceptance/failure assertions; current scope, known gaps and dependencies |
| Output brief | Domain verdict per assertion: pass, counterexample or missing evidence; exact source/target revisions and proof limits |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Domain-specific validator remains separate from Data/source review. Preview acceptance is not target delivery; effectful tests require authority. |
| Memory / ownership | configuration/offer (read/propose; planner owns canonical publication) |
| Role instructions | Independently exercise LB/LBRR dependencies, conflicts and restrictions; assert optionality, cover selection and TemplateOfferCode/OfferCode template resolution. Available cover is not selected cover; pricing one happy path is insufficient. Return the verdict to the requesting parent with evidence; do not edit the author's artefacts, approve the result or acquire target authority. |
| Runtime / connector access | scoped LB/LBRR reads connector.offer.resolve connector.abacus.getRates runtime.run-declared-assertions |
| Context and task continuity | Receive the original objective, approved source/specification revision and citations, current baselines, predecessor result references, acceptance assertions, dependencies and known uncertainties. Exclude the author's private reasoning/transcript, not these necessary facts. Request missing evidence from the parent before judging. |
| Eval set | configuration.offer — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.offer.validator |
Methods used. common.skill.validation, configuration.skill.offer-build, configuration.skill.product-proof.
Source basis. Stage Offer, Configuration case-specific work and falsifiers.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.offer.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.offer.validator.
configuration.serdica.validator — Validate product access and wiring
P Function. Check intended and excluded actors at navigation and backend boundaries; exact visible maintained-language text; sequence/grant/synonym principal; wanted and unwanted print documents/copies. Gate any number allocation, printing or issued test separately.
| Contract | Content |
|---|---|
| Input brief | Accepted product/template and seller/service audience; maintained languages; numbering/document requirements; current routes/endpoints and prerequisite ids Current proposed delta and observations with provenance; original acceptance/failure assertions; current scope, known gaps and dependencies |
| Output brief | Domain verdict per assertion: pass, counterexample or missing evidence; exact source/target revisions and proof limits |
| Scope and restrictions | One scoped task instance per selected responsibility; a reused profile never shares another case transcript, grants or mutable task state. Boot from the own domain index, task request, pinned profile and confirmed parent plan. Select relevant articles, skills, tools and helpers on demand. Domain-specific validator remains separate from Data/source review. Preview acceptance is not target delivery; effectful tests require authority. |
| Memory / ownership | configuration/serdica (read/propose; planner owns canonical publication) |
| Role instructions | Check intended and excluded actors at navigation and backend boundaries; exact visible maintained-language text; sequence/grant/synonym principal; wanted and unwanted print documents/copies. Gate any number allocation, printing or issued test separately. Return the verdict to the requesting parent with evidence; do not edit the author's artefacts, approve the result or acquire target authority. |
| Runtime / connector access | scoped route/endpoint/message/numbering/print metadata reads connector.browser.scoped-action runtime.run-declared-assertions |
| Context and task continuity | Receive the original objective, approved source/specification revision and citations, current baselines, predecessor result references, acceptance assertions, dependencies and known uncertainties. Exclude the author's private reasoning/transcript, not these necessary facts. Request missing evidence from the parent before judging. |
| Eval set | configuration.wiring — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | product-change; retrieval and index contract |
| Runtime profile identity | configuration.serdica.validator |
Methods used. common.skill.validation, configuration.skill.wiring-build, configuration.skill.product-proof.
Source basis. Stage Serdica, Configuration case-specific work and falsifiers.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.serdica.verifier resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.serdica.validator.
Domain procedures
configuration.skill.rating-verification — Check selected-cover pricing
P Function. Build factors from the declared current product and compare worked examples per selected cover, loading, tax and roll-up.
| Contract | Content |
|---|---|
| Input brief | Current product factors, selected covers, template/version and approved numeric oracle |
| Output brief | Per-component comparison and unsupported/missing-factor result |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. No invented rates, default factors or summing all offered covers; annual and period amounts remain distinct. Shared capability identity does not widen the requesting case data class, model route, environment, evidence visibility or permitted effects. |
| Runtime / connector access | connector.abacus.getRates runtime.run-declared-assertions |
Source basis. abacus getrates direct call, abacus rates presentation, Stage Abacus, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference common.skill.rating-replay resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.skill.rating-verification.
configuration.skill.normalization — Understand and reconcile product sources
P Function. Use LLM interpretation to turn tariff/workbook/message evidence into a scoped business model. Re-read the source to review omissions, implicit axes, unit/base/order and contradictions. Only then run mechanical model checks and generate the review view.
| Contract | Content |
|---|---|
| Input brief | Immutable source inventory with file/sheet/range or message identity, requested outcome and known baseline Raw tables/text plus layout evidence where headers, merged cells or page structure carry meaning |
| Output brief | Source-grounded product/delta model, tariff interpretation and evidence matrix Source-to-model semantic review, unresolved questions, helper-check results and a reviewable brief for H1 |
| Scope and restrictions | The LLM and isolated tariff analyst establish meaning; schema validation cannot prove the source was interpreted correctly. 9951: infer 3 deductible × 3 vehicle-group layers from titles/merged columns and reconcile all 9; a helper cannot assume the axes. CF-05: bind KOLHIDA_EUR/MAXO_EUR leaf tasks and selector/effective-version scope; compilation alone is insufficient. CF-18: later material changes the input revision and reopens affected assumptions/approvals. Existing-product deltas keep unchanged prerequisite facts as current baselines rather than fabricated sign-outs. |
| Runtime / connector access | Source readers; optional case-specific openpyxl/pandas extraction preserving coordinates and values LLM semantic interpretation and independent source review Model-only consistency checker; deterministic projection of the approved model |
Optional executable helpers. configuration.tool.product-validator. Helper availability and role scope are checked separately.
Source basis. Stage Normalization, skill, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference configuration.skill.tariff-extraction resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses configuration.skill.normalization.
configuration.skill.rating-build — Build the accepted rating version
P Function. Map tariff rules to supported engine version/templates/grids and classify rules requiring code; retain the configuration path per object.
| Contract | Content |
|---|---|
| Input brief | Accepted tariff/model, template selectors, fresh engine contracts and worked examples |
| Output brief | Prepared rating objects, coverage map, verified summary and sign-out |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. Unsupported filter/evaluator logic needs a Source consult and accepted change, not an invented grid rule. |
| Runtime / connector access | connector.rating.compile-version runtime.render-approved-parameters connector.abacus.getRates |
Source basis. Stage Abacus, abacus schema, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.skill.catalogue-build — Build catalogue and factor wiring
P Function. Use the rating summary to prepare the declared product, covers, definitions, factors and fitting existing process registration.
| Contract | Content |
|---|---|
| Input brief | Accepted product specification, rating sign-out, measured keys and process-fit evidence |
| Output brief | Catalogue/factor/process plan, returned logical identities, assertions and sign-out |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. A new process is Source work. Reuse shared codes only when semantics and keys match; assigned users are outside Configuration. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.process.read connector.browser.scoped-action |
Source basis. Stage IPAL, product config new product, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.skill.offer-build — Build packages and cover rules
P Function. Map commercial requirements to offers, mandatory/recommended/optional extents, limits, dependencies/conflicts and template selection.
| Contract | Content |
|---|---|
| Input brief | Current catalogue, approved offer rules, selector scope and worked examples |
| Output brief | Offer/limit/selection matrix with positive and deliberately forbidden cases |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. Available covers are not selected covers; M/R/O carries the inclusion meaning. No duplicate default switch. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.abacus.getRates runtime.run-declared-assertions |
Source basis. Stage Offer, lb offers and cover rules, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.skill.wiring-build — Make the product reachable
P Function. Prepare the routes, endpoints, role definitions, language messages, numbering and print configuration required for a usable product.
| Contract | Content |
|---|---|
| Input brief | Accepted wiring needs, tariff sign-out and current route/role/number/print evidence |
| Output brief | Wiring plan, shared-row dependencies, deployed-evaluator prerequisites and actor/language tests |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. Role definitions do not authorise assigning a user. Process registration belongs to S3; source appsettings requires Source ownership. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction connector.browser.scoped-action connector.process.read |
Source basis. Stage Serdica, cfg print docs template configuration, never promote config ahead of evaluator, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.skill.product-proof — Prove the usable product
P Function. Join selected stage sign-outs and test quote, issue, package restrictions, role reachability and output against the same accepted revision.
| Contract | Content |
|---|---|
| Input brief | Selected-stage sign-outs, worked examples, audience, working/target environment and proof scope |
| Output brief | Joint proof matrix with preview/target/delivery acceptance distinguished |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. Policy issue and document production are effectful tests. A working preview does not close target delivery. |
| Runtime / connector access | connector.abacus.getRates connector.browser.scoped-action runtime.run-declared-assertions scoped catalogue reads and semantic reconstruction |
Source basis. Components, Configuration Module, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
configuration.skill.release — Re-emit and release the target set
P Function. Re-derive target identities and ordered apply/recovery artifacts, classify shared rows/versions and coordinate the approved release.
| Contract | Content |
|---|---|
| Input brief | Accepted sign-outs, requested targets, current baselines and release policy |
| Output brief | Target-specific release packet, actual outcomes, smoke proof and recovery/residue record |
| Scope and restrictions | Selected for the current task when its applicability and published version fit; not loaded or executed merely because the stage exists. The procedure adds no target authority; every effect, including effectful verification, retains its own gate and journal. Never transplant environment ids or silently deploy every version consuming a changed file. |
| Runtime / connector access | runtime.render-approved-parameters scoped catalogue reads and semantic reconstruction runtime.request-effect connector.what-landed |
Source basis. Components, Configuration Module, Agents Memory, Actual configuration decisions, helpers and all 25 CF cases.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Executable helpers used by procedures
configuration.tool.product-validator — validate_product.py — model consistency helper
P Function. Run the existing Python checker against supplied product JSON: schema types, model recounts, cross-references and declared unresolved items. It does not read the original tariff/workbook and cannot confirm its meaning.
| Contract | Content |
|---|---|
| Input brief | Authored specification/delta and pinned schema/field catalogue |
| Output brief | Existing CLI: warning/blocker text and exit status. The governed adapter records input/schema hashes and structured check results; the current script does not emit that adapter envelope. |
| Scope and restrictions | A self-consistent but incorrectly interpreted model can pass. Source claims/counts are themselves supplied inputs. LLM source-to-model semantic review and the human scope decision remain required. No DB writes, tariff interpretation or business approval. The current seed warns if jsonschema is unavailable; a governed binding must treat an unrun structural check as incomplete, not as H1 acceptance. |
Executable evidence. Existing source helper; no published v2 binary/package is claimed.
- Source: validate_product.py.
Invocation: python validate_product.py product.json.
Source basis. validate product, Stage Normalization, configuration-change.schema, tool-package.schema.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed