Customer-specific development — 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
source.root — Customer-specific development root
P Function. Own accepted customer defect or bounded extension work, placement and release gates, source task coordination and return evidence to the originating root.
| Contract | Content |
|---|---|
| Input brief | Accepted customer defect/extension brief, lineage and estate ownership, requested target set |
| Output brief | Source task graph, accepted brief, root decision packets, scoped handovers and complete result/remaining 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. Core Development is outside this workflow: return that classification to Support root and the communicator for the request, offer and delivery estimate. Acceptance never opens a Core pipeline. |
| Memory / ownership | source 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 — What has gone wrong in this seam, service or product before? |
| Eval set | source.planning — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.root |
Methods used. common.skill.experience.
Selected segments. common.skill.experience: source-change strategy; current case/tenant and version filters
Source basis. Components, Core Development.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.planner — Placement planner
P Function. Establish reproduction and actual source lineage, choose the smallest fitting seam and prepare the brief for owner approval.
| Contract | Content |
|---|---|
| Input brief | Accepted request/evidence, current branch/source registry and source/configuration contracts |
| Output brief | Reproduction or evidence substitute, branch set, selected/rejected seams, blast radius, tests and recovery brief |
| 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 implementation before accepted H2. A config-only remedy returns to Configuration through root-accepted work, not an unneeded code patch. |
| Memory / ownership | source; selected repository nodes |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | connector.git.read connector.oracle.source-and-signature connector.logs.read experience.query (strategy-selected) |
| May consult | support/experience — What has gone wrong in this seam, service or product before? It sizes the blast radius. configuration/ipal — Is this behaviour a catalogue row rather than a code path? Yes ends the case here: Hd to Configuration. (a change is Hd) |
| Eval set | source.planning — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.planner |
Methods used. source.skill.placement, common.skill.experience, common.db-prober.
Selected segments. common.skill.experience: source-change strategy; current case/tenant and version filters
Source basis. Stage Planning, Components.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.implementer — Source implementer — one instance per declared source
P Function. The stage profile of Implementation: one bounded instance per declared source the accepted change touches, booted from that source's own memory node. No source is privileged — the backend services, the UI, stored code and any other source the customer plug-in declares are siblings — and a declared source with no node of its own is worked by this profile from the approved brief alone.
| Contract | Content |
|---|---|
| Input brief | Approved brief naming the seam, the branch set, the test plan and the cross-source contract paper where two sources are involved The declared source this instance is bound to, with its own build truths, layout and conventions |
| Output brief | Worktree commits in that one source, the failing test first where the change is a defect fix, and the pipeline run S4 reads Interface gaps and a source-domain learning proposal |
| 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 protected-branch push, deployment or editing another source outside the task. One instance works one declared source. A change that needs another source recruits that source's instance; widening the diff is not diligence. A source with a dedicated node uses that node's profile — serdica-backend, serdica-ui, db-plsql, intentgpt — and this profile is the seam that dispatches to them. |
| Memory / ownership | source/<declared source> |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | workspace.create / edit / diff connector.git.read runtime.run-pinned-build-command |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.implementer |
Methods used. common.db-prober.
Source basis. Components, Stage Implementation.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.implementer.serdica-backend — Backend implementer
P Function. Implement only the accepted task in this declared source and return reproducible change/test evidence.
| Contract | Content |
|---|---|
| Input brief | Accepted brief and cross-source shape, pinned source head, task scope and source-specific build/test contract |
| Output brief | Confined diff/build state, actual checks, interface gaps and source-domain learning proposal |
| 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 protected-branch push, deployment or editing other sources outside the task. Preserve the agreed DTO/GraphQL/RabbitMQ/DB contract and the target layout; a shared-library change names its consuming services. The Python riskmodel under Analytics/ is part of this source: use the repository's own pytest and golden-master toolchain (its analytics-* skills), not the .NET one; a rating-filter change ships as the anl-risk-model artefact. A declared-source variant of source.implementer: siblings, none of them the stage's default. |
| Memory / ownership | source/serdica-backend/<service> |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | workspace.create / edit / diff connector.git.read runtime.run-pinned-build-command |
| May consult | configuration/abacus — Engine questions on rating behaviour; after two failed attempts an unknown contract goes to the owning repository agent. |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.implementer.serdica-backend |
Source basis. Components, Stage Implementation, serdica-backend AGENTS.md — bulstrad-staging build truths, serdica-backend CLAUDE.md — layout, the two service patterns, build and test commands, create-plugin skill (customer seam), create-operation skill (premium, underwriting, confirmation operations), analytics-rating-engine skill (Python riskmodel).
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.implementer.serdica-ui — Frontend implementer
P Function. Implement only the accepted task in this declared source and return reproducible change/test evidence.
| Contract | Content |
|---|---|
| Input brief | Accepted brief and cross-source shape, pinned source head, task scope and source-specific build/test contract |
| Output brief | Confined diff/build state, actual checks, interface gaps and source-domain learning proposal |
| 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 protected-branch push, deployment or editing other sources outside the task. Operator journey, actual requests/state and maintained translations matter; a source/DOM test is not a real first-click browser test. bulstrad-staging is Angular 16 / Nx 16 / NGXS 3.8, one app with hash routing and no Module Federation; the master-only import aliases do not exist on this lineage (serdica-ui CLAUDE.md, verified 10.09.2026). A declared-source variant of source.implementer: siblings, none of them the stage's default. |
| Memory / ownership | source/serdica-ui |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | workspace.create / edit / diff runtime.run-pinned-build-command connector.browser.scoped-action |
| May consult | configuration/abacus — Engine questions on rating behaviour; after two failed attempts an unknown contract goes to the owning repository agent. |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.implementer.serdica-ui |
Source basis. Components, Stage Implementation, serdica-ui CLAUDE.md — bulstrad-staging lineage facts, formly-form skill, ngxs-state skill.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.implementer.intentgpt — Declared behaviour-source implementer
P Function. Implement only the accepted task in this declared source and return reproducible change/test evidence.
| Contract | Content |
|---|---|
| Input brief | Accepted brief and cross-source shape, pinned source head, task scope and source-specific build/test contract |
| Output brief | Confined diff/build state, actual checks, interface gaps and source-domain learning proposal |
| 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 protected-branch push, deployment or editing other sources outside the task. Bind only when that declared source is reachable. The current catalogue does not invent repository contents, tests or implementation when unavailable. A declared-source variant of source.implementer: siblings, none of them the stage's default. |
| Memory / ownership | source/intentgpt |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | workspace.create / edit / diff connector.git.read runtime.run-pinned-build-command |
| May consult | configuration/abacus — Engine questions on rating behaviour; after two failed attempts an unknown contract goes to the owning repository agent. |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.implementer.intentgpt |
Source basis. Components, Stage Implementation.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.implementer.db-plsql — Stored-code implementer
P Function. Prepare the smallest signature-guarded stored-code change, reconciling live edits and all called/remote dependencies.
| Contract | Content |
|---|---|
| Input brief | Accepted brief, exact live/baseline object snapshots and caller/dependency map |
| Output brief | Full object apply/inverse proposal, narrow diff, signature preconditions and runtime smoke plan |
| 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 direct DDL, stale-body replacement, schema-wide recompile or automatic INSIS-core ownership. A declared-source variant of source.implementer: siblings, none of them the stage's default. |
| Memory / ownership | source/db-plsql |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | connector.oracle.source-and-signature workspace.create / edit / diff connector.git.read runtime.validate-contract |
| May consult | configuration/abacus — Engine questions on rating behaviour; after two failed attempts an unknown contract goes to the owning repository agent. |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.implementer.db-plsql |
Methods used. source.skill.plsql-change.
Source basis. Components, Stage Planning, db resident procedure fix workflow.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.camunda_developer — Process developer
P Function. Author BPMN/task contracts and the migration/teardown statement for an accepted process change.
| Contract | Content |
|---|---|
| Input brief | Accepted process brief, current definitions/instances and generated source contract |
| Output brief | Process and generated source diff, tests, migration/residue statement and configuration registration prerequisite |
| 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 register/deploy/start/terminate/migrate without the separate effect path. Camunda 7 Community Edition is end-of-life and serdica-backend master carries a proposed migration to Elsa Workflows (Camunda obsoletion decision, 11.02.2026); the Bulstrad lineage still runs Camunda — 344 BPMN files and 242 C# references on bulstrad-staging (counted 10.09.2026) — so a process change is authored for Camunda on the customer lineage and its master counterpart is a brief decision (Stage Deployment Scripts, G27). |
| Memory / ownership | source/serdica-backend/core (proposal to backend owner) |
| Role instructions | Ask the owning profile for a bounded interpretation when knowledge is insufficient; return evidence to the asking case. |
| Runtime / connector access | workspace.edit-process-source connector.process.read runtime.run-pinned-build-command |
| Eval set | source.implementation — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.camunda_developer |
Methods used. source.skill.process-change.
Source basis. Components, Stage Implementation, create-bpmn-process skill (backend), camunda-process skill (UI stepper).
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.validator — Independent source validator — validates the change against the approved brief
P Function. Review placement, actual diff, contract compatibility and test coverage, then independently assess declared release/recovery proof when selected.
| Contract | Content |
|---|---|
| Input brief | Accepted brief, exact source head/diff, test results or scoped target evidence; no author transcript |
| Output brief | Head-bound review verdict, corrections, coverage gaps or target-proof disagreement |
| 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 author identity/reasoning, implementation of the reviewed change or approval of a changed unreviewed head. Checklist seeds are the repositories' own reviewer agents: security, EF Core misuse, async patterns, translations and plugins for the backend; Angular 16, NGXS, RxJS, Apollo and Formly patterns for the UI. |
| Memory / ownership | source (read/propose) |
| Role instructions | Re-derive the specific claim from objective, evidence and provenance; return the evidence separating agreement from disagreement. |
| Runtime / connector access | connector.git.read runtime.run-pinned-build-command compare current scoped evidence connector.logs.read |
| Context and task continuity | Receive original objective/requirements, relevant source citations, current baseline and observations, exact proposed or applied artefacts, assertions, dependencies and known uncertainties. Exclude author private reasoning/transcript, not necessary facts; ask the parent for missing evidence before judging. |
| Eval set | source.review — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.validator |
Methods used. common.skill.validation.
Source basis. Components, Stage Test and Corrections, backend code-reviewer agent, UI code-reviewer agent.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference source.reviewer resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses source.validator.
source.deployer — Source release coordinator
P Function. Prepare the complete target set and ports, coordinate current authorised release operations, and reconcile proof or recovery per consumer.
| Contract | Content |
|---|---|
| Input brief | Reviewed brief/head, exact artifacts and consuming services, current targets and release policy Accepted source/configuration artefact hashes and semantic change set; fresh target inventory, baseline and drift; ordered dependencies and external-owner obligations |
| Output brief | Release/preflight packet, operation outcomes, serving proof and per-target port/recovery/residue report 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. Target settings stay target-specific; one library artifact or one target pass does not close every affected consumer. Independent reviewer checks runtime result. |
| Memory / ownership | source (release/port/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 | connector.git.read workspace.create / edit / diff runtime.validate-contract connector.pipeline.dispatch (gated) connector.what-landed connector.oracle.source-and-signature |
| 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. |
| May consult | configuration/ipal — What state is this product in on this environment? It tells a defect apart from a half-deployed product at S8 verification. |
| Eval set | source.deployment — declared by the owning module manifest or platform-profiles.json; green before publication (Agents § 0.1) |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.deployer |
Methods used. source.skill.release, source.skill.port, source.skill.release-audit, common.effect-applier.
Source basis. Components, Stage Deployment Scripts, cross branch port check base commit.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Domain procedures
source.skill.release-audit — Check the serving release
P Function. Establish the actual serving artifacts and declared customer behaviour per target/consumer after a release.
| Contract | Content |
|---|---|
| Input brief | Release manifest, target receipts, current runtime source/version and smoke plan |
| Output brief | Serving revision and behaviour matrix, incomplete consumers and ports |
| 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. Build, pipeline receipt, deployment metadata and VALID alone cannot establish that the correct implementation runs. Shared capability identity does not widen the requesting case data class, model route, environment, evidence visibility or permitted effects. |
| Runtime / connector access | connector.logs.read connector.git.read connector.oracle.source-and-signature runtime.run-declared-assertions |
Source basis. Components, verify prod object state on prod connection, Agents Memory.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Former reference common.skill.target-release-proof resolves to this responsibility. Existing task/profile revisions are not rewritten; new work uses source.skill.release-audit.
source.skill.placement — Choose the smallest fitting source change
P Function. Reproduce or name the evidence substitute; compare configuration, customer seam and existing extension, or classify excluded core work for communication.
| Contract | Content |
|---|---|
| Input brief | Requested behaviour, mechanism, source/branch set and estate ownership |
| Output brief | Reviewable brief with selected/rejected seams, test plan, scope and release/recovery route, or excluded Core classification |
| 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. Core Development stops at classification and communication of the request, offer and delivery estimate. It never enters this or another development pipeline, including after acceptance. |
| Runtime / connector access | connector.git.read connector.oracle.source-and-signature connector.logs.read |
Source basis. Stage Planning, Core Development, Agents Memory.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.skill.plsql-change — Prepare a stored-code change
P Function. Capture and three-way reconcile live stored code, preserve untouched content and produce a signature-guarded apply/inverse with caller proof.
| Contract | Content |
|---|---|
| Input brief | Accepted brief, baseline/current spec and body, signatures, owner and target |
| Output brief | Full deployable object script, narrow diff, preconditions, callers/dependencies and runtime smoke |
| 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. Snapshot status or timestamp is insufficient; external writer exclusion is required. No schema-wide recompile or INSIS-core code ownership assumption. |
| Runtime / connector access | connector.oracle.source-and-signature workspace.create / edit / diff connector.git.read runtime.validate-contract |
Source basis. db resident procedure fix workflow, Stage Planning, Agents Memory.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.skill.process-change — Author a process with migration intent
P Function. Modify a process/task contract only after defining how existing running instances and generated consumers are handled.
| Contract | Content |
|---|---|
| Input brief | Accepted process brief, current BPMN/task code and running-instance constraints |
| Output brief | Process/source diff, generated contracts, tests and migration/teardown statement |
| 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. Changes to stored definition, registration and running instances are separate effect boundaries. |
| Runtime / connector access | workspace.edit-process-source connector.process.read runtime.run-pinned-build-command |
Source basis. Components, Stage Implementation, Agents Memory, create-bpmn-process skill (backend).
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.skill.release — Prepare every release component
P Function. Enumerate binaries, consuming services, DB objects, configuration/evaluator dependencies, target order, ports and inverse limits.
| Contract | Content |
|---|---|
| Input brief | Reviewed source head, consumers, target branch set, deployment mechanism and recovery requirements |
| Output brief | H4 packet with exact artifacts/jobs/targets, compatibility order, ports and rollback limits |
| 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 library build is not its consumers rollout; missing body/evaluator blocks an otherwise present configuration. |
| Runtime / connector access | connector.git.read connector.oracle.source-and-signature runtime.validate-contract connector.pipeline.dispatch (gated) |
Source basis. Stage Deployment Scripts, never promote config ahead of evaluator, Agents Memory.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
source.skill.port — Port only the real delta
P Function. Compare target base feature and layout before adapting the remaining fix; preserve target environment configuration.
| Contract | Content |
|---|---|
| Input brief | Source commit/delta, current target ref, layout map and tests |
| Output brief | Adapted target diff/commit, excluded target settings, no-op/no-counterpart decision and tested target evidence |
| 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 carry source appsettings/credentials across environment branches. Port outcome is explicit per target. |
| Runtime / connector access | connector.git.read workspace.create / edit / diff runtime.run-pinned-build-command |
Source basis. cross branch port check base commit, Stage Deployment Scripts, Agents Memory.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed
Executable helpers used by procedures
source.test_runner — Deterministic test task
P Function. Run the declared build/test/evaluation set on the pinned build state and record what actually ran.
| Contract | Content |
|---|---|
| Input brief | Build-state hash, permitted commands, test selection and declared result contract |
| Output brief | Actual command/exit status, discovered/run/skipped counts, reports and S3 correction evidence |
| 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. A tool, not an agent: code with a contract that the platform runs on a schedule or on request. It holds its task in the ledger like anything else, but there is no prompt, no model route and no eval set. execution_kind=deterministic: no model route/call per test, no silently ignored filter, no deployment. Pinned commands per lineage: dotnet build/test on the solution or the affected project; nx build, lint and test for serdica-ui; LSP diagnostics are not authoritative in serdica-backend. |
| Memory / ownership | No writable memory domain; reads only declared build/test instructions |
| Runtime / connector access | runtime.run-pinned-build-command |
| Experience strategy | source-change; retrieval and index contract |
| Runtime profile identity | source.test_runner |
Source basis. Components, Stage Test and Corrections, backend build-validator agent, backend test-runner agent, UI test-runner agent.
P Status. planned; seeded from the cited architecture and AISA workspace evidence; no published v2 capability is claimed