Strengthen review contracts and add AL reliability guidance (#196)

* Strengthen review contracts and HTTP guidance

- add outbound HttpClient transport and HTTP status review rules with paired fixtures`n- resolve layered action-skill overrides deterministically across enabled layers`n- validate findings reports and enforce measurable changed-fixture coverage

* Add data handling and test isolation guidance

- add SCM guidance for deriving base quantities through line unit-of-measure validation`n- add security guidance for parameterizing SetFilter with external text`n- add test isolation guidance for resetting per-test state before initialization guards`n- add web-service guidance for JSON null handling and invariant standard format 9`n- route and cover all five rules with paired evaluation fixtures

* Fix findings report rollup validation

* Validate findings report rollups

* Enforce merged finding identity

* Fix locationless finding deduplication

* Reject conflicting merged corrections

* Detect conflicting leaf corrections

* Route HTTP error checks to canonical web-services knowledge

Let the Error Handling leaf conditionally retrieve the existing HTTP owner articles, preserving applicability and exact-path provenance. Add deterministic source-contract and retrieval regressions without duplicating knowledge rules.

Copilot-Session-Id: a92a7788-103e-4651-9b84-19e34caffb94

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: wenjiefan <wenjiefan@microsoft.com>
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Jesper Schulz-Wedde <jesper.schulzwedde@microsoft.com>

microsoft/skills/review/al-code-review.md

@@ -46,8 +46,12 @@ The sub-skills invoked by this skill are those listed in frontmatter `sub-skills
 
 Hosts that orchestrate leaves mechanically SHOULD run
 `tools/Build-SkillIndex.ps1` and resolve this skill by `id: al-code-review`.
-The generated `subSkills` array preserves the frontmatter order and avoids
-host-specific Markdown parsing.
+The generated `subSkills` array preserves the frontmatter slot order and
+avoids host-specific Markdown parsing. Before invoking leaves, run
+`tools/Resolve-SkillWorklist.ps1` with this skill's path and the task's enabled
+layers and disabled skill paths. Each declared path supplies a leaf `id`; the
+resolver selects the highest-precedence enabled implementation with that `id`
+without changing slot order.
 
 ## Relevance
 

microsoft/skills/review/al-error-handling-review.md

@@ -22,6 +22,13 @@ An orchestrator invokes this skill with a `pr-diff`, `file-path`, or `folder-pat
 
 Use READ's **Bounded retrieval for review skills** workflow with `-Domain error-handling`. Consume every catalog page across enabled layers before applying this leaf's Relevance and Worklist; preserve each exact catalog path and open complete bodies only for exact paths selected by the Worklist. If the helper or prepared index is unavailable or invalid, use READ's explicit path-discovery and bounded native-read fallback.
 
+When the review scope contains outbound `HttpClient.Get` or `HttpClient.Post` calls, including resolved call paths, also retrieve every catalog page with `-Domain web-services`, using the same known task dimensions and enabled layers. Restrict supplementary candidates to these article slugs across enabled layers; the links identify their canonical owners:
+
+- [`handle-httpclient-platform-failure-before-response-access`](../../knowledge/web-services/handle-httpclient-platform-failure-before-response-access.md)
+- [`check-http-status-before-consuming-response-body`](../../knowledge/web-services/check-http-status-before-consuming-response-body.md)
+
+Retain each selected catalog row's exact `path`; do not invent paths for missing, pruned, or disabled entries. Apply this leaf's Relevance, Worklist, and READ layer precedence to the supplementary candidates before retrieving complete bodies with `Get-KnowledgeArticles.ps1`. Apply each selected article's own scope and exceptions when evaluating code and agent-finding candidates. Use READ's bounded path-discovery fallback over these same sources when needed. This supplements error-handling knowledge, not the scope of the review with unrelated web-services concerns.
+
 ## Relevance
 
 Apply the frontmatter matching rules defined in READ (*Frontmatter matching semantics*) against the task context:
@@ -40,6 +47,7 @@ Narrow the relevant files to the subset that applies to the changes under review
 - The changed AL object names and types — especially codeunits that post or validate, tables and table extensions with `OnValidate` triggers, and any procedure that raises errors or orchestrates a batch over records.
 - The changed procedures and triggers, weighted toward `OnValidate`/`OnInsert`/`OnModify` triggers, posting and validation routines, and procedures attributed with `[ErrorBehavior(...)]` or `[TryFunction]`.
 - Tokens extracted from the diff that relate to error surfacing and diagnostics (`Error`, `ErrorInfo`, `FieldError`, `TestField`, `Title`, `Message`, `DetailedMessage`, `AddAction`, `AddNavigationAction`, `RecordId`, `PageNo`, `ErrorBehavior`, `Collect`, `HasCollectedErrors`, `GetCollectedErrors`, `ClearCollectedErrors`, `ErrorType`, `Internal`, `Client`, `TryFunction`, `GetLastErrorText`, Boolean assignment).
+- For the outbound HTTP call paths identified in Source, include `HttpClient`, `Get`, `Post`, `HttpResponseMessage`, response use, and caller failure handling (including `[TryFunction]` call sites) in keyword and topic matching.
 - Resolve changed standalone call targets; when the target declaration has `[TryFunction]`, worklist the ignored-return rule even if the declaration itself is unchanged. Only assignment and conditional use activate try semantics.
 
 A file enters the candidate worklist when its `keywords` intersect the extracted tokens or its topic (derived from the index entry's `path`, `title`, and `description`) matches a changed object type. Read an article's full file — its `## Best Practice` / `## Anti Pattern` bodies — only after it makes the worklist; candidate selection uses the index alone.

microsoft/skills/review/al-scm-review.md

@@ -77,6 +77,7 @@ paths; they select articles, not findings. Facts and exceptions stay in articles
 | Registered warehouse quantity/physical-adjustment synchronization, `"Directed Put-away and Pick"`, `"Adjustment Bin Code"`, `"Warehouse Adjustment"`, or `"Calculate Whse. Adjustment"` and the resulting item-journal posting | `reconcile-warehouse-adjustments-with-the-item-ledger` |
 | `"Transfer Header"`/`"Transfer Line"` shipment/receipt completion, transfer posting publishers, in-transit/document-link changes, or item-journal posting presented as transfer-order completion | `post-transfers-through-shipment-and-receipt-codeunits` |
 | `Inventory`, `CalcQtyAvailableToPromise`, or stock sums used in a dated supply/demand promise, including changed location/variant/date filters and source-demand context | `use-date-aware-availability-for-promising` |
+| Direct assignment to `Quantity`, `"Unit of Measure Code"`, `"Qty. per Unit of Measure"`, or a `(Base)` quantity field on a persisted or posted item journal, sales, purchase, or transfer line, or a line quantity compared with a base-unit inventory value | `derive-base-quantities-through-the-line-unit-of-measure` |
 | `"Requisition Line"` action-message execution, accepted planning suggestions, `"Req. Wksh.-Make Order"`, `CarryOutBatchAction`, or linked supply creation/change plus requisition-line deletion | `carry-out-requisition-actions-through-the-standard-workflow` |
 
 Route clean supported calls through the same cues, not just suspicious writes.

microsoft/skills/review/al-security-review.md

@@ -39,7 +39,7 @@ Narrow the relevant files to the subset that applies to the changes under review
 
 - The changed AL object names and types — especially permission sets, codeunits handling authentication or authorization, objects touching `Isolated Storage`, `OAuth2` flows, web service endpoints, API pages, event publishers, and RecordRef helpers.
 - The changed procedures and triggers, weighted toward those that call `HttpClient`, validate or compose URLs, write to telemetry, read or write secrets, unwrap SecretText, manipulate record-level security, expose var Boolean guard parameters, or bypass the permission model (for example, `RecordRef.Open`, `Record.WritePermission`, direct table access from a non-owning app).
-- Tokens extracted from the diff that relate to security concerns (`IsolatedStorage`, `SetEncrypted`, `OAuth2`, `SecretText`, `Unwrap`, `NonDebuggable`, `Password`, `Token`, `HttpClient`, `Uri`, `AreURIsHaveSameHost`, `IsValidURIPattern`, `RecordRef`, `RecordId`, `TransferFields`, `Codeunit.Run`, `Access = Internal`, `internalsVisibleTo`, `Open`, `IntegrationEvent`, `SkipValidation`, `HasAccess`, `Permission`, `UserSecurityId`, `Commit`).
+- Tokens extracted from the diff that relate to security concerns (`IsolatedStorage`, `SetEncrypted`, `OAuth2`, `SecretText`, `Unwrap`, `NonDebuggable`, `Password`, `Token`, `HttpClient`, `Uri`, `AreURIsHaveSameHost`, `IsValidURIPattern`, `RecordRef`, `RecordId`, `TransferFields`, `Codeunit.Run`, `Access = Internal`, `internalsVisibleTo`, `Open`, `IntegrationEvent`, `SkipValidation`, `HasAccess`, `Permission`, `UserSecurityId`, `Commit`, `SetFilter`).
 
 A file enters the candidate worklist when its `keywords` intersect the extracted tokens or its topic (derived from the index entry's `path`, `title`, and `description`) matches a changed object type. Read an article's full file — its `## Best Practice` / `## Anti Pattern` bodies — only after it makes the worklist; candidate selection uses the index alone.
 

microsoft/skills/review/al-testing-review.md

@@ -39,7 +39,7 @@ Narrow the relevant files to the subset that applies to the changes under review
 
 - The changed AL object names and types — especially codeunits with `Subtype = Test`, test runner codeunits with `TestIsolation`, test libraries, and codeunits that define UI handlers.
 - The changed methods and attributes, weighted toward `[Test]`, `[TransactionModel(...)]`, `[TestPermissions(...)]`, `[HandlerFunctions(...)]`, handler attributes, `asserterror`, `ExpectedError`, `ExpectedErrorCode`, fixture initialization, and test-library calls.
-- Tokens extracted from the diff that relate to testing (`Subtype = Test`, `Subtype = TestRunner`, `TestIsolation`, `TestPermissions`, `Restrictive`, `NonRestrictive`, `Disabled`, `Permissions Mock`, `Library - Lower Permissions`, `TransactionModel`, `AutoRollback`, `AutoCommit`, `Commit`, `asserterror`, `ExpectedError`, `ExpectedErrorCode`, `HandlerFunctions`, `ConfirmHandler`, `MessageHandler`, `StrMenuHandler`, `ModalPageHandler`, `SendNotificationHandler`, `RecallNotificationHandler`, `Enqueue`, `Dequeue`, `AssertEmpty`, `Library Assert`, `LibraryVariableStorage`, `LibrarySales`, `LibraryPurchase`, `LibraryERM`, `LibraryInventory`, `LibraryRandom`, `Library - Utility`, `LibraryUtility`, `GenerateGUID`, `GenerateRandomCode`, `TestPage`, `.Visible(`, `.Enabled(`, `.Editable(`, `OpenNew`, `OpenView`, `OpenEdit`, `Init`, `Insert`).
+- Tokens extracted from the diff that relate to testing (`Subtype = Test`, `Subtype = TestRunner`, `TestIsolation`, `TestPermissions`, `Restrictive`, `NonRestrictive`, `Disabled`, `Permissions Mock`, `Library - Lower Permissions`, `TransactionModel`, `AutoRollback`, `AutoCommit`, `Commit`, `asserterror`, `ExpectedError`, `ExpectedErrorCode`, `HandlerFunctions`, `ConfirmHandler`, `MessageHandler`, `StrMenuHandler`, `ModalPageHandler`, `SendNotificationHandler`, `RecallNotificationHandler`, `Enqueue`, `Dequeue`, `AssertEmpty`, `Initialize`, `IsInitialized`, `OnTestInitialize`, `LibrarySetupStorage`, `Library Assert`, `LibraryVariableStorage`, `LibrarySales`, `LibraryPurchase`, `LibraryERM`, `LibraryInventory`, `LibraryRandom`, `Library - Utility`, `LibraryUtility`, `GenerateGUID`, `GenerateRandomCode`, `TestPage`, `.Visible(`, `.Enabled(`, `.Editable(`, `OpenNew`, `OpenView`, `OpenEdit`, `Init`, `Insert`).
 
 A file enters the candidate worklist when its `keywords` intersect the extracted tokens or its topic (derived from the index entry's `path`, `title`, and `description`) matches a changed object type. Read an article's full file — its `## Best Practice` / `## Anti Pattern` bodies — only after it makes the worklist; candidate selection uses the index alone. When the diff contains no testing-related changes by any of the above signals, return `outcome: "not-applicable"` without evaluating files.
 
@@ -49,6 +49,7 @@ The following targeted checks cover every current `testing` article. Treat each
 - An `AutoCommit` test runs under a `Subtype = TestRunner` codeunit that omits `TestIsolation` or sets it to `Disabled`, leaving committed data between tests — `testisolation-belongs-on-the-test-runner`. Require runner/repository context; a standalone test file cannot prove which runner executes it.
 - A permission-sensitive test uses `TestPermissions = Disabled`, claims to test a restricted user without `"Permissions Mock"`/`"Library - Lower Permissions"`, or declares `[TestPermissions(...)]` without applying that context — `permission-tests-must-lower-the-execution-context`.
 - Test fixture code manually calls `Init`/`Insert`, invents keys or prerequisite records, or bypasses available `LibrarySales`, `LibraryPurchase`, `LibraryERM`, `LibraryInventory`, `LibraryRandom`, or equivalent library codeunits — `use-library-codeunits-for-test-fixtures`.
+- A test codeunit's `Initialize` procedure exits on `IsInitialized` before per-test reset such as `LibraryVariableStorage.Clear`, `LibrarySetupStorage.Restore`, or `LibraryTestInitialize.OnTestInitialize`, or a `[Test]` method in a codeunit using that pattern does not call `Initialize()` first — `reset-per-test-state-before-the-isinitialized-guard`.
 - `asserterror` is added or changed without a following `Assert.ExpectedError`, `Assert.ExpectedErrorCode`, or a purpose-built assertion such as `ExpectedTestFieldError` — `asserterror-needs-expectederror-and-code`.
 - Test fixture code assigns a hardcoded literal to a primary-key field or a field the test relies on as a unique lookup identifier, hand-builds a "unique" value for such a field (string concatenation, a counter, `Format(CurrentDateTime)`), or truncates `LibraryUtility.GenerateGUID()`'s result with `CopyStr` for such a field shorter than 10 characters — `use-generateguid-for-unique-test-fixture-values`. Calling `GenerateGUID()` untruncated into a full-length field, `GenerateRandomCodeWithLength` for a shorter field needing real verified uniqueness, or `GenerateRandomCode20` specifically for a `Code[20]` field, is the compliant shape, not the signal to flag. `GenerateRandomCode20` is not a substitute for `GenerateRandomCodeWithLength` on a shorter field — it truncates `GenerateGUID()`'s sequential value down to the field's length by keeping the *leftmost* characters, which change the slowest, so retries against a short field can churn through the same truncated prefix far longer than `GenerateRandomCodeWithLength`'s equivalent. A hardcoded or deterministic value in an ordinary descriptive field is not this anti-pattern — that field carries no uniqueness constraint. Do not claim `GenerateRandomCode` (without `WithLength`/`20`) or `GenerateRandomXMLText` verify uniqueness against the real table, or that `GenerateRandomCode` is collision-free even within one test run for a short field — none of that is true.
 - A test asserts against a `TestPage` field's `.Visible()` or `.Enabled()` — `use-testpage-visible-enabled-to-verify-field-ui-state`. When the assertion is against `.Editable()`, or the page is opened with `OpenEdit()` specifically to check editability — `use-testpage-editable-to-verify-field-editability`.

microsoft/skills/review/al-web-services-review.md

@@ -3,7 +3,7 @@ kind: action-skill
 id: al-web-services-review
 version: 1
 title: AL web services review
-description: Reviews AL API surfaces and webhook integration handlers against web-services guidance from BCQuality.
+description: Reviews AL API surfaces, outbound HTTP integrations, and webhook handlers against web-services guidance from BCQuality.
 inputs: [pr-diff, file-path, folder-path]
 outputs: [findings-report]
 bc-version: [all]
@@ -14,7 +14,7 @@ application-area: [all]
 
 # AL web services review
 
-Reviews AL source changes against the `web-services` knowledge domain in BCQuality and emits a findings report. This is a leaf action skill: it invokes no sub-skills. It is one of the skills composed by `al-code-review`.
+Reviews AL source changes against the `web-services` knowledge domain in BCQuality and emits a findings report. This includes inbound API surfaces, outbound HTTP integrations, and webhook lifecycle code. This is a leaf action skill: it invokes no sub-skills. It is one of the skills composed by `al-code-review`.
 
 An orchestrator invokes this skill with a `pr-diff`, `file-path`, or `folder-path`. The skill produces a single JSON document conforming to the DO output contract.
 
@@ -39,10 +39,12 @@ Narrow the relevant files to the subset that applies to the changes under review
 
 - The changed AL object names and types — especially pages declared with `PageType = API`, API page `part` controls, queries declared with `QueryType = API`, and procedures that expose bound actions.
 - The changed properties and triggers, weighted toward API page metadata (`APIPublisher`, `APIGroup`, `APIVersion`, `EntityName`, `EntitySetName`, `ODataKeyFields`, `SourceTable`, `SourceTableTemporary`), navigation metadata (`SubPageLink`, `Multiplicity`, and visible singleton or collection semantics), CRUD guards (`InsertAllowed`, `ModifyAllowed`, `DeleteAllowed`, `Editable`), the `OnOpenPage` trigger, and `OnValidate` triggers on exposed fields.
+- Outbound HTTP integration code that constructs or sends requests, captures a client method's optional Boolean result, checks an `HttpResponseMessage`, or reads and parses response content.
+- Integration code that converts values to or from exchanged text: `Format` or `Evaluate` on a request URL, body, or file line, and `JsonToken`/`JsonValue` conversions of payload properties.
 - Webhook subscriber handlers and subscription lifecycle code, especially code that creates or renews subscriptions, handles `validationToken`, schedules from `expirationDateTime`, or targets resources whose eligibility is visible in the diff.
 - An API page (`PageType = API`) with `InsertAllowed = true` lists a field in `ODataKeyFields` and that same field control sets `Editable = false` — `api-page-key-fields-must-be-editable-on-insert.md`. A system-generated key such as `SystemId` marked read-only is not this anti-pattern.
 - An API page exposes a field via `Rec` directly, and that field is a stored value computed from other fields inside an `OnValidate` trigger rather than a FlowField recalculated in `OnAfterGetRecord` — `stored-derived-fields-must-not-be-exposed-directly.md`. Exposing a genuine FlowField, or a value already recalculated in `OnAfterGetRecord`, is not this anti-pattern.
-- Tokens extracted from the diff that relate to API surface and behaviour (`PageType`, `QueryType`, `API`, `api-page`, `page-part`, `APIPublisher`, `APIGroup`, `APIVersion`, `EntityName`, `EntitySetName`, `ODataKeyFields`, `SystemId`, `SubPageLink`, `subpagelink`, `Multiplicity`, `multiplicity`, `Many`, `ZeroOrOne`, `SourceTableTemporary`, `Job Queue Entry`, `webhook`, `webhookSupportedResources`, `webhook-supported-resources`, `subscriptions`, `notificationUrl`, `validationToken`, `validationtoken`, `expirationDateTime`, `expirationdatetime`, `ServiceEnabled`, `WebServiceActionContext`, `SetActionResponse`, `ReadIsolation`, `IsolationLevel`, `ReadCommitted`, `InsertAllowed`, `ModifyAllowed`, `DeleteAllowed`, `Editable`, `SourceTable`).
+- Tokens extracted from the diff that relate to API surface and behaviour (`PageType`, `QueryType`, `API`, `api-page`, `page-part`, `APIPublisher`, `APIGroup`, `APIVersion`, `EntityName`, `EntitySetName`, `ODataKeyFields`, `SystemId`, `SubPageLink`, `subpagelink`, `Multiplicity`, `multiplicity`, `Many`, `ZeroOrOne`, `SourceTableTemporary`, `Job Queue Entry`, `webhook`, `webhookSupportedResources`, `webhook-supported-resources`, `subscriptions`, `notificationUrl`, `validationToken`, `validationtoken`, `expirationDateTime`, `expirationdatetime`, `ServiceEnabled`, `WebServiceActionContext`, `SetActionResponse`, `ReadIsolation`, `IsolationLevel`, `ReadCommitted`, `InsertAllowed`, `ModifyAllowed`, `DeleteAllowed`, `Editable`, `SourceTable`, `HttpClient`, `HttpRequestMessage`, `HttpResponseMessage`, `Get`, `Post`, `Put`, `Delete`, `Send`, `IsSuccessStatusCode`, `HttpStatusCode`, `Content`, `ReadAs`, `JsonObject`, `JsonToken`, `JsonValue`, `AsValue`, `IsNull`, `SelectToken`, `Format`, `Evaluate`, `XmlDocument`).
 
 A file enters the candidate worklist when its `keywords` intersect the extracted tokens or its topic (derived from the index entry's `path`, `title`, and `description`) matches a changed object type. Read an article's full file — its `## Best Practice` / `## Anti Pattern` bodies — only after it makes the worklist; candidate selection uses the index alone.
 
@@ -58,7 +60,14 @@ For each worklist entry, evaluate the diff against the file's `## Best Practice`
 - When the diff contains code that contradicts a Best Practice without being a full anti-pattern, emit `minor` with the same reference shape.
 - Applicability alone is not a finding. Emit `info` only for a concrete, non-actionable observation the article explicitly defines; otherwise emit nothing when no violation is present.
 
-For API parts whose parent declares `ODataKeyFields = SystemId`, detect a child foreign key linked to a parent business field instead of `Field(SystemId)`. Do not apply the SystemId-link rule to APIs intentionally keyed by another field. Omitted `Multiplicity` is valid and means the documented default 1:N collection; never report omission alone. Report an explicit `ZeroOrOne` only when the visible contract clearly intends a collection or deep insert, and report an explicit `Many` only when it clearly intends a singleton. Singleton metadata requires an explicit `ZeroOrOne`; do not infer singleton intent from naming alone. For webhook eligibility, detect `QueryType = API`, `SourceTableTemporary = true`, composite `ODataKeyFields` (including an omitted property when a visible source primary key is composite), Job Queue Entry, and visible system-table sources; do not infer an unknown table number. For lifecycle code, require both create and renew paths to use a handler that returns the query-string `validationToken` verbatim with `200 OK`, and flag renewal scheduling that assumes subscriptions are permanent instead of using `expirationDateTime`. Do not emit generic HTTP or REST advice.
+For API parts whose parent declares `ODataKeyFields = SystemId`, detect a child foreign key linked to a parent business field instead of `Field(SystemId)`. Do not apply the SystemId-link rule to APIs intentionally keyed by another field. Omitted `Multiplicity` is valid and means the documented default 1:N collection; never report omission alone. Report an explicit `ZeroOrOne` only when the visible contract clearly intends a collection or deep insert, and report an explicit `Many` only when it clearly intends a singleton. Singleton metadata requires an explicit `ZeroOrOne`; do not infer singleton intent from naming alone. For webhook eligibility, detect `QueryType = API`, `SourceTableTemporary = true`, composite `ODataKeyFields` (including an omitted property when a visible source primary key is composite), Job Queue Entry, and visible system-table sources; do not infer an unknown table number. For lifecycle code, require both create and renew paths to use a handler that returns the query-string `validationToken` verbatim with `200 OK`, and flag renewal scheduling that assumes subscriptions are permanent instead of using `expirationDateTime`.
+
+For outbound HTTP calls, apply these targeted checks:
+
+- When code captures the optional Boolean result of `HttpClient.Get`, `Post`, `Put`, `Delete`, or `Send`, require the failed branch to stop before status, headers, or content are accessed. Do not report a call that omits the Boolean result: that form intentionally lets the runtime raise an error when the request cannot execute.
+- After platform success, require `IsSuccessStatusCode()` or an explicit acceptable `HttpStatusCode()` check before response content is interpreted as a success payload. Do not report code that reads a bounded non-success body solely for diagnostics and keeps it out of the success parsing path.
+
+Do not emit generic HTTP or REST advice beyond applicable knowledge articles.
 
 Set `confidence` to:
 
@@ -66,7 +75,7 @@ Set `confidence` to:
 - `medium` when detection relies on heuristics or when any frontmatter dimension was `unknown`.
 - `low` when the finding is an advisory derived only from applicability.
 
-After evaluating each worklist entry, also consider whether the diff exhibits a web-services defect the agent recognises from its general AL knowledge that no knowledge file in the worklist covers. Such candidates are agent findings within this skill's domain — emit them with `references: []`, an `id` slug prefixed with `agent:`, `confidence` capped at `medium`, `severity` capped at `minor` (agent findings are advisory and non-gating), and a `message` that is self-contained (describing both the issue and a concrete recommendation, since there is no knowledge-file footer for the consumer to fall back on). Hold every candidate to the precision bar in `skills/do.md` (*Agent findings*): emit only a concrete, material web-services defect a knowledgeable BC reviewer would agree is wrong — steelman it first and drop anything stylistic, speculative, dependent on code outside the diff, or merely a valid alternative; when in doubt, omit. The scope is strictly API pages and web-service surfaces; defects outside this domain belong to other leaves and MUST NOT be emitted here. Before emitting, check the worklist for a knowledge file that matches the candidate — if one exists, upgrade the candidate to a knowledge-backed finding instead. See `skills/do.md` for the full contract.
+After evaluating each worklist entry, also consider whether the diff exhibits a web-services defect the agent recognises from its general AL knowledge that no knowledge file in the worklist covers. Such candidates are agent findings within this skill's domain — emit them with `references: []`, an `id` slug prefixed with `agent:`, `confidence` capped at `medium`, `severity` capped at `minor` (agent findings are advisory and non-gating), and a `message` that is self-contained (describing both the issue and a concrete recommendation, since there is no knowledge-file footer for the consumer to fall back on). Hold every candidate to the precision bar in `skills/do.md` (*Agent findings*): emit only a concrete, material web-services defect a knowledgeable BC reviewer would agree is wrong — steelman it first and drop anything stylistic, speculative, dependent on code outside the diff, or merely a valid alternative; when in doubt, omit. The scope is strictly API pages, outbound HTTP integrations, and other web-service surfaces; defects outside this domain belong to other leaves and MUST NOT be emitted here. Before emitting, check the worklist for a knowledge file that matches the candidate — if one exists, upgrade the candidate to a knowledge-backed finding instead. See `skills/do.md` for the full contract.
 
 For every emitted finding, decide whether the fix is mechanical. A fix is mechanical when it is small, local, and unambiguous from the diff context (for example: set `ODataKeyFields = SystemId`; add the three `*Allowed = false` guards to a read-only page; add the missing `OnOpenPage` isolation assignment). For mechanical findings, emit `findings[].suggested-code` with the literal replacement for the source lines indicated by `location`. The payload must be a verbatim replacement — no diff markers, no fences, no commentary — that the consumer can render as a one-click suggestion. When a `.good.al` companion exists and the diff context matches the `.bad.al` shape, adapt the `.good.al` replacement into `suggested-code`.
 
@@ -76,7 +85,7 @@ Outcome selection:
 
 - `completed` — the skill evaluated every worklist item; default when the skill finishes normally, including when the resulting `findings` array is empty.
 - `no-knowledge` — no applicable web-services knowledge survived Source, Relevance, configuration filtering, and conflict resolution. `findings` is empty.
-- `not-applicable` — the task context contains no AL API surface, JavaScript webhook subscription lifecycle code, or JavaScript notification handler, or the `technologies` filter rejected the task.
+- `not-applicable` — the task context contains no AL API surface, outbound AL HTTP integration, JavaScript webhook subscription lifecycle code, or JavaScript notification handler, or the `technologies` filter rejected the task.
 - `partial` — a time or token budget was hit before the worklist was exhausted. `summary.coverage` reflects the evaluated subset; `outcome-reason` explains the cause.
 - `failed` — an unrecoverable error occurred. `outcome-reason` is required.