Module : Agents

Issue	Type	TrialGrid Version	Description
5380	Bug	91	The Bulk Edit Check Creator's created_high_confidence hook now exports "created" and "updated" as True/False rather than 1/0, so an action precondition such as agent("created") or agent("updated") runs the action instead of being rejected as not returning True or False.
5381	Bug	91	An Action precondition that can result in something other than True or False is now rejected when the Action is saved, instead of saving without error and being skipped every time it runs. For Agent actions, the agent configuration page and the Action editor also check the precondition against the variables of each hook the action is selected for.
5382	Feature	91	Every Bulk Edit Check Creator action hook now exports an "outcome" variable naming the hook that fired, so one action can be selected under several hooks, for example with the precondition agent("outcome") in ["custom_function_required", "build_failed"].
5383	Bug	91	When an agent hook action cannot run because its precondition is invalid, the agent's results now show it in a warning with the reason, instead of counting it among skipped actions in a collapsed detail list.
5290	Bug	91	The URL settings export now includes each Custom Property's Send to Agents setting and Agent Instruction, and the object type's own Agent Instruction, so an export shows which property values are sent to an AI agent.
5289	Feature	91	Object Definitions have an Agent Instructions tab showing what the AI agents are told about that object type — the type's own instruction, and for each of its properties whether the value is sent and what guidance goes with it. Editing needs the Manage Agent Configuration permission; anyone who can see the URL can read it.
5290	Bug	91	AI agents no longer send every populated custom property to the model. Each Custom Property now carries a "Send to agents?" setting which is off until you turn it on, so a value reaches the model only once someone has decided it belongs in the specification. Values recording who tested or reviewed an object are no longer sent. Existing object types need their specification properties marked as sent on the Object Definition's Agent Instructions tab before an agent will build from them.
5324	Feature	91	Custom Properties can carry their own agent instruction, sent alongside the property's value so the model is told what an individual property means rather than relying on one instruction for the whole object type.
5322	Bug	90.1	Asking the Edit Check agent for a check on a different form after a specification has been loaded now starts that check on its own, instead of offering it under the name of the check the specification described with overwrite pre-selected.
5340	Bug	90.1	A note explaining that something named in a clarification could not be found in the draft is now shown only on the turn that produced it, instead of standing above later summaries that had nothing to do with it.
5304	Feature	90.1	A clarification saying a form must not be wildcarded is now understood; the agent asks which folder it belongs to instead of using any folder.
5298	Bug	90.1	An agent no longer asks which field to use on a form that a check only adds, rather than reads.
5299	Bug	90.1	A generated check is no longer flagged as not matching its specification because a form was named in words rather than by its OID.
5300	Bug	90.1	Re-wording a clarification the agent did not understand no longer discards the specification it was building from.
5301	Bug	90.1	When a clarification cannot be applied, the agent now shows examples using the objects it found.
5302	Bug	90.1	A clarification the agent cannot use now keeps the objects it found on screen instead of reporting a problem reading the specification.
5303	Bug	90.1	After a clarification that could not be applied, the agent asks for a rephrase rather than re-offering the confirm buttons.
5306	Bug	90.1	Putting a form in a folder no longer removes another form already assigned to that folder.
5307	Bug	90.1	Removing a field or a folder in a clarification now affects only the form named, not every form.
5308	Bug	90.1	A correction typed while the agent is asking which field to use is now applied instead of being searched for as a field name.
5309	Bug	90.1	The message box is disabled while the agent is waiting for a confirm choice, so a typed reply cannot discard the loaded specification.
5310	Bug	90.1	Naming a folder in a clarification no longer also adds a form of the same name to the check.
5311	Bug	90.1	A clarification that does not name a form now applies to every form found, and a follow-up such as "for both" is understood as continuing the previous correction.
5313	Bug	90.1	A folder you name yourself is now flagged for review rather than blocking the save, and the note no longer claims the specification named a different one.
5314	Bug	90.1	When an edit cannot be applied the agent now says the check is unchanged, and explains why where it can.
5315	Bug	90.1	Naming a folder that does not exist no longer removes the folder a form already had, and the agent says the folder was not found.
5316	Bug	90.1	A generated custom function can now be changed by describing the change, as an edit check already could.
5317	Bug	90.1	Answers about a generated custom function now read the query flag correctly, instead of describing a guard clause that closes a query as one that opens it.
5318	Bug	90.1	Generated custom functions keep a consistent shape when edited, rather than alternating between two structures.
5278	Bug	90.1	Checks generated against a checkbox field now compare it to 1 or 0, and a comparison against any other value is rejected before the check can be saved.
5279	Bug	90.1	The Edit Check Agent no longer asks which field to use on a form that a check adds, since Add Form and Add Matrix actions operate on the whole form.
5280	Bug	90.1	Agents now resolve field references written as FORM.FIELD or FOLDER.FORM.FIELD directly, without asking which form or field was meant and without losing the folder.
5281	Bug	90.1	Choosing 'None of these' on an agent's field clarification no longer removes the option to generate; you can now type a correction or generate with the objects already found.
5282	Bug	90.1	Clarifying an agent's discovered objects now keeps the check's specification and can move a form to a named folder, instead of being ignored and replacing the specification with your correction.
5283	Bug	90.1	When an agent picks a coded value your request did not name, the warning now shows both values with their labels and keeps the value you asked for visible in the list.
5284	Feature	90.1	Checks can now compare a form's field across two folders, with each reference scoped to the folder the specification names.
5285	Bug	90.1	Agents can now build checks that test the value as entered, so an unknown date component named in the specification no longer forces a custom function.
5286	Bug	90.1	The Custom Function Agent now builds from an Edit Check's or Custom Object's specification when started from its metadata, instead of reporting that no forms or fields could be found.
5287	Bug	90.1	When the Edit Check Agent asks you to type a field name for a form with too many fields to list, your answer is now resolved against that form instead of being ignored.
5288	Bug	90.1	An agent no longer binds a form to a folder the form does not belong to, and says so when the folders a specification names do not match where the form lives.
5291	Bug	90.1	Clicking an agent option that is no longer available now says so and keeps the check's specification, instead of being read as a new description.
5178	Feature	90	AI usage is now limited by cost rather than tokens. Each organization has a monthly AI budget and a daily per-user cost limit, and the AI usage report shows the cost of each day's usage.
5114	Bug	90	Tooltips on datapoint widgets in the Test Case editor and on the LLM agent session list no longer raise a background error when the surrounding content refreshes while a tooltip is appearing.
5060	Bug	89	The Test Case agent's "Create from this check" action now works for Edit Checks which use a Custom Function or Set Dynamic Search List action, where it previously offered an empty test case. If generation fails the agent now reports an error instead of offering an empty test case.
5046	Bug	89	Custom Function Review results are now shown with a warning when finding validation could not run, so it is clear that the findings have not been verified.
4999	Feature	89	The Edit Check and Test Case agents are better at recognising specifications that need a custom function. Conditions on a single date component (such as "the month is unknown but the day is entered") and conditions on a form's status (such as "the form is not submitted") are now flagged as needing a custom function instead of being turned into a comparison that doesn't match the specification. After generating a check the agent now also verifies that the logic actually matches the specification and stops rather than offering a check that doesn't.
5000	Feature	89	A new bulk Edit Check Creator agent builds edit checks from specifications in bulk — either the metadata attached to existing empty checks (which it fills in) or custom specification objects (from which it creates new checks). Each specification's outcome is classified — created with high confidence, created but review required, custom function required, or failed — and drives configurable per-outcome action hooks. Each result is shown in the chat as soon as it is built, so progress is visible throughout the run. Only specifications that carry values, on an object type with an LLM instruction configured, are built; anything else is listed with the reason it was skipped.
4991	Feature	89	The Edit Check and Test Case agents now guard range comparisons with an IsNotEmpty step.
4992	Feature	89	When the Edit Check Agent opens on a check that has a specification but no logic yet, it now offers to build that check's logic from its specification and writes the result back onto the same check. Checks that already have logic offer to regenerate from the specification or edit the existing logic, and the "metadata" wording is now "specification" throughout. When working on a specific check the generated logic saves straight onto that check with one click ("Save to <name>"), with a "Save as new check" option to spin the same logic into a separate check.
4993	Feature	89	The Edit Check and Test Case agents now see the URL's configured marking groups when generating a query action, so they pick a real marking group by its exact name instead of inventing one. A new per-URL default marking group can be set by TrialGrid admins.
4994	Feature	89	The Edit Check and Test Case agents now recognise when a specification can only be satisfied by a custom function (for example regex matching, extracting date components, or aggregating across records) — logic that Rave edit check steps cannot express. In that case the agent explains that a custom function is required and stops instead of producing an invalid edit check.
4997	Feature	89	When a specification says its behaviour is handled by a named custom function, the Edit Check and Test Case agents now build the check correctly — a trigger step on the field (its .DataPoint.IsPresent) plus a single CustomFunction action that calls the function, with no separate Open Query. If the named custom function isn't in the draft yet, the check is still offered with a warning rather than being blocked.
4998	Feature	89	Improved Agents ability to recover after an unexpected failure.
4986	Bug	89	The Edit Check and Test Case agents now recognise fields that have a data dictionary and compare them using the coded value (the .CodedValue suffix). A value that is not in the field's dictionary blocks the save. When the value is valid but the request never clearly named it, the agent shows a "did you mean" suggestion but still lets you save. Editing an existing check no longer re-flags a coded value it already contained.
4987	Bug	89	The Test Case Agent now writes date values using the separators of the field's own Rave format (for example "15 MAR 2024" for a "dd MMM yyyy" field) instead of a fixed hyphenated format, so generated test cases no longer raise a "does not match the field format" warning.
4988	Bug	89	The Custom Function Agent now knows how to change field visibility (setting the DataPoint's IsVisible property) and generates the show/hide code when asked to make fields visible or hidden, instead of only clearing the field values.
4989	Feature	89	When a user explicitly asks the Custom Function Agent for a function using Medidata's new V2 custom function API, the agent now explains that only the V1 API is supported so far and asks them to describe the function using V1, instead of generating V1 code for a V2 request.
4970	Feature	89	Organization administrators can enable or disable Agents for individual team members from the team user edit page; the option is shown only when the organization has Agents enabled.
4929	Feature	89	The Custom Function AI assistant now recognises the action a function performs (merge/un-merge a matrix, add a form, derive a value, and more) and generates the correct action instead of always raising a query.
4844	Feature	88	The Test Case creator agent now supports selection of one or more folders.
4841	Bug	88	The Test Case Creator agent was incorrectly requiring a field to be chosen for the AddForm check action. This has been corrected.
4779	System	88	The Help agent and development Test agent have been migrated to the new orchestrator framework.
4811	System	87	All agents converted to the new orchestrator framework.
4781	Feature	87	The Test Case search agent has been migrated to the new orchestrator framework.
4760	Feature	87	The Test Case Creator agents have been migrated to the new orchestrator framework.
4763	Feature	87	The Custom Function creator agent has been migrated to the new orchestrator framework.
4757	Feature	87	The Custom Function Review agent has been migrated to the new orchestrator framework.
4756	Feature	87	The Field Visibility Advisor (Medidata) and Progressive Display Advisor (Veeva) agents have been updated for a more responsive and consistent experience. Any in-progress conversations with these agents will be cleared when this release is deployed; just start a fresh chat to continue.
4752	System	87	An Orchestration Framework has been introduced to help TrialGrid Ltd design and develop Agents.
4751	Bug	87	The Custom Diagnostic agent was creating a diagnostic when running tests; it should not have created a diagnostic at this point. This has been corrected.
4647	Feature	86	AI Summary document templates allow customers to create templates that render context data into an LLM prompt. The rendered prompt is sent to the AI for summarization, with the response saved as a text document.
4636	Bug	86	The Progressive Display Advisor agent now saves the Choice Code (not the Choice Label) when configuring Controlling Item Values on fields with code lists.
4637	Bug	86	The Medidata Agents were not always using the Coded Value for dictionary Fields. This has been corrected.
4638	Feature	86	The Medidata Test Case Agent can now use Edit Check custom properties.
4632	Feature	86	The Specification Review agent's object discovery has been improved.
4609	Feature	86	The Edit Check Agent can now generate CQL logic and actions from an edit check's metadata properties, allowing users to define specifications via Property Definitions and then use the agent to implement them.
4611	Feature	86	The Custom Function Agent can now generate custom functions from Custom Object properties or Edit Check metadata, allowing users to define specifications via Property Definitions and then use the agent to implement them.
4603	Feature	86	The Field Visibility Agent will run on the Forms selected in the Forms list.
4604	Feature	86	The Progressive Display Agent is now available from the Veeva forms list and runs on selected forms when invoked. If no forms are selected, it analyzes all forms.
4605	Feature	86	The Field Visibility Agent now displays analysis results progressively as each form is analyzed, rather than waiting for all forms to complete.
4606	Feature	86	The Progressive Display Agent now displays analysis results progressively as each form is analyzed, rather than waiting for all forms to complete.
4607	Feature	86	The Progressive Display Agent now offers a fix option to set Controlling Item and Controlling Item Values on items that are missing progressive display configuration.
4589	Feature	86	The 'Review' column in the Custom Function list has been removed. The Custom Function Review agent supports action hooks which can be used to add Labels or other actions to reviewed Custom Functions.
4585	Feature	86	Agents will now search for Folders by OID and Name.
4586	Feature	86	The Custom Function Review agent was producing too many low-quality findings and now focuses on errors.
4579	Feature	86	The Test Case Creator agent can also create Edit Checks at the same time as Test Cases.
4552	Bug	85	The Veeva and Medidata Test Case agents should have similar workflow and features.
4554	Feature	85	The Custom Diagnostic agent is now aware of TSDV configurations.
4543	Feature	85	There are Action Hooks for the Test Case creator agent.
4544	Feature	85	There are Action Hooks for the Edit Check creator agent.
4545	Feature	85	There are Agent Hooks for the Custom Function review agent.
4531	Feature	85	The Specification Review Agent can run on multiple specification objects.
4532	Feature	85	Agents now require user confirmation before running Actions.
4508	Feature	85	The Specification Review Agent will assess whether or not a Custom Function is required.
4486	Feature	85	Agents can be configured to run Actions.
4489	Feature	85	The Field Visibility agent is now available on the Form Review page.
4480	Feature	85	There is a URL Agent configuration page to control which objects Agents can operate on.
4466	Feature	85	The Test Case creator agent now lets users select multiple folders for the test scenarios.
4467	Feature	85	The Test Case creator agent now uses for loops for testing scenarios with multiple folders.
4468	Feature	85	The discovery of forms and fields by Agents has been improved.
4452	Feature	85	The Custom Function Review agent will respond to follow up questions after the initial review.
4445	Feature	85	There is an Agent to check the progressive display settings on Veeva forms.
4442	Feature	85	There is an Agent to review Specification custom objects.
4441	Feature	85	Changes made during an Agent session will be identified with the Agent name in Activity listings.
4423	Feature	85	There is an Agent to help create Custom Functions.
4409	Feature	85	There is an Agent to review Custom Function code and look for possible issues.
4371	Feature	85	There is an AI Agent to help create Edit Checks.
4375	Feature	84	There is a new URL/Vault level setting controlling whether AI Agents are enabled for that URL/Vault.
4370	Feature	83	When the test case creator agent creates a test case from an object it will include it as a related object.
4345	Feature	83	There is a Custom Diagnostic agent to help create and edit Custom Diagnostics.
4348	Feature	83	New Field Visibility Agent helps users assess whether form fields have appropriate visibility settings and corresponding edit checks to show/hide fields dynamically.
4343	Feature	83	There is a Test Case creator agent for Veeva drafts.
4331	Feature	83	The Test Case creator agent supports all Rave Check actions, except Balance and CTMS check actions.
4333	Feature	83	The Test Case creator agent is context aware, if opened from an Edit Check, Custom Function or Custom Object page it will have details of that object available to use.
4334	Feature	83	The Test Case creator agent can use the Test Case Advisor to create test cases from existing edit checks and custom functions.
4335	Feature	83	Object definitions have a property which can be used to provide a description of the properties for AI Agents. This property can only be set by TrialGrid Ltd staff.
4330	Feature	83	Organization settings page now displays the status of AI Features and AI Agents settings.
4329	Feature	83	There is a user confirmation step before Agents create, update or delete objects.
4324	Feature	83	A flag now controls whether users can access AI Agents.
4327	Feature	83	It is now possible for users to rate the performance of the agent in a particular session. Rating the agent performance will allow TrialGrid Ltd to gather metrics to improve agent performance.
4311	Feature	83	There is an AI agent to search Test Cases.
4310	Feature	83	There is an Agent to create and edit Test Cases.
4304	Feature	83	There is Help agent which will search TrialGrid documentation to answer questions.
4299	Feature	83	There is now a framework for introducing AI Agents into TrialGrid. AI Agents are not yet generally available.
4300	Feature	83	AI Agent sessions are stored and can be viewed.
4301	Feature	83	Previous AI Agent sessions can be soft-deleted.
4302	Feature	83	AI Agent sessions can be exported to PDF documents.