Work on ServiceNow without fighting the UI.

Each instance opens in its own browser window: your SSO, your second factor, the normal login page. vikstra keeps that session and makes every call through it. There is no API account to request, no OAuth app to register, nothing to install on the instance.You can do what your user can do, and nothing more.

The session stays alive the way the platform UI keeps it alive, and lets go after four idle hours or as soon as the session ends. Each instance gets its own tabs, its own index and an environment tag, and a tab opened on one instance can never save to another.

Business rules, script includes, client scripts, ACLs, scripted REST, transform maps, widgets, catalog items with their variables: 35 artifact types open in a Monaco editor. Completion comes from the instance’s live dictionary, so current. offers the fields that table really has, and your own Script Include methods show up next to the Glide API.

The validator runs on open and on save. It catches unknown tables and fields, Glide classes that do not exist, and GlideAjax mistakes, with quick fixes where a fix is safe. Before you save, the footer tells you what you are about to touch.

Lists open in the instance’s own views, with columns sized to what they hold and reference cells that open the record they point at. Forms follow the form layout configured for the view. The All fields view shows every field of the table and its parents, the raw value beside the display value, the table it was declared on, and which fields the form leaves out.

The record’s own UI actions run through the platform. The record explorer answers the questions you would otherwise answer about any record: what runs on save and in which order, why a field is mandatory or read-only, which ACLs decide whether this user can do this.

Why did this record change? vikstra joins the audit trail to the business rules that write each field and ranks them likely or possible. With a model configured, plain-language list filters resolve display values and show the row count before anything runs.

Run scripts with saved snippets and parameters, keep a scratchpad, read parsed output. When/sys.scripts.do is not available to you (disabled, role missing, a scoped session), run the script as a temporary Fix Script or a run-once Scheduled Script Execution instead. In auto mode vikstra switches to a Fix Script by itself when a scoped session is refused, and a script that ran and failed is never run twice.

Then it deletes that record and its sys_update_xml capture row. If the app stops halfway, it finishes the cleanup the next time you are signed in to that instance. If anything is left, you are told what and where. Every save and every run is checked in the instance logs afterwards.

Each entry in an update set opens with a real diff. The validator runs over every script in the payload, and a dependency check against the target instance runs before you promote. Collisions across open sets are flagged, and the capture audit finds what did not get captured: an ACL without its roles, a field without its table, a cross-scope reference.

After an upgrade, skipped records are sorted into three piles: only the out-of-box version changed, only you changed it, or both did. Mark them reviewed or revert to out-of-box.

vikstra keeps an encrypted index of every script body on each instance, with full-text search and a table of who references what. A where-used question becomes a local query that sends nothing to the instance.

On top of it: sweeps that rank anti-patterns by how far you exceed the out-of-box baseline, a security review of eight deterministic checks with evidence, an ACL matrix of users by operations, a dependency graph several hops deep with rename impact, and code rules you can write yourself.

A flow is drawn as a graph: trigger, steps, if and else side by side, loops with their next-item edge. Add a step on any edge, set data pills, publish with a backup. Run it and the path lights up as it executes. Compare two runs step by step to see the timing difference and the paths only one of them took.

For ATF, with a model configured, point vikstra at a business rule or script include and ask for tests. Each proposed case runs once. Only cases that pass and leave nothing behind are kept, and generated code cannot forge the verdict. Generation is refused on production.

Everything above works without a model, except plain-language list filters and ATF test generation. Add one and it reads the live instance through about 70 read-only tools: records, schema, logs, flows, ACLs, and whether a rule would run for a given record. Answers cite their sources, and a sys_id that no tool returned is flagged.

Scripts it writes are checked against the live dictionary, then every table, field and property they name is looked up on the instance. They are marked verified only when both checks ran. Larger changes arrive as a plan; your approval is bound to the plan’s hash, so the plan you read is the plan that runs.

A personal-data filter sits in front of every request, hosted or local, and cannot be switched off. If the detector is unavailable, the request is blocked rather than sent.

User story mode writes one ServiceNow story. It reads the instance and the docs, asks you what it needs, and can picture a form or list as you see it, with fields or buttons that do not exist yet drawn in. It always runs on a local model, keeps its pictures encrypted with the chat, and writes nothing to the instance. The story exports as a PDF.

They hold for the editor, the agent, your own automations and anything connected over MCP.