Set rules without editing configuration
Choose a reviewed rule or write one in plain English. Set its agents, scope and strength in the interface, then inspect the result before applying it.
Free VISINVOKE utility
Set the rules once.
Make the task clear.
Review the evidence.

A clear working contract
Define how the coding agents work on your projects.
Choose a reviewed rule or write one in plain English. Set its agents, scope and strength in the interface, then inspect the result before applying it.
Record the project, read-first authorities, required work, forbidden scope and exact done conditions. Keep the checkpoint and existing work beside the active task.

A shared policy and separate agent workstreams
The illustration shows blue and purple workstreams passing through a shield and visible checkpoints into separate project spaces.
Choose Guidance, Must Check or Prevent When Possible. Review inherited rules, conflicts and the guard support available to each selected agent.
What you can control
Bring the working requirements into one visible control model.
Choose reviewed rules for repository safety, scope, existing work, testing, task discipline and completion. Read the plain-English explanation before applying a rule.
Write the behaviour you require normally. Select the agents, category, scope and enforcement strength in the visual interface.
Use Guidance for working preferences and Must Check for evidence at preflight or completion. Prevent When Possible uses supported agent guards and shows limitations.
Target Claude, Codex or both. Each selected agent receives instructions and checks through the mechanisms it supports.
Keep shared protections at global scope, then add project, profile or current-task rules. Review the origin of each resolved instruction.
Collect rules, authority locations, checkpoint expectations and supported agents in a profile. Review inherited and local rules together before applying them.
Identify the project, stage and task type. Record required authorities, existing work, prohibited changes and acceptance criteria before work starts.
Keep the job, project, read-first material, Do, Do Not and Done When in one durable task record. Continue from the checkpoint already reached.
Define required scoped checks, acceptance evidence, live proof and completion records. Report further work without silently expanding the task.
Review managed output and conflicts before apply. Keep rule, profile, task and apply history, with backup and restore for managed changes.

A deliberate browser-to-desktop handoff
The illustration shows a person reviewing a checklist between a browser conversation and a desktop workspace. A document connects the two screens.
Optional ChatGPT assistance uses a deliberate copy-and-paste handoff. Choose context, copy a prompt, paste the returned structure, validate it and review its effect before applying.
Before work starts
Give the agent a clear place to begin.
Record the product or project, active stage and task type. Describe a continuation, correction, bug fix, design change, validation or release.
Name the relevant authority, checkpoint and completion records. Keep the existing implementation and concurrent work visible.
Write what must happen, what must not change and what evidence counts as done. Keep unrelated work outside the task.
Inspect the resolved instruction sources and conflicts. A required check stays labelled as a check where the installed agent cannot enforce a technical gate.

Protected files and a retained proof record
The illustration shows locked folders, document review, code, a checklist and a shield leading to a retained record.
Define completion requirements and inspect agent status, generated rules and task records. Review missing evidence, unsupported controls or conflicts before applying a bounded correction.
Keep the control state inspectable
Retain the decisions behind your project instructions.
Use managed regions or generated files for Agent Control output. Review the target files and existing manual instructions before apply.
Create a backup before changing managed output. Restore the previous generated state when that is the result you need.
Review rule, profile, task and apply history. Keep the control state beside the checkpoint you continue from.
Build an optional browser prompt from selected facts. Review the prompt before copying it; credentials, secrets and file contents are not automatically included.
Your workspace and related tools
Open your desktop tools through your account.

Open Agent Control in DCC. Get the desktop workspace through your account installer page.

Turn your figures, notes or an existing deck into an editable presentation. Refine the slides and export PowerPoint.
Agent Control illustrations show a human-led command centre, separate workstreams, a reviewed browser handoff and retained evidence.
Questions and answers
Choose the requirements that suit your project.
A free VISINVOKE utility for setting rules, scope, preflight, task records and completion requirements for Claude and Codex without manually editing their configuration files.
Agent Control runs on Windows inside VISINVOKE Desktop Control Centre. Create an account or sign in and get DCC through your account installer page.
No. Guidance is instructional, Must Check requires evidence at a gate, and Prevent When Possible uses a guard only where the selected installed agent supports it. Review the indicated support and limitations.
Use global rules as a foundation, reusable profiles for a type of work, and project or current-task rules for specific requirements. Inspect the resolved rule sources and conflicts before apply.
No. Browser assistance is optional: choose the context, copy a prompt, paste the response back, validate and review it, then apply deliberately. Agent Control has no embedded model, API key or token billing.
Agent Control uses managed blocks or generated files, with review, backup and restore for managed output. Review the affected files and conflicts before applying changes.

Continue in VISINVOKE
Create an account or sign in. Get Desktop Control Centre through your account installer page and open Agent Control on Windows.
Agent Control is free with your VISINVOKE account. It has no embedded AI model, API key or token billing.