Boundaries and safety
The guarantees behind every part of Ysra — scope, isolation, credentials, approvals, credits, recovery, honest results and audit — and where each one stops.
On this page
Boundaries are the controls that keep Ysra’s work inside what you gave it: a folder, sources, permissions and credits. This page is for anyone who needs to know what Ysra will and won't do on its own, such as a team lead or a security reviewer. It says what each control guarantees, not how it is built. For Code on your computer, in the cloud and on your servers, see Security.
The principle#
Scope is set before work starts, and nothing leaves Ysra without the permission you gave. Anything that reaches outside — a message, an email, a push to the default branch, a deploy — shows the exact content or command first. Ysra never blocks an action you asked for explicitly: it says plainly what it ran to check the work, and does what you asked.
What each control guarantees#
| Control | Guarantee |
|---|---|
| Scope | The folder, sources, capabilities and permissions are fixed before work starts. Ysra works inside them and nothing wider. A Code thread's access mode sets what it may do without asking. |
| Isolated work | Code works only in the folder you opened, or on its own cloud machine. |
| Credentials | Held by the platform; never given to the model or put in the workspace beyond what one task needs. |
| Approvals | An action outside Ysra shows its exact content first; an approval covers that content and nothing else. |
| Credits | Work pauses before its next paid step when credits run out, and keeps everything. |
| Recovery | Progress is saved as work happens; an interruption resumes from the last saved point. |
| Clear end states | Every ending says in one line what Ysra ran and how it went; a check that didn't run is never reported as passed. |
| Audit | Each piece of work records what it did and what it used. |
| Uploads | Every file is checked before Ysra uses it; a file that fails isn't added. See Files you can upload. |
Where work can reach#
| Where | What it can reach |
|---|---|
| Code on your computer | Only the folder you opened; commands are contained to it. What always asks is set by the access mode. See Security & privacy. |
| Code in Ysra’s cloud | A copy of the repository on a machine of its own. Internet only to the code host and the project’s package registries. Secrets are for the setup script only, unless you allow one for Ysra’s commands. See Code in Ysra’s cloud. |
| Ask | Only the sources a conversation was given. Slack, Jira, Gmail and other tools are connected and selected by their owner. |
| A public colleague | Only the files chosen for it, and the web if allowed. Never your connected tools; it can’t send or change anything. See Public colleagues. |
| A routine | Only the actions listed on it; anything not listed is forbidden. See How routines work. |
| Capabilities (installing, network, destructive actions) | Allowed, asked about or denied by the autonomy policy. |
Repository delivery, deployment and third-party actions are separate permissions: being allowed to edit code is not being allowed to publish it. Every change is scanned for credentials and secrets before it can be delivered.
What happens when …#
An action would leave Ysra#
A Slack message, a Jira update, an email: the exact content is shown before it runs. Editing it means it waits for approval again. For routines, sending email and posting messages wait for approval of the exact text on every run; an approval nobody gives within 7 days lapses and the step isn’t done.
You ask to deploy or push#
Every deploy asks first, with the exact command. A push to the default branch asks every time; in Auto, a branch Ysra made is pushed without asking. A push is never forced. Ysra says what it ran to check the work, and still does what you asked.
Credits run out#
New paid work doesn’t start. Running work pauses before its next paid step, keeping the changes, the saved points and the evidence; nothing continues until there are credits again and you tell Ysra to continue. See When credits run out.
Work is interrupted#
It resumes from the last saved point instead of starting over; a task whose machine is lost is picked up again on a new one. Rewinding to an earlier saved point marks the restored version unchecked until it is checked again.
Something fails#
An infrastructure problem is never labelled a success, nor a defect in your code. Repeated failures stop the work with an explanation instead of looping. See How a Code thread works.
Audit#
Each piece of work records its events, tool activity, usage, files made and delivery, so you can see what happened and what it used. Reasoning is shown as the actions and observations that drove the work; internal model reasoning is not exposed.
Error reports#
When an app breaks (a crash, an uncaught error, a desktop app that stops responding), it sends the Ysra team a report of what failed and where. A report never contains your questions, answers, files, file names, email addresses or keys. Each app has a switch, on to start with; turning it off stops reports at once.
| App | Where |
|---|---|
| Web app | Settings → Send error reports, for this browser. See Settings |
| Desktop app | Settings → General → Send error reports. See Ysra Desktop settings |
| Phone | Settings → Send error reports. See Mobile app settings |
Limits of these controls#
| Control | What it doesn’t cover |
|---|---|
| The security scan | A baseline, not a penetration test or a substitute for your own review. |
| Browser proof | The behaviour that was exercised, not every path through the app. |
| Permissions | Ysra doesn’t mirror every permission in your organisation; it searches the sources you select for it. |
For security reviews
Bring your questionnaire. The Ysra team walks your team through data flows, isolation, credential handling and the audit trail for the workflows you plan to use, and says plainly where the boundaries are today.
Related#
- Security — Code on your computer, in the cloud, devices and machines.
- Data, privacy and deletion — what is kept, and how to delete it.
- Autonomy policy — what is allowed, asked about or denied.
- How routines work — approvals and caps for standing jobs.
- Limits and costs — credits and every cap.