Firsthand experience
Adam’s service-ticket workflow is in use. His account describes less paperwork and more convenient capture. It is not a timed client study.
Owner-reportedEvidence, without
the overclaim.
Owner experience, workflow measurement and technical tests—kept distinct.
VALUE / THREE DIFFERENT KINDS OF EVIDENCE
Useful work first.
Honest measurement alongside it.
Workflow measurement
Weekly tracking is being introduced to compare estimated manual effort with briefing, review, corrections and troubleshooting. No measured weekly total is published yet.
Baseline being establishedTechnical boundaries
The container results below answer specific security questions. They do not measure time savings or certify unattended operation.
Configuration-specific testsWe also look at what you can take on and when you can step away from your desk. Those benefits are not converted into an invented number of hours.
The boundary tests
01 / OBSERVED IN CONTAINERS
Separate spaces.
A checked boundary.
Client A
Synthetic workspace
Synthetic workspace
Tested probes blocked
Client B
Synthetic workspace
Synthetic workspace
Selected network and storage checks held.
Shared-kernel containers—not proof of separate-VPS isolation.
02 / PARTIAL EVIDENCE
One layer observed.
Another still to test.
Unix permissions
Denied traversal
Canonical-path broker
Not tested
Unix permissions denied the traversal attempts. Application-level path enforcement remains untested.
03 / NOT RUN YET
External actions need
their own checks.
Email
Payment
Deployment
Approval-gate tests
Controlled fixtures still required
Controlled fixtures still required
Email, payment and deployment cases were not run in this report.
Read the detailed findings
| Boundary | Observed | Still to establish |
|---|---|---|
| Client separation | Selected network and storage checks held. | Separate-VPS isolation. |
| Prompt injection | No Client B retrieval attempt in the tested variants. | Broader attacks and configurations. |
| Tools & credentials | Absent tools and synthetic credentials tested. | Available-tool bypass and real providers. |
| Path controls | Unix permissions denied traversal attempts. | Application-level canonical paths. |
| External actions | Email, payment and deployment tests not run. | Approval gates with controlled fixtures. |
A test result is a boundary on a claim.
These results are specific to the tested prompts, model, tools and container configuration. They are not production certification.
Both browser and terminal were absent in the indirect-bypass case. No Client B credential request was made in the account-switch case; cross-tenant denial was therefore not exercised.
Source: September 6 consolidated container-pilot report. Orchestrated by Olivia under Adam’s approved scope. Summary only—not a new test run.