Red Team Findings and Schedules

Track red-team findings through to a fix, and re-run the same authorized assessment on a schedule.

Both live in the shared sub-navigation of the Agentic and Model Red Teaming tabs: Findings and Schedules. They are shared between the two tabs.

Findings tracker

The tracker collects findings from agent and model assessments so you can assign, schedule, and close them out.

Open a report

Runs → a completed run

Track findings

Copies its findings

into the tracker

Review

Status, assignee, due date

Reason for change

Verify

Re-run or guardrail verify

Then close

QuilrAI

Add findings

Click Track findings at the top of a report. It copies that report's findings into the tracker, recorded in the audit history as "Imported from assessment evidence".

note

Track findings writes to the tracker. It is not just a link.

The list

Column	Shows
Objective	The objective name, plus the target type and outcome (for example "http · vulnerable", "model · partial")
Severity	Critical, High, Medium, or Low
Workflow	The current workflow status
Assignee / Due	Owner and due date
Action	Review

Use Refresh findings to reload. The list is paged at up to 100 findings per page.

Review a finding

Click Review on a row.

1. Click Open source assessment to see the evidence in the original report.
2. Set Workflow status, Assignee, and Due date (UTC) as needed.
3. Enter a Reason for change (required). Record the decision here; keep credentials and sensitive evidence in the source assessment.
4. Click Save workflow.

Workflow status	Use it when
open	New, not yet looked at
triaged	Confirmed and prioritized
in progress	A fix is being made
awaiting verification	Fixed; waiting for a re-run to prove it
risk accepted	Not fixing, with a recorded reason
false positive	Not a real issue

Every save is kept in the Audit history below the form: an append-only list of revisions with actor, status, timestamp, and note (the latest 200 changes are shown).

warning

Workflow status does not certify a fix. Move a finding out of awaiting verification only after a re-run, or a guardrail verification, shows it held.

Schedules

Schedules repeat one authorized red-team assessment against the same target, so you can track it over time. Each recurrence appears in Runs with the scheduled time appended to the campaign name.

Click New schedule to open New recurring assessment.

1. Schedule details

Field	Default	Notes
Schedule name	Empty (required)
Start this schedule immediately	On	Inactive schedules keep their configuration but do not launch runs.

2. Cadence

Field	Default	Notes
Cadence	Weekly	Daily, Weekly, or Monthly
IANA time zone	Your browser's time zone	Scheduled times follow this zone, including daylight-saving changes.
Weekday	Monday	For weekly schedules
Hour / Minute	09 / 00

3. Assessment target

Every recurrence runs against one authorized HTTP agent.

Field	Notes
HTTP agent name	Required
HTTP method	GET or POST (default POST). Only these two are authorized.
Agent endpoint URL	Absolute HTTPS URL only. Credentials, query strings, and fragments are not allowed.
Request headers	Values are masked, encrypted by the schedule service, and never returned.
Request body template	Required. Include {{message}} for the current turn or {{history}} for the conversation.
Response path	Optional, such as reply or result.text
Replay-safe checkbox	Same meaning as on the interactive form
Advanced HTTP settings	Session ID path, Tool calls path, Request encoding (default Auto-detect), Timeout (seconds), Maximum response bytes

note

Credentials are sent securely only when you save and are never returned in schedule details.

4. Test plan and coverage

• Shared system prompt and Shared tools JSON (both optional).
• Attack coverage: Full library (all 64 catalog attacks) or Select attacks (search the catalog and pick). Schedules have no Quick scan option.
• Custom objectives, as on the interactive form.

5. Authorization and evaluation

Field	Notes
Authorized assessment scope	Required
Assessment depth	Low (default), High, or Maximum. The interactive form's Standard matches Low and Deep matches High.
Evaluation policy	Same options as the interactive form: Hybrid / Deterministic / Judge, judge count 1, consensus threshold 0.67, confirmation runs, deterministic evidence override
Acknowledgement	Required, as on the interactive form

Click Save schedule.

Related

• Reading Red Team Results
• Attack Library