{% extends "base.html" %} {% import "_ui.html" as ui %} {% block title %}Feedback{% endblock %} {% block crumbs %}Feedback & lessons{% endblock %} {% block content %} {%- set active_lessons = lessons|selectattr("active")|list|length -%}

Feedback & lessons

Reviewer corrections on generated rules. Applied feedback becomes a lesson that is injected into future generation prompts for the same technique and log source.

{{ ui.icon("download", "sm") }}Export JSONL (evaluation dataset)

{{ ui.icon("refresh") }}How feedback works

One loop per mistake: the rule gets fixed now, and future rules avoid the same mistake.

  1. 1
    ReportOn a rule's Feedback tab: what is wrong (artifact, category, severity) and why. Saved against that rule version.
  2. 2
    FixPaste a corrected rule (validated before it is accepted){% if ai_enabled %}, or let the AI rewrite it from your comment{% endif %}. The fix becomes the next version; the old one is kept in history.
  3. 3
    Re-reviewThe new version starts as a draft, queries are re-converted and test results reset. Someone other than its author approves it.
  4. 4
    LearnThe comment becomes a lesson, redacted and added to future generation prompts for the same ATT&CK technique or log source.

Feedback on behaviours, simulations or the ADS is recorded for evaluation (export below) but does not change prompts unless it is turned into a lesson.

Feedback

{{ items|length }} item(s){% if status %} with status {{ status }}{% endif %}.

{% for f in items %} {% else %}{% endfor %}
#StatusRuleFeedbackByResolution
{{ f.id }} {{ ui.status(f.status) }}
{{ f.severity }}
{{ f.slug }} v{{ f.rule_version }}
{{ f.run_id }}
{{ f.artifact }}{% if f.target %} ({{ f.target }}){% endif %} ยท {{ categories.get(f.category, f.category) }}
{{ f.comment }}
{{ f.reviewer }}
{{ ui.short_date(f.created_at) }}
{{ f.resolution or "" }}
{{ ui.empty("No feedback yet" if not status else "Nothing " ~ status, "Open a rule and use its Feedback tab to report a mistake.", "feedback") }}

Lessons

Injected into future generation prompts. Disable a lesson that turned out to be wrong.

{% for l in lessons %} {% else %}{% endfor %}
#TechniqueLog sourceLessonFromActive
{{ l.id }} {{ l.technique_id }} {{ l.product }}/{{ l.category }} {{ l.text }} #{{ l.feedback_id }}
{{ ui.short_date(l.created_at) }}
{{ ui.status("active" if l.active else "disabled", "ok" if l.active else "muted") }}
{{ ui.empty("No lessons yet", "Applied feedback becomes a lesson.", "book") }}

Audit log

Last 100 state-changing actions.

{% for a in audit %} {% else %}{% endfor %}
TimeActorActionDetails
{{ ui.short_date(a.ts) }}{{ a.actor }}{{ a.action }} {{ a.details }}
{{ ui.empty("No audit entries", "", "history") }}
{% endblock %}