Skip to content

Escalation messages need decision context — screenshots, scores, options #63

@gregario

Description

@gregario

Problem

Even if the Slack notification worked (#62), the human receiving it would have no context to make a decision. They'd need to SSH in, read cycle_context.json, grep logs, and piece together what happened.

What the human needs to see

When the loop escalates, the Slack message should include:

  1. What feature area and cycle number
  2. Health score and confidence trend (e.g., "80/100, confidence plateaued at 0.82")
  3. The specific decision needed — from analysis_recommendation.rationale
  4. Options as actionable buttons — e.g., "Accept & advance to FA2", "Deepen one more cycle", "Need more info"
  5. Key screenshots — last walk's main screen screenshot attached to the thread
  6. Summary of what was built — file count, test count, acceptance criteria pass rate
  7. What the evaluator said — one-line summary from each lens (QA, Design, PO)

Example message

🟡 fleet-manager needs your input (FA1: Fleet Dashboard, cycle 2)

Health: 80/100 | Confidence: 0.82 (plateaued) | Tests: 139/139 | Criteria: 9/9 (7 env-limited)

Decision needed: Confidence stuck at 82%. Responsive polish didn't improve it. WebGL map can't be verified in headless.

Options:
/rouge feedback fleet-manager accept — Ship FA1 at 80, advance to FA2
/rouge feedback fleet-manager deepen — One more cycle with design focus
/rouge feedback fleet-manager info — Show me the screenshots and details

[📸 Dashboard screenshot attached]

Priority

Blocker — paired with #62. The escalation pathway is: notification reaches human (#62) + human has context to decide (this issue).

Metadata

Metadata

Assignees

No one assigned

    Labels

    blockerMust fix before public launchenhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions