Skip to content

feat(console): decision cockpit frontend redesign - #1011

Draft
Pigbibi wants to merge 20 commits into
mainfrom
feat/cockpit-frontend-redesign
Draft

Pigbibi wants to merge 20 commits into
mainfrom
feat/cockpit-frontend-redesign

Conversation

@Pigbibi

@Pigbibi Pigbibi commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

AI assistance

OpenAI Codex.

What changed

Redesign the six-page console around decision review, evidence, lifecycle and policy readback. Keep unavailable and unknown states explicit, and distinguish recorded owner intent from execution authority.

Fix the JavaScript check by exporting AccessError and HttpError from the display-validation client substitute, including their status/payload behavior. The production client already exports both classes; its behavior is unchanged. Patch Vite to 7.3.7 and the Worker test runtime's undici/sharp dependencies to compatible security-fixed versions. #1019 layers paginated audit history and runtime readback on this branch.

How verified

  • Reproduced the missing-export failure, then passed console_display_gaps_validation.mjs with the unchanged display assertions.
  • Passed the full frontend test:operations suite, TypeScript/Vite build and size budget, generated-asset validation, Worker API and human-decision tests, and documentation check (zero blocking findings).
  • Regenerated the Worker asset map; the security update produces identical frontend assets. Both npm projects report zero vulnerabilities.
  • Full-page desktop/mobile screenshots are collected from the final stacked frontend in the local synthetic environment and recorded in the local REPORT.md; they do not establish production acceptance.
  • Remote checks are tracked on the current head. The risk gate requires an owner label, which this task explicitly prohibits adding. No merge, deployment, labels, IAM or secret changes.

@coderabbitai

coderabbitai Bot commented Oct 31, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

Pigbibi and others added 10 commits November 1, 2026 16:02
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
@Pigbibi
Pigbibi force-pushed the feat/cockpit-frontend-redesign branch from 92522c6 to a1d9e0c Compare November 1, 2026 09:08
An unreadable decision state fell back to pending and an applied operation
claimed a runtime confirmation regardless of whether the target can be read
back. Both overstated what the platform actually reports.

States outside the server enum now render as unknown, stale/expired/
unavailable support each name its own blocker, and the applied badge follows
the server readback flag. The pause control states that pausing neither
liquidates nor cancels orders, the lifecycle stepper names the axis it shows,
and the operation strip refreshes instead of leaving the owner guessing.
Opening the queue from Overview always landed on the default tab, so the
owner had to find the decision again and the URL never said which one was
open. Sharing or returning to that page lost the target.

A queue row now carries its decision in the route, the page restores the tab
and filters when the target arrives, and Back clears the parameter. Index
links drop a stale target, and a queue row keeps its button role.
…M07, P01-P07) (#1024)

* fix(console): render machine values in cockpit UI only through bilingual copy

* test(console): cover text-leak fallbacks for enrollment, decisions and cockpit

* chore(i18n): drop unused regime copy superseded by lifecycle keys

* fix(console): localize unknown change-log outcomes and tier labels

* chore(console): regenerate v2 asset map from current build
…-redesign

# Conflicts:
#	web/strategy-switch-console/v2_asset_map.js
@Pigbibi Pigbibi added the owner-approved Owner approval for high-risk PRs. AI agents must never apply this label. label Nov 1, 2026
* fix(console): map audit events and gate decisions by server fields

* test(console): cover audit mapping, action gating, promotion body

* build(console): regenerate cockpit assets

* build(console): regenerate cockpit assets

* test(console): align audit fixtures with the merged envelope contract
…-redesign

# Conflicts:
#	web/strategy-switch-console/v2_asset_map.js

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

owner-approved Owner approval for high-risk PRs. AI agents must never apply this label.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant