Skip to content

feat(issue-workbench): classify issues and coordinate organization members - #80

Open
YukinaXSer wants to merge 24 commits into
zts212653:mainfrom
YukinaXSer:feat/issue-workbench
Open

YukinaXSer wants to merge 24 commits into
zts212653:mainfrom
YukinaXSer:feat/issue-workbench

Conversation

@YukinaXSer

@YukinaXSer YukinaXSer commented Oct 14, 2026 •

Copy link
Copy Markdown

The Issue workbench combines GitHub READ, classifications and local investigation with a candidate module for cat analysis and organization-member assignment. Opening an issue now loads its members immediately. Owners can assign directly without first requesting analysis, or select a Host-routable cat to receive a classification and rule-limited assignment task. Returned classifications update the queue, filters and detail together.

This remains plugin PR #80. Host PR #1639 was withdrawn; hostInstallable:false still applies. Installed Host access/authorization, live-cat invocation and real two-account acceptance remain incomplete. This PR is not approved for merge.

Behavior

  • Browse issues, Feature and semantic classifications; save local annotations, plans, relationships and verification. Adjustable columns retain widths and drafts across desktop/mobile layouts.
  • Organization members come from verified, paginated GitHub reads. Opening an issue loads the roster; member reporters are preferred. Direct assignment requires explicit confirmation and a current source version, has a separate durable record, and never invents a cat analysis or score.
  • Choose among the configured thread's Host-routable cats, including cats not yet joined. The selected destination is persisted and revalidated; retry never substitutes the default cat. Only the bound cat/thread can submit results.
  • The default task requests classification and rule-limited personnel assignment. Organization membership, reporter/Feature/category priority, repository permission, source state/version and existing assignees remain enforced. With automatic assignment off, the task recommends and waits for owner confirmation.
  • Existing classification appears automatically; pending results refresh into the queue, category filters and detail without another button. Reading alone does not wake a cat. Unclassified issues remain explicit until analysis returns. Provider identity/update stamps and fresh content verification prevent stale or cross-source results; Feature bindings retain precedence.
  • Existing team intake, uncertain-publication recovery and explicit feedback remain available. A failed one-time manual assignment never becomes authorization for automatic background retry. Statistics distinguish direct assignments from analyses and do not imply training or repair acceptance.

Integration and validation

Public SDK remains 0.2.0-beta.21. The Core export remains pinned to 031244712b8b62d1dc017ed379efd09b15dd4fc6; explicit plugin export adapters keep integration reproducible. No new dependency, shared Host/SDK contract change or production configuration change is included. Stored local work and team history remain intact without expiry.

Final verification: HEAD 267d7b1d421673101d8ea2e9f280b0e2b302c936.

  • All 35 repository gate steps passed. The five package-local changes made during that run were audited and followed by a frozen rerun of the entire Issue package: lint, all tests, build, independent packed-artifact installation and all browser journeys passed; all 117 captured input digests matched on readback.
  • 52 team checks cover eager reads, selected-cat authority/recovery, direct assignment/replay, policy changes, organization/permission boundaries and classification source isolation.
  • Six packed browser/READ journeys include direct assignment before analysis, selected cat, automatic results, queue/filter/detail agreement, reload, comment-only update stamps, stale source withdrawal, roster failure, feedback/recovery and responsive layouts.
  • Final packed preview uses live meow-club membership reads with clearly simulated Issue/cat/assignment data. The exact committed revision, immediate 10-member roster, selectable demonstration cats, opening an issue without prior analysis, reload, desktop/mobile/dark views and unchanged persisted data were verified.
  • Evidence: docs/evidence/issue-workbench-assignment-flow.md, packages/issue-workbench/VALIDATION.md, and local artifacts/assignment-flow-2026-10-21/.

GitHub CI is not green: current-head [Contract CI](https://github.com/zts212653/clowder-ai-plugins/actions/runs/39542657908) and [Personal Chrome Companion CI](https://github.com/zts212653/clowder-ai-plugins/actions/runs/39542658075) report action_required. assert-pr-head confirms this exact PR HEAD and both new runs exist.

The owner requested solo execution while other cats have no remaining quota. These are author checks, not independent approval. No real GitHub assignments/comments/label writes or merge were performed. The original Task retains real installed access, live-cat/two-account acceptance and applicable independent review.

Author: cat-xooblf7l / gpt-6-astra.

Why: preserve the existing Issue workflow as an independently buildable pre-SDK artifact while providing a repeatable, conflict-aware path for continued Core optimizations. Keep Host authority, real installation and task dispatch explicit until fixed SDK integration is accepted.

Source: clowder-ai feat/issue-workbench@3aca4242a03081edbfae32b8f34e316f845d5daf. Includes per-file provenance, packed consumer/browser checks and SDK cutover policy.

[砚砚/gpt-6-astra🐾]
CatXooblf7l-GPT-6-astra added 4 commits October 14, 2026 17:24
Why: the owner added an internal-repository use case. Record the agreed source configuration journey, authority and identity boundaries, platform-specific implementation steps, and a synthetic isolation counterexample before wiring real connections. Package runtime bytes remain unchanged.

[砚砚/gpt-6-astra🐾]
… verification

Why: coordinate upstream and internal reports without copying issue lifecycles or exposing internal evidence through public tasks. Preserve the one-way Core export and the original trial records.
…vision

Why: a single reading panel must survive issue switching, and the workspace preview needs exact-build attestation. Keep the inline bridge exception limited to the approved local preview hostname.
…rowser CI gate

Why: the navigation and authorization regressions must run whenever the package browser gate runs.
@YukinaXSer YukinaXSer changed the title feat(issue-workbench): package pre-SDK workbench with reproducible Core sync feat(issue-workbench): package standalone workbench with multi-source investigation Oct 14, 2026
CatXooblf7l-GPT-6-astra added 11 commits October 14, 2026 19:25
Why: bootstrap local candidate labels from current public issue reports while keeping source workflow decisions separate. Include a bounded 12-case semantic sample with source evidence, and record the operator's update-only PR boundary.

[砚砚/gpt-6-astra🐾]
Why: long issue bodies buried local notes and save actions. Export the reviewed design direction from Core e562a37b, retain issue identity and drafts in narrow views, and verify the packed desktop/mobile journey. No merge requested. [砚砚/gpt-6-astra🐾]
Why: selecting a related issue from hundreds of unfiltered options is impractical. Export partial-text and source/number search, bounded results and keyboard selection; verify the actual packed journey with 803 issues and stale-target invalidation. [砚砚/gpt-6-astra🐾]
Why: The operator could not inspect the twelve proposals in the workbench, and the dataset belonged beside the plugin rather than in a plan. Package the evidence-backed public seed, display authorized candidates with source evidence, and require explicit local draft/save actions.

Author: 砚砚 / gpt-6-astra
Why: users need the current issue inventory grouped visibly, with explicit Feature bindings taking precedence over semantic candidates. Ship the first 708 proposals beside the plugin, preserve manual records, and make the isolated public READ preview independent of an expired bridge.
Why: make source lifecycle visible across queues, classification cards and details, with counted state filters that preserve local investigation drafts.
Why: real-page verification exposed shell styles overriding compact button sizing; preserve component typography and assert text stays inside each button.
Why: preserve full state labels for real inventory counts in the narrow desktop queue, including selected typography.
Why: PR zts212653#80 conflicts with the consolidated CI and dependency graph. Preserve upstream locked resolutions and route existing workbench checks through the shared gate while keeping the packed browser journey. This updates the feature PR only; no merge to main.
Why: The semantic-only headline total made Feature-bound issues appear missing. Remove the result total and treat Feature groups as classifications when displaying empty results, while retaining category filters and counts.
Why: Classified issues need accountable cat analysis, configured member routing, durable offline intake, and explicit feedback. Implement the public module boundary and fixed GitHub provider with recovery and provenance, while recording the remaining Host, UI, and real-account integration work.
…seline

Why: The local Core checkout predated C1 module support. Verify the current Host, export its build-compatible shared imports, and record isolated archive admission and the narrow Host grant companion without claiming the unfinished installed user journey is complete.

Thread-Context: threadId=thread_mucwsgsjtlbabfcq catId=cat-xooblf7l
@YukinaXSer YukinaXSer changed the title feat(issue-workbench): package standalone workbench with multi-source investigation feat(issue-workbench): add classified workbench and team analysis candidate Oct 20, 2026
…tions

Why: users need to request cat analysis, inspect assignment evidence, record feedback and recover uncertain publications from the workbench. Keep Host owner authority separate from native cat callers and preserve assignment change history.
@YukinaXSer

Copy link
Copy Markdown
Author

@codex review

@chatgpt-codex-connector

Copy link
Copy Markdown

To use Codex here, create a Codex account and connect to github.

CatXooblf7l-GPT-6-astra added 3 commits October 20, 2026 19:47
…ions

Why: native calls and Settings recovery could outlive their admitted authority during GitHub queries and still publish or return data. Bind the current authority inside each serialized operation, recheck before effects and results, and release it before the next request or scheduled recovery.

Validation: three observed failing regressions became green; all five authority scenarios, 27 team checks, the full 35-step gate and five packed browser journeys passed. Independent review and real installed/live-cat/two-account acceptance remain open; neither PR is approved or merged.

[砚砚/gpt-6-astra🐾]
Why: The owner limited development to plugin PR 80 and requested rollback of the Host expansion. Mark PR 1639 withdrawn, qualify historical acceptance evidence and preserve the uncompleted installation boundary without changing plugin behavior or user data.

Validation: documentation diff checked; Host companion full tree matches pre-expansion base and GitHub reports CLOSED with mergedAt null.

Thread-Context: threadId=thread_mucwsgsjtlbabfcq catId=cat-xooblf7l sourceMessageId=0001792509520767-000070-c6435188

[砚砚/gpt-6-astra🐾]
Why: repository access and routing rules must not implicitly define staff membership. Prefer a member reporter, support revision-bound owner assignment, and suggest reporter acceptance before merge while preserving existing assignees and classifications.

Validate: 38 team checks, 5 packed browser/READ journeys and all 35 repository gate steps passed. Group and request authority are rechecked before effects. No Host, SDK, production configuration or real issue assignment changes.

Author: cat-xooblf7l / gpt-6-astra
@YukinaXSer YukinaXSer changed the title feat(issue-workbench): add classified workbench and team analysis candidate feat(issue-workbench): classify issues and dispatch to staff groups Oct 20, 2026
Why: fixed side columns and viewport breakpoints prevent users from arranging the reading workspace inside a narrower Host. Add accessible drag and keyboard separators, retain browser width preferences without expiry, and adapt to actual container width without losing drafts.

Validate: all 35 repository gate steps and 6 packed browser/READ journeys passed; 112 input digests matched. Inspected desktop, dark and mobile pages. Core stays unchanged; the explicit export adapter preserves the frozen source provenance.

Author: cat-xooblf7l / gpt-6-astra
Why: the operator needs staff linked to the GitHub organization instead of a manually maintained list. Verify active organization membership, read every member page, show and refresh the roster, and recheck current organization eligibility before assignment and publication while preserving historical work.

Validate: 44 team checks, all 35 repository gate steps and 6 packed browser/READ journeys passed; 114 captured inputs matched. The packed provider and local page read and refresh 10 meow-club members. No real GitHub assignment/comment writes, Core changes or merge.

Author: cat-xooblf7l / gpt-6-astra
@YukinaXSer YukinaXSer changed the title feat(issue-workbench): classify issues and dispatch to staff groups feat(issue-workbench): classify issues and coordinate organization members Oct 21, 2026
Why: opening an issue should expose members immediately and dispatch a classification plus constrained assignment task to the chosen cat. Direct owner assignments are recorded separately, and persisted classifications drive the queue, filters and detail together.

Validation: 35 repository gate steps passed; final affected package lint, complete tests (52 team checks), build, packed artifact and 6 browser/READ journeys passed. All 117 captured inputs matched on readback. No real GitHub assignment/comment writes or Core changes.

Author: cat-xooblf7l / gpt-6-astra
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant