Skip to content

Commit 65f31a8

Browse files
authored
fix: harden Discussion and Agent Core v2 workflows (#273)
## Related Issue No public issue. This is direct repository-owner requested work for reported Discussion regressions and approved Agent Core v2 reconciliation. ## Problem Discussion did not preserve the selected model pair, had no per-role reasoning effort, exposed inactive state in the composer, and rejected valid catalog models through the unrelated subagent pool. Live exchanges could also stop after a late tool call, lose reasoning updates, collapse completed work, or let one model panel overgrow the other. Several applicable Agent Core v2 fixes were also missing from this branch, including provider compatibility, safe command handling, config formatting preservation, transcript recovery, and Tower workspace-state safeguards. ## What changed - Persist Discussion model and effort preferences across sessions, validate configured catalog models directly, and make New Discussion reuse the current workspace and saved pair. - Stream live reasoning in equal-height scrollable panes, keep completed exchanges expanded, add a persistent reasoning visibility setting, restore the final agreement/differences/uncertainty comparison, improve dark model menus, align Dynamic Workflow status dots, and anchor action/manual-stop toasts above the composer. - Harden the Discussion workflow against late tool calls, provider-idle failures, cancellation races, partial participant creation, cleanup failures, malformed Fusion output, and stale live snapshots without increasing request limits. - Reconcile applicable non-breaking Agent Core v2 fixes for Pythinker Responses, secondary-model routing, telemetry, Tower state, transcript identity, TOML writeback, and dangerous Bash command approval. - Update Web, TUI, gateway, klient, docs, tests, changesets, and the committed Web bundle together. ## Verification - `pnpm lint` — 0 errors - `pnpm typecheck` - `pnpm exec vitest run --maxWorkers=8` — 1,244 files and 21,143 tests passed - Focused Agent Core v2, gateway, and Web Discussion suites - `pnpm run build:web && pnpm run check:web` - `npm run build` in `docs/` - `node scripts/check-nix-workspace.mjs` - `nix build .#pythinker-code --print-build-logs` - `pnpm package:desktop` and strict deep code-sign verification - Launched the exact packaged worktree app successfully Two unrelated timing tests each failed once under unbounded full-suite saturation. Both passed immediately in isolation; the complete bounded-worker suite passed cleanly. ## Checklist - [x] I have read the [CONTRIBUTING](https://github.com/PyModel/pythinker-code/blob/main/CONTRIBUTING.md) document. - [ ] I have linked a related issue. Not applicable: direct repository-owner requested work; no public issue was authorized. - [x] I have added tests that prove the behavior works. - [x] Ran `gen-changesets` skill. - [x] Ran `gen-docs` skill.
1 parent 0d57540 commit 65f31a8

253 files changed

Lines changed: 8091 additions & 2076 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Add OpenAI Responses support to the Pythinker provider.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Enable configured secondary-model routing by default.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Place action and stopped notices above the composer.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Add per-role thinking effort controls and preserve saved model pairs in Discussion.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Preserve usable Discussion responses across transient provider and workflow failures.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Align Dynamic Workflow status dots with their subagent row titles.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Keep active and undone turns aligned after session reloads.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Ask before dangerous shell commands in interactive modes and block them in Auto mode.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Preserve comments and formatting when Pythinker updates config.toml.
Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
---
2+
"@pymodel/pythinker-code": patch
3+
---
4+
5+
Preserve staged, unstaged, and untracked workspace changes when Tower agents start and merge work.

0 commit comments

Comments
 (0)