You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/next-steps.md
+7-5Lines changed: 7 additions & 5 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -19,19 +19,21 @@ Focused follow-up work for `@knighted/develop`.
19
19
- Suggested implementation prompt:
20
20
- "Add a deterministic E2E execution mode for `@knighted/develop` that serves pinned runtime artifacts locally (instead of live CDN fetches) and wire it into CI as a required check on every PR. Keep a separate lightweight CDN-smoke E2E check for real-network coverage. Validate with `npm run lint`, deterministic Playwright PR checks, and one CDN-smoke Playwright run."
21
21
22
-
4.**Issue #18 continuation (resume from Phase 3)**
- Phase 2 complete: separate AI chat drawer UX, streaming-first responses with non-stream fallback, selected repository context plumbing, and README fine-grained PAT setup links.
27
-
- Implement the next slice first (Phase 3):
27
+
- Phase 3 partially complete: PR-prep filename/path groundwork landed via the Open PR drawer with repository-scoped persistence and stricter path validation.
28
+
- Phase 4 complete: open PR flow from editor content (branch creation, file upserts, PR creation), confirmation UX, loading/success states, and toast feedback.
29
+
- Post-implementation hardening complete: traversal/path validation edge cases, trailing-slash rejection, writable-repo select reset behavior during loading/error states, and a JS-driven Playwright readiness check.
30
+
- Implement the next slice (remaining Phase 3 assistant features):
28
31
- Add mode-aware recommendation behavior so the assistant strongly adapts suggestions to current render mode and style mode.
29
32
- Add an editor update workflow where the assistant can propose structured edits and the user can apply to Component and Styles editors with explicit confirmation.
30
-
- Add filename groundwork for upcoming PR flows by allowing user-defined Component and Styles file names, persisted per selected repository.
31
33
- Keep behavior and constraints aligned with current implementation:
32
34
- Keep everything behind the existing browser-only AI feature flag.
33
35
- Preserve BYOT token semantics (localStorage persistence until user deletes).
34
36
- Keep CDN-first runtime behavior and existing fallback model.
35
37
- Do not add dependencies without explicit approval.
- "Continue Issue #18 in @knighted/develop from the current Phase 2 baseline. Implement Phase 3 with three deliverables. (1) Add mode-aware assistant guidance: when collecting AI context, include explicit policy hints derived from render mode and style mode, and ensure recommendations avoid incompatible patterns (for example, avoid React hook/state guidance in DOM mode unless user explicitly asks for React migration). (2) Add assistant-to-editor apply flow: support structured assistant responses that can propose edits for component and/or styles editors; render these as reviewable actions in the chat drawer, require explicit user confirmation to apply, and support a one-step undo for last applied assistant edit per editor. (3) Add PR-prep filename metadata: introduce user-editable fields for Component filename and Styles filename in AI controls, validate simple safe filename format, and persist/reload values scoped to selected repository so Phase 4 PR write flow can reuse them. Keep all AI/BYOT behavior behind the existing browser-only AI feature flag and preserve current token/repo persistence semantics. Do not add dependencies. Validate with npm run lint and targeted Playwright tests covering: mode-aware recommendation constraints, apply/undo editor actions, and repository-scoped filename persistence."
- "Continue Issue #18 in @knighted/develop from the current baseline where PR filename/path groundwork and Open PR flow are already shipped. Implement the two remaining Phase 3 assistant deliverables. (1) Add mode-aware assistant guidance: when collecting AI context, include explicit policy hints derived from render mode and style mode, and ensure recommendations avoid incompatible patterns (for example, avoid React hook/state guidance in DOM mode unless user explicitly asks for React migration). (2) Add assistant-to-editor apply flow: support structured assistant responses that can propose edits for component and/or styles editors; render these as reviewable actions in the chat drawer, require explicit user confirmation to apply, and support a one-step undo for last applied assistant edit per editor. Keep all AI/BYOT behavior behind the existing browser-only AI feature flag and preserve current token/repo persistence semantics. Do not add dependencies. Validate with npm run lint and targeted Playwright tests covering mode-aware recommendation constraints and apply/undo editor actions."
0 commit comments