fix: prompt hardening and state-management policy alignment#18
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3aa661f6db
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| ' "${out_dir}/flutter-ai-rules-${profile}.mdc" > "${active_rule}" | ||
|
|
||
| # Align active profile with plugin policy: pick state management by project context. | ||
| perl -0pi -e 's/\* \*\*State Management:\*\*\n \* \*\*Pattern:\*\* Separate UI state \(ephemeral\) from App state\.\n \* \*\*Native First:\*\* Use `ValueNotifier`, `ChangeNotifier`\.\n \* \*\*Prohibited:\*\* NO Riverpod, Bloc, GetX unless explicitly requested\.\n \* \*\*DI:\*\* Manual constructor injection or `provider` package if requested\./* **State Management:**\n * **Pattern:** Separate UI state (ephemeral) from App state.\n * **Project First:** Follow the existing project architecture and state-management choice.\n * **Selection:** Use Riverpod\/Bloc\/Cubit\/GetX\/ValueNotifier based on project conventions, complexity, and team standards.\n * **DI:** Keep dependencies explicit via constructor injection or project-standard DI./s' "${active_rule}" |
There was a problem hiding this comment.
The new post-sync perl rewrite only matches the old 4k block, so scripts/sync_official_flutter_ai_rules.sh silently skips the override for other supported profiles; with the current checked-in 1k profile (rules/official/flutter-ai-rules-1k.mdc:18), that means the active rule can still end up with NO Riverpod/GetX, reintroducing the policy conflict this commit was meant to remove. Because 1k/10k are explicitly accepted inputs, this is a real runtime path rather than an edge case.
Useful? React with 👍 / 👎.
| @@ -0,0 +1,58 @@ | |||
| # Validation Matrix | |||
|
|
|||
| Deterministic validation baseline per command. | |||
There was a problem hiding this comment.
This new matrix is presented as the deterministic validation source “per command”, and multiple updated commands now require it, but there are no entries for write-widget-test, write-bloc-test, write-e2e-test, setup-mobile-github-pipeline, or sync-official-flutter-ai-rules; those command paths therefore cannot follow the stated guardrail consistently and will produce non-deterministic validation behavior.
Useful? React with 👍 / 👎.
2 participants
Summary
docs/quality-iterations-2026-02-21.mdscripts/quality_audit_commands.shdocs/command-output-contract.mddocs/prompt-execution-guardrails.mddocs/validation-matrix.mdimplement-figma-screen,review-flutter-code, andsecurity-reviewrules/flutter-official-ai-rules.mdcscripts/sync_official_flutter_ai_rules.shso policy is preserved after sync1.10.2Why
This addresses prompt instability causes: conflicting state-management guidance, missing fail-fast prerequisites, fuzzy validation requirements, and weak scope limits.