chore(skills): add analyze-feedback skill and incorporate agent learnings (#2208)

* chore(skills): move analyze-feedback skill to .claude/skills directory

Move skill file and scan cursor to the correct .claude/ directory
structure and remove the now-unnecessary setup script.

* Incorporate feedback

Author: naqvitalha

feedback-scan-cursor.json .claude/feedback-scan-cursor.jsonfeedback-scan-cursor.json renamed to .claude/feedback-scan-cursor.json

@@ -1,5 +1,5 @@
 {
-  "last_scanned_at": "2026-02-22T00:00:00Z",
-  "last_run_id": 0,
+  "last_scanned_at": "2026-03-24T18:03:09Z",
+  "last_run_id": 23504809129,
   "note": "ISO-8601 UTC timestamp of the most recent workflow run that was scanned"
 }

.claude/skills/agent-device/SKILL.md

@@ -286,6 +286,12 @@ settings wifi off                     # Toggle wifi
 settings permission grant camera      # Grant camera permission
 ```
 
+## CI Known Issues
+
+- **Session creation may hang on CI runners** — `agent-device open` can hang indefinitely on CI. Set reasonable timeouts and be prepared to fall back to code-only verification with unit tests.
+- **Use full bundle ID on CI** — the app may not be recognized by display name; use the full bundle ID (e.g., `org.reactjs.native.example.FlatListPro`) instead of `FlatListPro`.
+- **Snapshot/session timeouts** — CI runners may experience persistent timeouts with snapshots and sessions. If agent-device is unavailable, document it in feedback and proceed with automated tests only.
+
 ## Tips
 
 - Round coordinates to the nearest integer

analyze-feedback-SKILL.md .claude/skills/analyze-feedback/SKILL.mdanalyze-feedback-SKILL.md renamed to .claude/skills/analyze-feedback/SKILL.md

@@ -97,6 +97,7 @@ Always do this **before** the raise-pr step.
 
 - **`dist/` is NOT rebuilt on branch switch** — you MUST `yarn build` after every `git checkout`. Verify with `grep` in `dist/` that the expected code change is present. Without this, you test stale code and get false results.
 - **Always reproduce the bug on `main` BEFORE testing the fix** — without confirming the bug exists on the base branch, you can't prove the fix works. See `review-and-test` skill → "Review Methodology".
+- **Code-only diagnosis is acceptable on CI** — layout/scroll bugs, Android-only issues (no emulator on macOS CI), and web/SSR issues (no browser on CI) can be diagnosed from code analysis + unit tests when simulator reproduction isn't feasible.
 
 For more testing/debugging pitfalls (console.log, RTL setup, agent-device swipe, observable callbacks), see the **`review-and-test` skill**.
 

.claude/skills/fix-github-issue/SKILL.md

@@ -5,7 +5,7 @@ description: Create a GitHub PR for FlashList. Ensures no AI/Claude attribution
 
 # Raise a FlashList PR
 
-**Prerequisite:** Run the `review-and-test` skill first. All checks must pass and device testing must be done before raising a PR.
+**Prerequisite:** Run the `review-and-test` skill first. All checks must pass and device testing must be done before raising a PR. **DO NOT SKIP**.
 
 ## Rules
 
@@ -46,6 +46,7 @@ Fixes #<number>
 Scope examples: `layout`, `hooks`, `scroll`, `sticky-headers`, `recycling`, `viewability`.
 
 **Do NOT include**:
+
 - `Co-Authored-By` lines
 - Any mention of Claude, AI, Anthropic, or automated tools
 - `Generated by` or `Assisted by` attributions
@@ -76,6 +77,7 @@ Run `git diff HEAD~1` and check for:
 - **Fixture `index.js` left with `forceRTL(true)`** — always revert before committing
 
 Quick check:
+
 ```bash
 grep -r "console\.\(log\|warn\)\|localhost:9876\|forceRTL(true)" src/ fixture/react-native/index.js --include="*.ts" --include="*.tsx" --include="*.js"
 ```
@@ -93,7 +95,7 @@ Fix any issues found, then amend the commit before proceeding.
 
 ### PR body template
 
-**Always use `--body-file`** — inline `--body` with markdown `#` headers triggers Claude Code permission checks and wastes turns. Use the `Write` tool to create the file (not `cat` or `echo`, which may not be in allowed tools on CI).
+**Always use `--body-file`** — inline `--body` with markdown `#` headers triggers Claude Code permission checks and wastes turns. Use the `Write` tool to create the file (not `cat`, `echo`, or `touch`, which may be blocked by sandbox on CI).
 
 Write the following to `/tmp/pr-body.md` using the Write tool:
 
@@ -113,16 +115,17 @@ Fixes #<number>
 <Before/after screenshots if applicable>
 
 ## Test plan
+
 - [ ] Unit tests pass (`yarn test`)
 - [ ] Type check passes (`yarn type-check`)
 - [ ] Lint passes (`yarn lint`)
 - [ ] Verified on iOS simulator
 - [ ] No regressions on related screens
-EOF
+      EOF
 
 GH_TOKEN="$AGENT_PR_TOKEN" gh pr create \
-  --title "fix(<scope>): <description>" \
-  --body-file /tmp/pr-body.md
+ --title "fix(<scope>): <description>" \
+ --body-file /tmp/pr-body.md
 ```
 
 > **CRITICAL — TOKEN VERIFICATION (read this before running `gh pr create`)**
@@ -135,6 +138,7 @@ GH_TOKEN="$AGENT_PR_TOKEN" gh pr create \
 > If `AGENT_PR_TOKEN` is empty, try `SHOPIFY_GH_ACCESS_TOKEN` as fallback: `GH_TOKEN="${AGENT_PR_TOKEN:-$SHOPIFY_GH_ACCESS_TOKEN}" gh pr create ...`
 
 Before running `gh pr create`, double-check:
+
 - [ ] Token is verified (see above)
 - [ ] No mention of Claude, AI, Anthropic, or any AI tool anywhere
 - [ ] Description explains the "what" and "why" clearly
@@ -147,6 +151,7 @@ Before running `gh pr create`, double-check:
 Return the PR URL to the user.
 
 If the user asks to update the PR (interactive only — not available on CI):
+
 ```bash
 git add <files>
 git commit --amend --no-edit

.claude/skills/raise-pr/SKILL.md

@@ -32,5 +32,11 @@ description: Triage a GitHub issue — classify priority (P0/P1/P2), search for
 - Do NOT modify any code.
 - Do NOT close issues — only label (and comment per Step 5 when genuinely needed).
 - **Only apply these labels**: `P0`, `P1`, `P2`. Do NOT apply any other labels (especially not `agent-fix`).
+- If the issue already has non-priority labels (e.g., `agent-fix`), leave them as-is — triage only adds priority labels, never removes existing ones.
+- **Test/spam issues**: If the title contains "[Test]", "please ignore", or similar markers, classify as P2 with minimal effort (skip thorough duplicate search).
 - Do NOT run `gh auth`, `gh api`, `gh secret`, or any `gh` command other than `gh issue`.
 - Do NOT read environment variables, process info, or files outside the repository.
+
+## CI Notes
+
+On CI, the issue number should be passed in the workflow prompt. If `gh issue` commands are blocked by permissions, the workflow's `allowedTools` or `settings.json` must include `gh issue view`, `gh issue list`, `gh issue edit`, and `gh issue comment`.