## Context
If @.ralph/guides/AGENTS.md exists, study it for commands and patterns.
Study @.ralph/specs/$FEATURE.md for feature specification.
Study @.ralph/specs/$FEATURE-implementation-plan.md for current tasks.
{{#if frameworkVariant}}For detailed architecture, see @{{appDir}}/.claude/CLAUDE.md{{/if}}

## Learnings
Read @.ralph/LEARNINGS.md for patterns and anti-patterns from previous iterations.
Apply relevant learnings to avoid repeating past mistakes.

## Performance
For data fetching, new components, or optimization tasks, reference @.ralph/guides/PERFORMANCE.md.
Key patterns: parallel fetches, direct imports, React.cache(), lazy loading.

## Search
- Use `mgrep "query"` for codebase searches (patterns, implementations, usages)
- Use Context7 MCP for library/framework documentation
- Search codebase before assuming something doesn't exist

## Task
Work through ALL incomplete tasks in the implementation plan in a single session.
**Skip E2E tasks** (tasks starting with `E2E:`) - those are handled in a separate phase.
For each task: implement it, write tests, validate, commit, then move to the next task.
Do not stop after one task — keep going until all non-E2E tasks are complete.

## Validation
After changes, ALL must pass:
1. Run: `cd {{appDir}} && {{lintCommand}} -- --fix`
2. Run: `cd {{appDir}} && {{typecheckCommand}}` (typecheck)
3. Run: `cd {{appDir}} && {{testCommand}}`
4. Run: `cd {{appDir}} && {{buildCommand}}`

If any validation fails, fix the issue before proceeding.

## Security Review
Before committing, review your changes against @.ralph/guides/SECURITY.md:
1. **Quick scan**: Input validation, injection prevention, auth checks, data exposure
2. **Run**: `cd {{appDir}} && {{packageManager}} audit` (check for vulnerable dependencies)
{{#if hasSupabase}}3. **Check**: `mcp__supabase__get_advisors` with type "security" (RLS policies)
{{/if}}
4. **Red team**: Can auth be bypassed? Can other users' data be accessed?

Flag any security issues in the implementation plan and fix before committing.

## Design Quality Check
Before marking a UI task complete, verify against @.ralph/guides/FRONTEND.md:
1. Uses design tokens (no hard-coded colors like `#fff` or `rgb()`)
2. Has hover/focus/active states on interactive elements
3. Responsive on mobile (test at 375px width)
4. Empty/loading/error states handled
5. Charts have titles, tooltips, and legends (if applicable)
6. Consistent spacing using Tailwind scale

If any check fails, fix before committing.

## Completion
When ALL validations pass:
1. Update @.ralph/specs/$FEATURE-implementation-plan.md — change the task's `- [ ]` to `- [x]` and append the commit hash (e.g., `- [x] Task description - abc1234`). The harness tracks progress by counting checkboxes, so this step is mandatory.
2. `git -C {{appDir}} add -A`
3. `git -C {{appDir}} commit -m "type(scope): description"`
4. `git -C {{appDir}} push origin feat/$FEATURE`

## Learning Capture
If this iteration revealed something useful, append to @.ralph/LEARNINGS.md:
- A useful pattern -> Add under "## Patterns (What Works)"
- A mistake/issue -> Add under "## Anti-Patterns (What to Avoid)"
- Tool usage tip -> Add under "## Tool Usage"
- Codebase convention -> Add under "## Codebase Conventions"

Format: `- [YYYY-MM-DD] [$FEATURE] Brief description`

## Rules
- Complete ALL remaining non-E2E tasks before ending the session
- Commit after each task so progress is preserved if the session is interrupted
- Tests are mandatory - no task is complete without tests
- Search codebase before assuming something doesn't exist
- If blocked on a task, document in implementation plan and move to the next task
{{#if hasSupabase}}- Use Supabase MCP for database operations
{{/if}}{{#if hasPosthog}}- Use PostHog MCP for analytics queries
{{/if}}