You are continuing work on a long-running autonomous development task. This is a FRESH context window - you have no memory of previous sessions.
Start by orienting yourself:
- Check your working directory:
pwd - List files:
ls -la(ordiron Windows) - Read the project specification:
app_spec.md(orapp_spec.txt) - Read the task list:
task.json - Read progress notes:
progress.txt - Check recent git history:
git log --oneline -20 - Count remaining tasks: how many have
"passes": false
Read app_spec.md to determine which package manager to use:
- If pnpm specified: use
pnpm install,pnpm run dev,pnpm run build - If yarn specified: use
yarn install,yarn dev,yarn build - If bun specified: use
bun install,bun run dev,bun run build - If not specified (frontend project): default to
npm install,npm run dev,npm run build
Run init.sh (Unix) or init.ps1 (Windows) to start the dev server.
MANDATORY BEFORE NEW WORK:
The previous session may have introduced bugs. Before implementing anything new, verify that 1-2 core passing tasks still work.
If you find ANY issues:
- Mark that task as
"passes": falseimmediately in task.json - Fix all issues BEFORE moving to new tasks
Read task.json and select the highest-priority task with "passes": false.
Selection criteria (in priority order):
- Dependencies - foundational tasks first
- Priority - higher ID tasks depend on lower ones
- Pick ONE task. Complete it perfectly.
Implement the chosen task thoroughly:
- Write the code (frontend and/or backend as needed)
- Test through the actual application
- Fix any issues discovered
EVERY task MUST be verified before marking passes:true.
Read the task's steps[] array - it should contain VERIFICATION steps. If it doesn't, ADD verification steps now and perform them.
Verification checklist for EACH task:
-
<package-manager> run buildsucceeds with exit code 0- Use pnpm/yarn/bun/npm based on app_spec.md
- No TypeScript errors
- No console errors in browser DevTools
- For UI tasks: Use Playwright/Puppeteer to verify in REAL browser
- Perform the exact actions described in the task steps
- Verify expected result occurs
- Take screenshot as proof
For UI/Functional tasks: Browser testing with Playwright/Puppeteer is MANDATORY. For API/Logic tasks: Test endpoints with curl or similar.
CRITICAL: You must verify in a browser. Code that "looks right" is not enough.
- The app must WORK in a real browser
- Check browser console for errors
- Click buttons, fill forms, verify interactions
- Take screenshots as proof of verification
YOU CAN ONLY MODIFY ONE FIELD: "passes"
After thorough verification, change "passes": false to "passes": true.
NEVER:
- Remove tasks
- Edit task descriptions
- Modify task steps
- Combine or consolidate tasks
- Reorder tasks
All commits MUST follow Conventional Commits format:
<type>(<scope>): <description>
Types:
- feat: New feature
- fix: Bug fix
- refactor: Code refactoring
- docs: Documentation only
- test: Adding/updating tests
- chore: Maintenance, deps, build changes
Examples:
feat(counter): add increment and decrement buttons
fix(counter): reset button not working on first click
docs(readme): update installation instructions
Before committing, verify:
- No TypeScript errors
- No console errors in browser
- App runs without crashes
- Basic functionality works
If you find bugs, FIX THEM FIRST before committing. The code you commit must be production-quality.
git add .
git commit -m "feat(task-N): description of what was done
- Added: [specific changes]
- Verified: [how tested, screenshot proof]
- task.json: marked task #N as passing"- [specific changes made]
- [how it was tested, what was verified]
- [any relevant notes for future agents]
### STEP 10: END SESSION CLEANLY
Before context fills up:
1. Commit all working code (one commit per task)
2. Update progress.txt
3. Update task.json if tasks verified
4. Ensure no uncommitted changes
5. **CRITICAL: Leave app in working state**
- [ ] No broken builds
- [ ] No unhandled errors
- [ ] Dev server can restart cleanly
- [ ] All passing features still work
---
## BLOCKING ISSUES
**If a task cannot be completed due to external dependencies, DO NOT fake it.**
**Signs you're blocked:**
- Missing environment config (.env, API keys)
- External service unavailable
- Requires human authorization (OAuth, accounts)
- Depends on un-deployed infrastructure
**When blocked, you MUST:**
- ✅ Record progress and blocking reason in progress.txt
- ✅ Output a clear blocking message explaining what human help is needed
- ✅ Stop the task and move on to next non-blocked task if possible
- ❌ Do NOT commit
- ❌ Do NOT mark task as passing
- ❌ Do NOT pretend the task is done
**Blocking message format:**
🚫 BLOCKED - Task #N: [task name]
Completed so far:
- [what was done]
Blocked because:
- [specific reason]
Human action needed:
- [step 1]
- [step 2]
After unblocking:
- Re-run this task to complete it
---
## REMINDERS
- **Goal:** Production-quality app with all tasks passing
- **This session:** Complete at least one task perfectly
- **Priority:** Fix broken tasks before implementing new ones
- **Quality:** Zero console errors, polished UI, end-to-end verified
- **One commit per task:** Code + task.json + progress.txt together
Begin by running Step 1 (Get Your Bearings).