Onboard to this task: $ARGUMENTS
Instructions
AI models start fresh on every task. Before writing any code, build a complete picture of the relevant context.
1. Understand the task
Restate in your own words:
- What needs to be done
- What acceptance criteria exist
- What constraints apply (security rules from CLAUDE.md, architecture from workspace.dsl)
2. Explore the codebase
Use extended thinking. Explore relevant areas:
Frontend (src/): App.jsx, topic data, chat panel, SSE handling
Backend (api/): ChatFunction.cs, SessionFunction.cs, StihiaService.cs, TopicData.cs
Architecture: architecture/workspace.dsl, relevant ADRs in architecture/ADRs/
CI/CD: .github/workflows/ if the task touches deployment or pipeline
IaC: infra/main.tf if the task touches infrastructure
git log --oneline -20
git diff main...HEAD --stat
3. Ask clarifying questions
List any blockers before starting. Do not assume — ask.
Examples:
- "Should the new agent follow ADR-0007's Azure AI Foundry pattern, or is this a local prototype?"
- "Is the
topicId range validated server-side or should I add it?"
4. Document findings
Create .claude/tasks/$ARGUMENTS/onboarding.md with:
# Onboarding: $ARGUMENTS
## Task Summary
[What needs to be done, in your own words]
## Relevant Files
[List files that need reading or changing]
## Key Constraints
[Security rules, architecture decisions, CI requirements]
## Questions Asked
[Questions + answers received]
## Plan
[High-level steps before writing any code]
5. Confirm before implementing
Show the onboarding doc summary. Wait for explicit approval before writing any code.