Claude-Pipeline: Command for Claude Code

Create a technical design grounded in live documentation research.

---

Prerequisites

Check that a session exists:

# Read current session
if [ ! -f ".claude/artifacts/current.txt" ]; then
  echo "ERROR: No active session. Run /arm first."
  exit 1
fi
SESSION_DIR=$(cat .claude/artifacts/current.txt)
if [ ! -f "$SESSION_DIR/brief.md" ]; then
  echo "ERROR: No brief.md found in $SESSION_DIR. Run /arm first."
  exit 1
fi

If no session exists, stop and tell the user to run /arm <task> first.

---

Technical Design

Use the architect agent.

Input: Read brief.md from the current session directory.

Process:

1. Extract key requirements and technology keywords from the brief
2. Research live documentation via WebSearch and WebFetch:
   • For each library/API mentioned, find current best practices
   • Get specific API patterns, configuration options, gotchas
   • Note the URLs and relevant quotes
3. Analyze existing codebase patterns via Glob/Grep/Read:
   • How does RDO already handle similar features?
   • What patterns should we follow for consistency?
4. Make design decisions:
   • Each decision must cite live docs OR existing codebase
   • Document alternatives considered and why rejected
5. Define component interfaces, data models, API contracts
6. Output design.md to the session directory

---

Documentation Research Guidelines

For each technology decision:

1. Search: [technology] documentation 2024 or [library] best practices
2. Fetch: Read the actual docs page, not just search results
3. Extract: Specific recommendations, code patterns, warnings
4. Cite: Include URL and relevant quote in the design

Example citation:

**Decision:** Use React Query's `staleTime` of 5 minutes for company data.

**Source:** https://tanstack.com/query/latest/docs/framework/react/guides/important-defaults
> "staleTime defaults to 0, meaning queries will be considered stale immediately"

We set 5 minutes because company data changes infrequently.