repo-native-alignment: Agent for Claude Code

/friction

Log a friction event with an RNA MCP tool. This is how the product improves — every friction point is signal.

The Rule

Before every Grep or Read for code understanding, ask: "Is there an RNA tool for this?"

• If yes and you use it → great, no log needed (unless it fails)
• If yes and it fails → log the failure
• If yes and you skip it → log why you skipped it (this IS friction)
• If no RNA tool applies → no log needed

Using Grep/Read for code understanding without trying the RNA tool first is itself a friction event. Not trying is not "no friction" — it's the worst kind of friction, because it means the tool wasn't even worth attempting. That's critical signal.

The only Grep/Read uses that are NOT friction events:

• Searching for non-code content (config files, CI yaml, shell scripts)
• Reading a specific file you already know the path to (e.g., after RNA told you the file)
• Searching for string literals, error messages, or log output
• Grep/Read that RNA genuinely can't do (regex patterns, multi-file sed-like operations)

When to Log

Call /friction (or apply this process inline) whenever:

• An RNA tool returned wrong or irrelevant results
• An RNA tool was missing data you expected (symbol, edge, metadata)
• The API was awkward — needed N calls where 1 should suffice, params confusing
• A tool was too slow and disrupted your flow
• You used Grep/Read instead of an RNA tool — log which RNA tool you should have used and why you didn't
• A tool errored or hung

How to Log

Write friction to .oh/friction-logs/<pipeline-or-context>.md — one file per pipeline run or work session. Create the file if it doesn't exist, append if it does.

File format:

# Friction Log: <context>
**Date:** <date>
**Pipeline/Issue:** #<number> or <description>

| Phase/Step | Tool | What happened | Workaround | Severity |
|------------|------|---------------|------------|----------|
| <phase> | <tool name> | <what happened> | <what you did instead> | <severity> |