Paste into Claude Code / CLAUDE.md task

# DevAgentRadar → Claude Code

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

## Context
Assistant: Goose
Proposal: Upgrade to Goose v1.46.0 and use per-message cost UI with cache-token tracking
Summary: Ship Goose v1.46.0 so sessions show per-message tokens/cost/TTFT/tok/s, cache-token-aware totals, streaming shell output, and cache-safe append-only request assembly from the unrolled agent loop.
Primary source: https://github.com/aaif-goose/goose/releases/tag/v1.46.0

## Why it matters
v1.46.0 makes agent cost and latency visible at message granularity (tokens, cost, TTFT, tok/s) and rolls those into session totals, including cache tokens for accurate billing. Cache-safe request assembly with append-only turn context and declared cache semantics reduces broken or wasteful prompt-cache use. Streaming shell output shortens the feedback loop while commands run. Pair that with markdown session export, optional disabling of built-in skills, and structured compaction summaries so teams can inspect runs, trim skill noise, and keep long sessions cheaper without guessing from the release notes alone.

## Suggested actions
1. Install or upgrade Goose to v1.46.0 from https://github.com/aaif-goose/goose/releases/tag/v1.46.0 and restart the CLI/IDE surface so the unrolled agent loop and cache-safe request assembly are active.
2. Open a real multi-tool coding session, run at least one long shell command, and confirm shell output streams while the command runs and each assistant message shows tokens, cost, TTFT, and tok/s with session totals including cache tokens.
3. Export the same session with the markdown format option and archive it next to the PR or runbook so per-message usage and tool traces are reviewable outside the UI.
4. For agents that should not load defaults, disable built-in skills you do not want, then re-run one task and compare per-message cost/token stats against the previous run.
5. If you use local voice input, set LOCAL_WHISPER_LANGUAGE to your team’s primary transcription language and verify multilingual local transcription on a short dictation before a working session.

## Config surfaces this release may change
- background and headless runs — check your repo before applying
- slash commands — check your repo before applying
- context and compaction settings — check your repo before applying
- agent context files — check your repo before applying
- hooks — check your repo before applying
- MCP servers — check your repo before applying
- skills — check your repo before applying


## Your job
1. Restate the change in one sentence.
2. Propose a minimal plan for my repo (or a throwaway pilot).
3. Implement only what I approve; prefer small diffs and tests.
4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.