Test Fat Llama

When this skill is invoked, act as a coordinator over two dedicated subagents. Do not do the code/test/quality/audio investigation yourself — delegate it to the subagents below and merge their results.

Before doing anything else, read .claude/skills/rules/test-fat-llama.md (merge contract and relaying policy) and .claude/rules/scope-and-safety.md (filesystem write scope and safety boundaries every skill/agent follows) — both may be updated over time without this file changing.

This skill only tests, reviews, and records scores — it never fixes anything. Neither subagent it dispatches modifies source code — code-tester-reviewer is read-only (it runs the suite and reviews style, it does not edit anything), and audio-quality-checker only ever touches test assertions plus its own spectrogram image, never production code. The one file this coordinator writes itself is README.md, and only the two regions defined in .claude/skills/rules/test-fat-llama.md's README update contract. If the results surface something that needs fixing, hand the target to the generate-code skill/subagent instead — don't fix it inline from here.

Logging

Before Step 1, open this run's log file per .claude/rules/logging.md (name: test-fat-llama-<time>-<user>.log). Append one entry per step below, including each subagent dispatch and its own log filename (from its JSON report's "log" field).

Steps

1. Launch the code-tester-reviewer subagent via the Agent tool with subagent_type: "code-tester-reviewer" and run_in_background: false (wait for it — the caller needs the result in this turn). Give it a self-contained prompt: it should run fat_llama's test suite and review code quality/PEP8 compliance per its own instructions and rules file (.claude/agents/rules/code-quality.md); no extra context is needed, the agent definition already covers scope and paths.

2. Launch the audio-quality-checker subagent the same way (subagent_type: "audio-quality-checker", run_in_background: false). Since code-tester-reviewer no longer edits any files, there's no shared-file race anymore — you may launch both subagents in parallel (single message, two Agent tool calls) rather than sequentially.

3. Each subagent's final message is JSON (see .claude/agents/rules/code-quality.md and .claude/agents/rules/audio-quality.md for the exact schemas):

   • code-tester-reviewer → {"tests": [...], "quality": [...], "log": "..."}
   • audio-quality-checker → {"tests": [...], "scores": {...}, "spectrogram_image": "...", "log": "..."}

   Merge them per the merge contract in .claude/skills/rules/test-fat-llama.md.

4. If audio-quality-checker's report has scores and spectrogram_image, update README.md per the README update contract in .claude/skills/rules/test-fat-llama.md (the scores table and the Spectrogram Results image reference). If either field is missing, skip this step and note why in the relayed result — don't write partial data.

5. Output the merged JSON as your final result, per the relaying policy in .claude/skills/rules/test-fat-llama.md.