Tier

Use this skill when the user wants to rate the current coding tool with /tier.

Examples:

• /tier s
• /tier a clear, direct, strong patch loop
• /tier meh
• /tier bad broke the workflow

Workflow

1. Parse the command as /tier <rating> [comment...].
2. Normalize <rating> using:
   • s -> S
   • a, good -> A
   • b -> B
   • c, meh -> C
   • d -> D
   • f, bad -> F
3. Run the bundled vote runner:

python3 scripts/cast_vote.py a clear, direct, strong patch loop

4. The vote runner detects context first.
   • In Codex, it reads only model and model_reasoning_effort from ~/.codex/config.toml.
   • Outside Codex, it does not read ~/.codex/config.toml.
   • It uses the detected tool_slug as the primary vote target when present.
   • If the tool is unknown but the model is known, it falls back to the model slug.
   • If neither can be determined, it exits with an error instead of guessing.
5. Use --dry-run --json when you want to inspect the exact request before sending it:

python3 scripts/cast_vote.py --dry-run --json s strong edits

6. Give a short confirmation with:
   • whether the vote was counted or over_quota
   • which targets were counted
   • the current tier and total votes for the primary item

Response Handling

• If the API returns 201, summarize the result briefly.
• If the API returns 429, tell the user voting is rate-limited.
• If the API returns 400, surface the validation error directly.

Notes

• The backend deduplicates targets, stores one logical vote, and can count it for both tool and model.
• Unknown model slugs are still useful as source metadata, but only known catalog items are counted publicly.
• When the detector returns model_reasoning_effort, include it only in local confirmation text for now. The production vote payload does not store it yet.