Agent skill
The vibetracker-rate skill lets compatible agent harnesses submit honest ratings for AI sessions through vtcli.
Install
Install and authenticate vibetracker-cli first. The skill uses the CLI for auth, validation, submission, and sentiment output.
bash
npm install -g vibetracker-cli
vtcli auth login
vtcli skill install| Field | Type | Required | Description |
|---|---|---|---|
| vtcli skill install | Install vibetracker-rate into the current project when possible | ||
| vtcli skill install --global --agent codex | Install globally for Codex | ||
| npx skills add https://github.com/Skeegan123/vibetracker-rate --skill vibetracker-rate | Install directly through the open skills installer |
Rating workflow
Ask your agent to rate the current model session as good, okay, or bad. The skill maps those words to 1, 0, or -1, preserves your comment exactly, and avoids guessing optional context when it is unclear.
text
Rate this Codex session as bad.
Comment: It lost track of the requested file changes.If the model is missing or ambiguous, the skill should ask one short question or use
vtcli options list --type model --search <query> to inspect candidates.Submission fields
The skill deliberately keeps submissions small. It only includes optional context when the transcript or harness exposes a high-confidence match.
| Field | Type | Required | Description |
|---|---|---|---|
| model | string | Yes | Model identifier from the harness or user request |
| score | -1 | 0 | 1 | Yes | Negative, neutral, or positive session rating |
| comment | string | No | Exact human wording when a comment is provided |
| useCase | string | No | Included only when the conversation clearly supports it |
| interface | string | No | Included only when the agent can infer it confidently |
| toolId | string | No | Included only when the selected interface/tool is clear |
For CLI command details, including vtcli skill install, see the CLI docs.