Rote Classifier
Decide whether an issue is ROTE — safe to run plan, implement, PR and auto-merge with no human review — or needs a person. Jev (TypeSafe System One, through an @swamp/typesafe-ai definition) answers a calibrated 15-question rubric about each issue, and a fail-closed gate turns the probabilities into ROTE or HUMAN with the blockers named. Source-agnostic: classify takes plain {id, title, body, type} records from any tracker. Ships rubric v6 and its gate (holdout precision 0.97, recall 0.97 on hand-labelled issues), both overridable for calibration.
First publish: classify issues from any source as ROTE or HUMAN. Jev answers rubric v6 (15 questions) through an @swamp/typesafe-ai definition; the v6 docs-branch gate decides and names its blockers. Bundled rubric and gate overridable for calibration.
Global Arguments
| Argument | Type | Description |
|---|---|---|
| jev? | string | Name of the @swamp/typesafe-ai definition that answers the rubric. Its own apiKey (a vault reference) is used. |
| model? | string | Jev checkpoint to ask (e.g. jev-latest). Defaults to the jev definition's own. |
| Argument | Type | Description |
|---|---|---|
| issues | array | |
| name? | string | Instance name of the run summary. |
| questions | preprocess | Override the bundled v6 rubric (question id → {type, instructions, criteria}). For calibration. |
| gate | preprocess | Override the bundled v6 gate. For calibration. |
Resources
- Has README or module doc2/2earned
- README has a code example1/1earned
- README is substantive1/1earned
- Most symbols documented1/1earned
- No slow types (deprecated)1/1earned
- Dependencies pass trust audit2/2earned
- Docs keep secrets in a vault1/1earned
- Has description1/1earned
- Platform support declared (or universal)2/2earned
- License2/2earned
- Verified public repository2/2earned