Kiln Model Sweep
The scheduled model sweep.
Install
npx skills add Kiln-AI/Kiln --skill kiln-model-sweep -g -a claude-code -y- Repository
- Kiln-AI/Kiln5.2k
- Category
- Testing
- Updated
- 3d ago
- License
- See repository
- Context cost
- Mediumabout 3,000 tokens
- Runs scripts
- No, instructions only
What it does
The scheduled model sweep. Reads the team's models channel for hints, discovers new LLM models, adds them to ml_model_list.py, runs the paid tests, marks dead provider entries deprecated, opens PRs under the team's rules, announces them in Slack, and checks whether remote_config needs a publish PR. Use when asked to run the model sweep by hand, to add or deprecate models unattended, or to change how the daily sweep behaves.
Kiln model sweep
The prompt a scheduled cloud routine runs each weekday morning. It is a thin wrapper: the procedure lives in claude-maintain-models, kiln-check-deprecation and open-pr; this file pre-answers their gates so a run finishes with nobody watching. The design, the reasons behind each rule, and the smoke-test record are in specs/projects/model_sweep/.
Run settings
The routine's message, or the person running the sweep by hand, supplies these. Nothing personal is hard-coded here.
- The models channel to read for hints: default
#models. SLACK_MODELS_WEBHOOK: incoming-webhook URL for draft announcements, as an environment variable. When missing, the post is skipped and the report says so.REVIEWERS: GitHub logins (ororg/teamslugs) to request on every ready PR. The team's PR bot posts the review card in the PR channel for any open, non-draft PR with reviewers requested, so the sweep never posts there itself. Never includes the account the sweep runs as; GitHub refuses a PR's author as reviewer. When unset, no reviewer is requested and the report says so.SLACK_CC_USER_ID: the Slack user id to cc in draft announcements. When unset, the cc line is omitted.- Provider API keys as environment variables, with the names Kiln's
Configuses.
The text below is what the routine runs.
You are the Kiln model sweep, running unattended on a schedule. Nobody will answer a question; where a skill would stop to ask, use the pre-answered rules below.
In the Kiln-AI/Kiln checkout, run git fetch origin and branch from origin/main. Read and follow, in this order:
.agents/skills/claude-maintain-models/SKILL.md, including its Phase 1 discovery for models released in the last 14 days and its Phase 1B lagging-provider backfill..agents/skills/kiln-check-deprecation/SKILL.md..agents/skills/open-pr/SKILL.mdfor every PR.
