mirror of
https://github.com/getcargohq/cargo-skills.git
synced 2026-09-14 19:17:15 +08:00
dependabot[bot]
cda2ee39cb
Bumps [actions/checkout](https://github.com/actions/checkout) from 4.4.0 to 7.0.1. - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](https://github.com/actions/checkout/compare/11d5960a326750d5838078e36cf38b85af677262...3d3c42e5aac5ba805825da76410c181273ba90b1) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: 7.0.1 dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
64 lines
2.3 KiB
YAML
64 lines
2.3 KiB
YAML
name: routing-evals-llm
|
|
|
|
# The real-model tier of the routing evals: a model is shown nothing but the
|
|
# skill descriptions and asked to route each prompt in evals/routing.jsonl.
|
|
#
|
|
# Kept out of skills-lint.yml on purpose — it costs money and is
|
|
# non-deterministic, so it must never be the thing standing between a typo fix
|
|
# and a merge. It runs only when routing could actually have changed, plus
|
|
# weekly and on demand.
|
|
|
|
on:
|
|
pull_request:
|
|
paths:
|
|
- "*/SKILL.md"
|
|
- "evals/routing.jsonl"
|
|
- ".github/scripts/routing-eval.ts"
|
|
- ".github/workflows/routing-evals-llm.yml"
|
|
push:
|
|
branches: [main]
|
|
paths:
|
|
- "*/SKILL.md"
|
|
- "evals/routing.jsonl"
|
|
- ".github/scripts/routing-eval.ts"
|
|
schedule:
|
|
# Weekly, so model drift shows up even when nobody touches a description.
|
|
- cron: "17 6 * * 1"
|
|
workflow_dispatch:
|
|
inputs:
|
|
model:
|
|
description: "Model id to route with"
|
|
required: false
|
|
default: "claude-haiku-4-5-20251001"
|
|
|
|
permissions:
|
|
contents: read
|
|
|
|
jobs:
|
|
llm-routing:
|
|
name: Route prompts with a real model
|
|
runs-on: ubuntu-latest
|
|
# A missing key (forks, or before the secret is set) should skip, not fail.
|
|
if: github.event.pull_request.head.repo.full_name == github.repository || github.event_name != 'pull_request'
|
|
steps:
|
|
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
|
|
|
|
- uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
|
|
with:
|
|
node-version: "22"
|
|
|
|
- name: Route every eval prompt from descriptions alone
|
|
env:
|
|
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
|
|
ROUTING_EVAL_MODEL: ${{ inputs.model || 'claude-haiku-4-5-20251001' }}
|
|
run: |
|
|
if [ -z "${ANTHROPIC_API_KEY}" ]; then
|
|
echo "ANTHROPIC_API_KEY is not available to this run — skipping the model tier." >> "$GITHUB_STEP_SUMMARY"
|
|
exit 0
|
|
fi
|
|
# Baseline on 2026-08-12 was 98% (92/94: core 79/79, hard 13/15).
|
|
# The floor sits below it, not at it — a model tier that fails on one
|
|
# non-deterministic flip trains people to ignore it. Raise the floor
|
|
# when the baseline moves up and holds.
|
|
node .github/scripts/routing-eval.ts --llm --llm-min=93
|