Skip to content
PiinfraDevOps1 file · +9 −5

Set macOS-26 GitHub Actions runner in conda-forge feedstock

Session from conda-forge-agent-traces on Hugging Face by xhochy · BSD-3-Clause · condensed by Coders Talk

8msession
0interventions
GOAL

Switch a conda-forge feedstock's CI to use the macos-26 GitHub Actions runner via conda-forge.yml/recipe config, then re-render with conda-smithy.

  1. Prompt · +0
    Use macos-26 as the GitHub Actions runner; use conda-forge skill and search web on how to do this via conda-forge.yml or recipe/conda_build_config.yaml, then conda-smithy rerender.
  2. Agent did · +0m
    Inspected the feedstock's conda-forge.yml, recipe files, and conda-forge skill docs to understand current config.
  3. Agent did · +0m
    Searched conda-forge docs and conda-smithy source (configure_feedstock.py) to find how github_actions_labels / runs_on is set, and inspected .ci_support and workflow files.
  4. Agent did · +5m
    Edited recipe/conda_build_config.yaml to add github_actions_labels with osx/arm64 selector set to macos-26.
  5. Agent did · +7m
    Ran conda-smithy rerender to regenerate CI configs from the updated recipe.
  6. Agent did · +7m
    Verified changes by inspecting conda_build_config.yaml, .ci_support/osx_arm64_.yaml, the workflow file, git diff, and ran conda-smithy lint, confirming clean output.
  7. Outcome · +8m
    Feedstock updated: runs_on changed from macos-15 to macos-26 across recipe/conda_build_config.yaml, .ci_support/osx_arm64_.yaml, and .github/workflows/conda-build.yml; conda-smithy lint passed clean.
VERDICT

I'd tell the agent upfront to check conda-smithy's GITHUB_ACTIONS_RUNS_ON mapping directly instead of spending several web searches first, since the source code fetch is what actually gave the answer. No corrections were needed here—the agent researched, edited the config, re-rendered, and the lint passed clean on the first try.

I tried this · 0

Discussion · 0

How Builds work