You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Allow configurable model once available upstream #232
TL;DR: this should be implemented, but only after upstream github/gh-aw#17265 / github/gh-aw#17296 lands; today all workflow source files use hardcoded engine.model values and expose no model input.
Recommendation
Proceed with a follow-up implementation once upstream support is merged and released, by adding a workflow_callmodel input (with workflow-specific defaults) and wiring engine.model to $\{\{ inputs.model }} across gh-aw workflow sources.
Findings
Current workflow sources hardcode model in frontmatter, e.g.:
workflow_call inputs currently include fields like additional-instructions, setup-commands, etc., but no model input (example: .github/workflows/gh-aw-issue-triage.md:17-33).
Workflow READMEs do not document a model caller input (example table: gh-agent-workflows/issue-triage/README.md:25-31; also gh-agent-workflows/pr-review/README.md:25-33).
Caller examples currently only pass secrets and optional existing inputs, no model passthrough (e.g. gh-agent-workflows/issue-triage/example.yml:13-17, gh-agent-workflows/pr-review/example.yml:12-18).
Upstream status:
github/gh-aw#17265 is open and describes why expression-based model config is currently blocked.
github/gh-aw#17296 is open draft/WIP to support env-var-backed model passing for expressions.
Verification
I ran repository checks to confirm current state:
$ cd /home/runner/work/ai-github-actions/ai-github-actions &&echo"model lines in workflow sources: $(grep -R '^[[:space:]]*model:' .github/workflows/gh-aw-*.md | wc -l)"&&echo"workflows using inputs.model expression: $(grep -R '\$\{\{ inputs.model }}' .github/workflows/gh-aw-*.md | wc -l)"&&echo"workflow READMEs documenting model input row: $(grep -R -F '| `model` |' gh-agent-workflows/*/README.md | wc -l)"
model lines in workflow sources: 31
workflows using inputs.model expression: 0
workflow READMEs documenting model input row: 0
Detailed Action Plan
Wait for upstream support to merge/release (github/gh-aw#17265, github/gh-aw#17296), then bump GH_AW_VERSION in Makefile.
In each source workflow (.github/workflows/gh-aw-*.md), add a workflow_call input:
model (string, required: false)
default should preserve existing per-workflow behavior (e.g., gpt-5.3-codex for most, claude-opus-4.6 for gh-aw-upgrade-check.md).
Replace static engine.model literals with $\{\{ inputs.model }} in those same files.
Update docs to match caller behavior:
standard inputs section in docs/workflows/gh-agent-workflows.md:88-96
per-workflow input tables in gh-agent-workflows/*/README.md.
Update selected gh-agent-workflows/*/example.yml files to show an optional with.model override example.
Run make compile and make lint to regenerate lock workflows and validate.
Add a focused regression check to ensure compiled workflows contain caller-expression model wiring instead of hardcoded literals.
Activity
github-actions commentedon Feb 20, 2026
TL;DR: this should be implemented, but only after upstream
github/gh-aw#17265/github/gh-aw#17296lands; today all workflow source files use hardcodedengine.modelvalues and expose nomodelinput.Recommendation
Proceed with a follow-up implementation once upstream support is merged and released, by adding a
workflow_callmodelinput (with workflow-specific defaults) and wiringengine.modelto$\{\{ inputs.model }}across gh-aw workflow sources.Findings
.github/workflows/gh-aw-issue-triage.md:11-13.github/workflows/gh-aw-pr-review.md:14-16.github/workflows/gh-aw-upgrade-check.md:11-14(usesclaude-opus-4.6)workflow_callinputs currently include fields likeadditional-instructions,setup-commands, etc., but nomodelinput (example:.github/workflows/gh-aw-issue-triage.md:17-33).modelcaller input (example table:gh-agent-workflows/issue-triage/README.md:25-31; alsogh-agent-workflows/pr-review/README.md:25-33).modelpassthrough (e.g.gh-agent-workflows/issue-triage/example.yml:13-17,gh-agent-workflows/pr-review/example.yml:12-18).github/gh-aw#17265is open and describes why expression-based model config is currently blocked.github/gh-aw#17296is open draft/WIP to support env-var-backed model passing for expressions.Verification
I ran repository checks to confirm current state:
Detailed Action Plan
github/gh-aw#17265,github/gh-aw#17296), then bumpGH_AW_VERSIONinMakefile..github/workflows/gh-aw-*.md), add aworkflow_callinput:model(string, required: false)gpt-5.3-codexfor most,claude-opus-4.6forgh-aw-upgrade-check.md).engine.modelliterals with$\{\{ inputs.model }}in those same files.docs/workflows/gh-agent-workflows.md:88-96gh-agent-workflows/*/README.md.gh-agent-workflows/*/example.ymlfiles to show an optionalwith.modeloverride example.make compileandmake lintto regenerate lock workflows and validate.Related Items
.github/workflows/gh-aw-issue-triage.md:11-33.github/workflows/gh-aw-upgrade-check.md:11-14docs/workflows/gh-agent-workflows.md:88-96gh-agent-workflows/issue-triage/README.md:25-31What is this? | From workflow: Issue Triage Agent
Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #232