GPT-5 finding
Per-skill model override is not detected due to incorrect YAML parsing
mediumbughigh
- .github/workflows/aeon.yml
The code greps only the skill header line (e.g., " my-skill:") and tries to sed a model=… value from that same line. In valid YAML the model key is on a subsequent indented line, so SKILL_MODEL remains empty even when a per-skill override exists. This silently falls back to the default model, contradicting the comment and README behavior.
Recommendation
Parse the model key from the indented block. For example: - grep-based: SKILL_MODEL=$(awk -v s="$SKILL_NAME" ' $0 ~ "^ " s ":" {inblk=1; next} inblk && $0 ~ "^ [^ ]" {inblk=0} inblk && $1 ~ /model:/ {gsub(/^[^:]*:[[:space:]]*|"/ ,"", $0); print $0; exit} ' aeon.yml) - Or use yq if available: command -v yq >/dev/null && SKILL_MODEL=$(yq -r ".skills[\"$SKILL_NAME\"].model // \"\"" aeon.yml) || true