[squad-plan] Investigate and fix Avenger model mismatch between frontmatter and executed engine model

[github-actions]

Objective

Find and fix why Avenger (.github/workflows/avenger.md) declares model: claude-haiku-4.5 but audited runs execute on claude-opus-4-8.

Issue context

Issue #50312 found the largest cost driver on the Avenger workflow is a model mismatch: frontmatter/intended budget points to Haiku, but agent_usage.json shows Opus 4.8 in all 3 sampled runs (5x more expensive per token). This suggests a compiler/runtime bug in model propagation, alias resolution, or engine argument generation.

Implementation guidance

• Start in compiler/engine code, likely pkg/workflow and possibly pkg/cli audit/reporting paths.
• Trace the model value through:
  1. frontmatter model
  2. engine config extraction
  3. activation job outputs / aw_info.json
  4. runtime engine env/args passed to the Claude engine
  5. any alias expansion/fallback behavior
• Confirm whether the bug is:
  • compile-time model dropped/overwritten
  • alias resolved incorrectly
  • aw_info/audit reporting wrong while runtime is actually correct
  • engine silently defaulting to a higher-tier model
• Add a targeted regression test for the failing path.

Acceptance criteria

• Compiling Avenger preserves the configured model through activation outputs and engine invocation.
• A targeted regression test fails before the fix and passes after.
• aw_info.json model metadata and actual runtime/usage model data are consistent after the fix.
• No unrelated engine/model behavior regresses in targeted tests.

Dependency / ordering notes

Do this first — highest-impact cost fix (~70-80% of AIC/run) and other Avenger changes should be validated with this fix in place.

Generated by 🧑‍🤝‍🧑 Squad Plan · auto · 37.1 AIC · ⌖ 2.24 AIC · ⊞ 9.5K · ◷
_{Comment /squad-plan to run again}