Fix bare Claude model alias resolution in OpenCode converter

[waltbeaman]

Thank you all for this amazing plugin and especially for the port to opencode! I use this tool daily at work and it has quickly become my favorite way to do agentic coding. Appreciate Kieran, Every Inc, and all the maintainers of this awesome project.

Summary

• Fix ProviderModelNotFoundError when OpenCode invokes agents that use bare model aliases (e.g., model: haiku)
• Add CLAUDE_FAMILY_ALIASES map to resolve haiku, sonnet, opus to full model IDs
• Warn during conversion when a bare alias is used, so users know to update if newer models are available

Problem

When using the compound-engineering plugin in OpenCode, running /workflows:plan fails with:

ProviderModelNotFoundError: ProviderModelNotFoundError

This happens because:

1. The learnings-researcher agent (and lint agent) use model: haiku in their Claude Code frontmatter -- a bare alias that Claude Code resolves internally
2. The normalizeModel() function in the OpenCode converter turns haiku into anthropic/haiku via its catch-all fallback (return \anthropic/${model}``)
3. OpenCode requires fully-qualified model IDs (e.g., anthropic/claude-haiku-4-5). The bare anthropic/haiku doesn't exist in its model registry

The error occurs every time /workflows:plan runs because it always invokes learnings-researcher in Step 1 (Local Research).

Fix

Add a CLAUDE_FAMILY_ALIASES map to normalizeModel() that resolves bare Claude family names (haiku, sonnet, opus) to their full model IDs before the catch-all fallback. A console.warn alerts during conversion so the map can be updated when new model versions are released.

Changes

File	Change
src/converters/claude-to-opencode.ts	Add CLAUDE_FAMILY_ALIASES map and alias resolution to normalizeModel()
tests/fixtures/sample-plugin/agents/bare-alias-agent.md	New fixture agent with model: haiku
tests/converter.test.ts	Test that haiku resolves to anthropic/claude-haiku-4-5
tests/claude-parser.test.ts	Update agent count expectation (2 -> 3) for new fixture

Test Plan

All 87 tests pass:

 87 pass
 0 fail
 308 expect() calls
Ran 87 tests across 12 files. [562.00ms]

[kieranklaassen]

THANKS!