v0.3.4

Fix: Fleet was silently stuck on stale Claude model IDs

The problem

apra-fleet, and its vendored apra-pm workflow, hardcoded dated Claude model IDs (e.g. claude-sonnet-4-6) for the cheap/standard/premium tiers. As a result, even after Anthropic released Claude Sonnet 5, the fleet kept dispatching the old Sonnet 4.6 model for every task. This was confirmed via a live end-to-end run that showed stale model usage despite the new model being available.

The fix

• Replaced pinned dated model IDs with the bare Claude model-family aliases (haiku / sonnet / opus).
• Both the Claude CLI and settings.json's defaultModel resolve these aliases to the current model generation automatically, so no code change is needed when a new model ships.
• Verified end-to-end: after updating the CLI on the runners, execute_prompt with the standard tier resolved to claude-sonnet-5.

Why this matters

• Immediate: Claude Sonnet 5 can now actually be used by the fleet.
• Future-proof: subsequent releases (Sonnet 6, Opus 5, etc.) will be picked up automatically via the alias mechanism, with no further hardcoded-ID hunting or patching across the codebase.

Vendored dependency

The vendored apra-pm submodule was updated to pull in the equivalent upstream fix: Apra-Labs/apra-pm#16.

---

What's Changed

• fix(models): use bare Claude model aliases instead of pinned dated IDs (v0.3.4) by @kumaakh in #316

Full Changelog: v0.3.3...v0.3.4