chore(deps): bump umm-actually to v0.4.0

[aliasunder]

Summary

Bump umm-actually from v0.3.14 to v0.4.0 (65ccbe7).

Changes in v0.4.0

• feat: staged review phases — parallel and sequential modes (#79)
• feat: content-based cross-run dedup for shifted/reworded findings (#85)
• fix: content dedup follow-ups — coalesceAnchors, titleSimilarity, logging (#86)
• fix: filter self-negating findings (#84)
• chore: bump default context_budget_tokens from 80K to 300K (#87)
• docs: drop the early-development status section (#83)

[umm-actually]

Align the context_budget_tokens default with the 300K release default
Low severity · subtle bugs · medium confidence

.github/workflows/umm_review.yml:130 — beyond the diff's line ranges, in code the changes touch or depend on.

PR description says the default context_budget_tokens was bumped from 80K to 300K, but this workflow still pins an 80000 default when the UMM_CONTEXT_BUDGET_TOKENS variable is unset. The config therefore does not use the released default and the surrounding comment describes a value that is not what runs.

Failure scenario: A contributor opens a PR on a repo where UMM_CONTEXT_BUDGET_TOKENS is not defined (the common case) and the action's shared context pool is capped at 80K tokens instead of the v0.4.0 default 300K — the increased budget that the new default was meant to provide, and that the comment claims, is never applied.

Suggested fix

Update the fallback to '300000' in the context_budget_tokens input (e.g. `${{ vars.UMM_CONTEXT_BUDGET_TOKENS || '300000' }}`) or, if an explicit 80K cap is deliberate, update the PR description and the workflow comment accordingly.

---

umm-actually · deepseek/deepseek-v4-flash-0731

[umm-actually]

umm-actually re-reviewed at feac09b

No new findings (1 tracked finding(s) across all runs).

---

umm-actually · deepseek/deepseek-v4-flash-0731

[aliasunder]

Fixed — the context_budget_tokens fallback was updated from '80000' to '300000' in the same branch (commit feac09b). The re-review at that commit confirms no new findings.

---

🔍 ship-check · pr-monitor · claude-opus-4-6