deps-dev(deps-dev): bump the patch-and-minor group with 3 updates

[dependabot]

Bumps the patch-and-minor group with 3 updates: @types/node, darwin-agents and tsx.

Updates @types/node from 22.20.0 to 22.20.1

Commits

• See full diff in compare view

Updates darwin-agents from 0.10.0 to 0.11.0

Release notes

Sourced from darwin-agents's releases.

v0.11.0 — GEPA budget discipline (skip-perfect feedback + merge cap)

Two opt-in "budget discipline" knobs adapted from GEPA (verified against the DSPy GEPA API and the gepa-ai engine source) to Darwin's forever-online loop. The values mirror upstream's defaults; the mechanisms are Darwin's own adaptation. Default behaviour is unchanged with the flags off.

Added

• skipPerfectFeedback (adapted from GEPA's skip_perfect_score, src/evolution/feedback-filter.ts). Upstream skips a whole reflection iteration when an entire sampled minibatch is perfect; Darwin generalizes it to per-report filtering, which it can afford because the critic scores on real runs are already paid for. A run that already scored a perfect critic score carries no improvement gradient — its "nothing to fix" report only dilutes the pool. When true, such reports are dropped from both the legacy optimizer feedback (getRecentFeedback) and the GEPA reflective feedback (getReflectiveFeedback); skipped items do not count toward the feedback window, so it fills with actionable reports. perfectFeedbackScore (default 10, the critic-scale max) sets the threshold — lower it to also skip near-perfect runs. A non-finite score is never treated as perfect, so a genuinely broken run still surfaces. If every recent run is perfect the reflective path falls back to the legacy stats optimizer and the legacy path proceeds on aggregate stats (so the loop keeps exploring). With useDemos on, perfect runs are still used — harvested as demonstrations. CLI: --skip-perfect / --no-skip-perfect. Pure helpers (isPerfectScore / resolvePerfectFeedbackScore / filterPerfectFeedback) are exported from the package root.
• maxMergeInvocations (adapted from GEPA's max_merge_invocations, default 5 upstream). A per-agent lifetime cap on merge-derived challengers, persisted in DarwinState.mergeInvocations (a process-scoped counter would reset every cron tick and never trigger). The GEPA paper leaves merge-budget allocation as open research; the reason Darwin needs a cap is its own — an uncapped mergeEveryK cadence would merge forever and crowd out reflective exploration late in an agent's life. The count is the number of merge challengers actually created — a merge that fails the alignment guard is not counted (it consumed no A/B slot) — and is written only when a cap is set, so an uncapped useMerge agent's persisted state is unchanged from v0.10. Once the cap is reached the merge branch is skipped and the loop falls back to the reflective path for the rest of the agent's life (the budget does not re-arm). Only consulted when useMerge is on. Left unset it is uncapped (v0.10 behaviour); set it to 5 to match GEPA's default. CLI: --max-merge <n> (non-negative integer; 0 disables merge).

Notes

• Both knobs flow through the existing override machinery (EvolutionConfigOverride + OVERRIDE_KEYS + resolveEvolutionConfig + describeOverride/describeConfig), so they are CLI-settable, persist across processes, and CLI overrides win over persisted ones.

... (truncated)

Changelog

Sourced from darwin-agents's changelog.

[0.11.0] — 2026-07-09

Two opt-in "budget discipline" knobs adapted from GEPA (verified against the DSPy GEPA API and the gepa-ai engine source) to Darwin's forever-online loop. The values mirror upstream's defaults; the mechanisms are Darwin's own adaptation. Default behaviour is unchanged with the flags off.

Added

• skipPerfectFeedback (adapted from GEPA's skip_perfect_score, src/evolution/feedback-filter.ts). Upstream skips a whole reflection iteration when an entire sampled minibatch is perfect; Darwin generalizes it to per-report filtering, which it can afford because the critic scores on real runs are already paid for. A run that already scored a perfect critic score carries no improvement gradient — its "nothing to fix" report only dilutes the pool. When true, such reports are dropped from both the legacy optimizer feedback (getRecentFeedback) and the GEPA reflective feedback (getReflectiveFeedback); skipped items do not count toward the feedback window, so it fills with actionable reports. perfectFeedbackScore (default 10, the critic-scale max) sets the threshold — lower it to also skip near-perfect runs. A non-finite score is never treated as perfect, so a genuinely broken run still surfaces. If every recent run is perfect the reflective path falls back to the legacy stats optimizer and the legacy path proceeds on aggregate stats (so the loop keeps exploring). With useDemos on, perfect runs are still used — harvested as demonstrations. CLI: --skip-perfect / --no-skip-perfect. Pure helpers (isPerfectScore / resolvePerfectFeedbackScore / filterPerfectFeedback) are exported from the package root.
• maxMergeInvocations (adapted from GEPA's max_merge_invocations, default 5 upstream). A per-agent lifetime cap on merge-derived challengers, persisted in DarwinState.mergeInvocations (a process-scoped counter would reset every cron tick and never trigger). The GEPA paper leaves merge-budget allocation as open research; the reason Darwin needs a cap is its own — an uncapped mergeEveryK cadence would merge forever and crowd out reflective exploration late in an agent's life. The count is the number of merge challengers actually created — a merge that fails the alignment guard is not counted (it consumed no A/B slot) — and is written only when a cap is set, so an uncapped useMerge agent's persisted state is unchanged from v0.10. Once the cap is reached the merge branch is skipped and the loop falls back to the reflective path for the rest of the agent's life (the budget does not re-arm). Only consulted when useMerge is on. Left unset it is uncapped (v0.10 behaviour); set it to 5 to match GEPA's default. CLI: --max-merge <n> (non-negative integer; 0 disables merge).

Notes

• Both knobs flow through the existing override machinery (EvolutionConfigOverride + OVERRIDE_KEYS + resolveEvolutionConfig + describeOverride/describeConfig), so they are CLI-settable, persist across processes, and CLI overrides win over persisted ones.

... (truncated)

Commits

• 33fce20 feat(v0.11.0): GEPA budget discipline — skip-perfect feedback + merge-invocat...
• f74041b deps-dev(deps-dev): bump tsx in the patch-and-minor group (#17)
• See full diff in compare view

Updates tsx from 4.23.0 to 4.23.1

Release notes

Sourced from tsx's releases.

v4.23.1

4.23.1 (2026-07-13)

Bug Fixes

• support tsImport after global preload (8d4ffc2)
• watch: avoid clearing piped output (95d0672)
• watch: treat script and dependency paths literally (79fddde)

Performance Improvements

• index transform cache lazily (e818ad6)
• load esbuild lazily in CLI (d067938)
• map Node TypeScript formats directly (cdcc623)
• use sync module hooks on Node v22.22.3+ (f8992f1)

---

This release is also available on:

• npm package (@​latest dist-tag)

Commits

• 79fddde fix(watch): treat script and dependency paths literally
• e818ad6 perf: index transform cache lazily
• cdcc623 perf: map Node TypeScript formats directly
• d067938 perf: load esbuild lazily in CLI
• 95d0672 fix(watch): avoid clearing piped output
• 6fd4607 docs: add per-page metadata
• f4176d8 docs: generate sitemap
• 8d4ffc2 fix: support tsImport after global preload
• f0e89b2 docs: document Node's public type-stripping API vs internal loader path
• f8992f1 perf: use sync module hooks on Node v22.22.3+
• See full diff in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore <dependency name> major version will close this group update PR and stop Dependabot creating any more for the specific dependency's major version (unless you unignore this specific dependency's major version or upgrade to it yourself)
• @dependabot ignore <dependency name> minor version will close this group update PR and stop Dependabot creating any more for the specific dependency's minor version (unless you unignore this specific dependency's minor version or upgrade to it yourself)
• @dependabot ignore <dependency name> will close this group update PR and stop Dependabot creating any more for the specific dependency (unless you unignore this specific dependency or upgrade to it yourself)
• @dependabot unignore <dependency name> will remove all of the ignore conditions of the specified dependency
• @dependabot unignore <dependency name> <ignore condition> will remove the ignore condition of the specified dependency and ignore conditions

[dependabot]

Labels

The following labels could not be found: dependencies. Please create it before Dependabot can add it to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.