feat: consolidate evolution gates stack (#55-#61)

[steezkelly]

Summary

Consolidation of evolution-gates PR stack: #55, #56, #57, #58, #59, #60, #61.
Also supersedes the narrower consolidation PR #67 (which covered #60-#66, now fully absorbed here).

This is a lower-overhead alternative to reviewing 7 overlapping PRs individually. All merged locally (conflicts resolved), tests pass clean.

What's included (18 files, 1441 additions, 95 deletions vs upstream origin/main)

From #55: ingestion reports and promotion gates

• evolution/core/benchmark_gate.py (new)
• evolution/core/dataset_builder.py (updated)
• evolution/core/external_importers.py (updated)
• evolution/core/pr_builder.py (new)
• evolution/core/run_report.py (new)
• evolution/skills/evolve_skill.py (updated — wiring for all gates)
• tests/core/test_issue54_ingestion.py (new)
• tests/core/test_issue54_promotion.py (new)

From #56: DSPy 3.x GEPA construction fix

• evolution/skills/evolve_skill.py (GEPA optimizer helper)
• tests/skills/test_evolve_skill_gepa.py (new)

From #57: LLM judge feedback for skill fitness

• evolution/core/fitness.py (rubric-based LLM judge scoring)
• evolution/skills/evolve_skill.py (wired into metric)
• tests/core/test_fitness.py (new)

From #58: enforce run-tests evolution gate

• evolution/core/constraints.py (pytest gate)
• evolution/skills/evolve_skill.py (wired into pipeline)
• tests/core/test_constraints.py (updated)
• tests/skills/test_evolve_skill_gates.py (new)

From #59: reject empty holdout datasets

• evolution/skills/evolve_skill.py (holdout gate)
• tests/skills/test_evolve_skill_dataset_gates.py (new)

From #60: declare reportlab dependency

• pyproject.toml (reportlab>=4.0)
• tests/test_generate_report.py (new)

From #61: fail fast on invalid baseline skills

• evolution/skills/evolve_skill.py (baseline constraint gate)
• tests/skills/test_evolve_skill_constraint_gates.py (updated)

Conflict resolution

Two conflict sites in evolution/skills/evolve_skill.py:

1. fix: enforce run-tests evolution gate #58 vs fix: update GEPA construction for DSPy 3 #56: Both added helper functions in the same region. Resolution keeps _create_gepa_optimizer(), _run_pytest_gate_if_requested(), and _save_failed_variant() — all independently needed.

2. fix: reject empty holdout datasets #59 + fix: fail fast on invalid baseline skills #61 vs earlier merges: Added _require_non_empty_holdout(), _require_constraints_pass(), and _validate_baseline_constraints() — all preserved.

No other conflicts. Merge order used: #55 → #56 → #57 → #58 → #59 → #60 → #61.

Local test evidence

GitHub reports no configured checks on any of these PRs (statusCheckRollup length 0). Local verification:

Targeted tests (41 tests covering all changed modules):
  41 passed, 11 warnings in ~2.3s

Full suite:
  164 passed, 11 warnings in ~2.4s

Warnings are DSPy deprecation noise (prefix argument in InputField/OutputField) from installed dependency paths only — not from changed code.

Test environment: local venv .venv-review, installed via pip install -e '.[dev]' at commit 74f5da3.

Superseded PRs

PR	Status	Action
#55	feat: add ingestion reports and promotion gates	→ close (absorbed here)
#56	fix: update GEPA construction for DSPy 3	→ close (absorbed here)
#57	fix: use LLM judge feedback for skill fitness	→ close (absorbed here)
#58	fix: enforce run-tests evolution gate	→ close (absorbed here)
#59	fix: reject empty holdout datasets	→ close (absorbed here)
#60	fix: declare reportlab dependency	→ close (absorbed here)
#61	fix: fail fast on invalid baseline skills	→ close (absorbed here)
#67	earlier #60-#66 consolidation	→ close (fully absorbed here)

Local evidence file at issue-55-61-superstack-local-evidence.md in the fork worktree.

Post-merge

Closes #54 (ingestion and promotion gates slice is now a single reviewable unit).

[steezkelly]

Closing — unintentional duplicate of the earlier #68 (already closed). Active development has moved to steezkelly/hermes-agent-self-evolution where PRs #8-#16 have been merged. See https://github.com/steezkelly/hermes-agent-self-evolution for current state.