Skip to content

v3.4.0

Choose a tag to compare

View all tags
@imewei imewei released this 19 Apr 17:56
· 230 commits to main since this release
v3.4.0
6a9e008

v3.4.0 (2026-04-19)

New Plugin: research-suite

Fourth plugin suite extracted from science-suite. Three workflow tracks:

  • scientific-review — journal-ready peer review → .docx referee report
  • research-spark — 8-stage artifact-gated refinement pipeline
    (articulate → landscape → claim → theory → prototype → experiment → premortem)
  • research-practice — methodology hub (design, reproduce, evaluate, write, synthesize)

Ships 2 opus agents (research-expert, research-spark-orchestrator),
4 registered hubs → 12 sub-skills, 3 hook events (SessionStart artifact-resume,
TaskCompleted audit log, SubagentStop prompt-based stage gate).

Opus 4.7 Sweep

  • context-specialist elevated from sonnet → opus (1M context)
  • 7 fabricated hook event handlers removed (not supported by CC v2.1.113
    schema): PreSubagentUse, ExecutionError, PermissionPrompt, ContextOverflow,
    CostThreshold across agent-core, dev-suite, science-suite
  • Hook event total across suites: 27 (agent-core 12, dev-suite 7,
    research-suite 3, science-suite 5)

science-suite Focus

Now purely computational (JAX, Julia, HPC, ML/DL, physics, nonlinear dynamics):
11 agents (was 12), 14 hubs → 112 sub-skills (was 117). research-expert and
5 methodology skills moved to research-suite.

Version Consistency

All 13 canonical version surfaces aligned at 3.4.0 (pyproject.toml,
docs/conf.py, Makefile, marketplace.json, 4 plugin manifests, 4 suite READMEs,
root README, agent-core/commands/team-assemble.md).

Tooling Hardening

  • `doc_checker.py`: code-block awareness for heading-space check (closes 17
    false positives); placeholder word-match relaxed (closes 3); per-plugin
    iteration wired into `make validate`
  • `skill_validator.py`: reports `n/a ⚪ no corpus` instead of misleading
    `0.0% ❌` when no test corpus is loaded
  • 8 new regression tests in `test_doc_checker.py`

Validator State (Final)

  • metadata_validator: 0 errors / 0 warnings on all 4 suites
  • doc_checker: 0 warnings across all 4 suites (was 73 at session start)
  • xref_validator: 531/531 valid (0 broken)
  • context_budget_checker: 217/217 skills fit 200K budget
  • pytest: 188/188 passing (was 154 in v3.3.0)
  • ruff + mypy: clean
  • Security audit (pip-audit + gitleaks + bandit + vulture): clean

Stats at a Glance

  • 4 suites (was 3)
  • 25 agents (11 opus, 13 sonnet, 1 haiku)
  • 14 registered commands + 21 skill-invoked reference templates
  • 31 hub skills → 186 sub-skills (217 SKILL.md files total)
  • 27 lifecycle hook events
Assets 2
Loading