Skip to content

Releases

v0.1.0 — initial release

Choose a tag to compare

@fadymondy fadymondy released this 29 Apr 11:36

Initial release — three pillars: generator, /meet, evaluator.

Added

  • /team-gen skill — generate a complete .claude/ agent team (orchestrator + specialists, skills, rules, hooks) from a product description. Picks archetypes, fills templates, scaffolds, runs the static linter on every produced file before exit. End-to-end test: a 4-agent Sample Shop team scores A 100/100 across the board. (#4)
  • /meet skill — streamed multi-agent meeting on a topic. Each agent answers in a structured format (constraint check, top-3 risks, dependencies, open questions, effort). Orchestrator consolidates to a global top-10 and writes minutes to .meetings/YYYY-MM-DD-{topic}.md. (#7)
  • /evaluate-agent skill — static linter (Phase 1) — citation-backed deterministic rules across six dimensions (frontmatter, description, tool hygiene, model fit, body structure, anti-patterns). Runs on all 14 sentra-hub + 42 orch agents without crashing. Every rule cites Anthropic spec or published prior art. (#10)
  • /evaluate-agent --deep — LLM-as-judge (Phase 2, skeleton) — per-dimension isolated calls, evidence-before-score, file-hash + model + rubric-version cache. Judge model swappable via ANTHROPIC_JUDGE_MODEL (default: claude-sonnet-4-6). (#11)
  • /evaluate-agent-behavior — behavioral harness (Phase 3, skeleton) — replay grader scores tool-whitelist adherence and step efficiency from a transcript. (#12)
  • Output schema v1 + Markdown renderer — single shape for all three evaluator phases. JSON Schema + jq-based renderer. Exit codes 0/1/2 for ship/revise/reject. (#9)
  • 8 agent + 4 skill archetypes — orchestrator, tech-leader, domain-engineer, designer, qa-engineer, security-engineer, devops-engineer, monitor; domain-skill, coordination-skill, lint-skill, init-skill. (#3)
  • 13 numbered rule templates — plan-first, service-boundaries, definition-of-done, etc. Each opens with a "When this applies" header so /team-gen can pick a subset. (#5)
  • 4 hook templates — notify, session-init, teammate-idle-gate, post-commit-check. (#6)
  • Plugin scaffolding — plugin.json, layout, plugin CLAUDE.md, root README, MIT LICENSE. (#2)
  • Documentation — root README + docs/{team-gen,meet,evaluator,contributing}.md. (#8)

Repo plumbing

  • gh-pms labels (32) + 7 issue templates + PR template.
  • GitHub Sponsors link for @fadymondy.
  • CHANGELOG with auto-update via git-cliff.

What's next

See v0.2 milestone — quality gates, calibration, real behavioral grader.