Skip to content

✨ feat: add artifact-centered team collaboration - #3

Open
pftg wants to merge 24 commits into
masterfrom
vcr-failure-integration
Open

✨ feat: add artifact-centered team collaboration#3
pftg wants to merge 24 commits into
masterfrom
vcr-failure-integration

Conversation

@pftg

@pftg pftg commented Aug 27, 2026

Copy link
Copy Markdown
Member

Summary

  • make Team valuable for named coworker registration, immutable artifact handoffs, atomic call budgets, thread/fiber fan-out, normalized failures, revision lineage, and inspectable traces
  • add a thin Team#run API while keeping routing, review loops, validation, and escalation in ordinary application Ruby
  • simplify the blog workflow around Team artifacts: current You.com MCP-backed research, a deliberately weak writer, GPT editors, bounded feedback, senior-writer escalation, and final Ruby API validation
  • keep the simple offline team example as cross-domain proof of the same Run and artifact primitives
  • replace copied workflow coverage with behavioral VCR tests that prove a bad draft receives exact feedback, is reworked, escalates after bounded failure, and is reviewed again
  • add a Tribunal eval beside the blog and preserve the live post and collaboration trace
  • supersede generic refine and repair with a concise ADR and v5 roadmap requiring a second non-writing workflow before further lifecycle extraction

Product boundary

Team owns reusable collaboration state and safety. RubyLLM owns models, tools, provider retries, timeouts, and instrumentation. Application Ruby owns workflow order, quality policy, validators, escalation, persistence, and approvals. MCP and You.com remain example integrations, not core gem dependencies.

Verification

  • bundle exec rake spec — 49 examples, 0 failures
  • bundle exec rubocop — 20 files, no offenses
  • gem build ruby_llm-team.gemspec — success; generated package removed afterward
  • strict OKF validation — conformant, no issues
  • live OpenRouter workflow — completed research, weak-writer revisions, senior escalation, fact re-audit, publication review, and Ruby validation
  • bundle exec rake eval:blog — deterministic, relevance, faithfulness, and hallucination gates all pass
  • VCR cassette contains only the three current hosted-review interactions

Review entry points

  • README.md — why Team earns a Gemfile line and how the examples use it
  • examples/blog/workflow.rb — explicit policy over Team artifacts
  • examples/blog/research.rb — shared MCP-backed search and extraction tool
  • examples/blog/trace.md — exact live collaboration and escalation path
  • docs/decisions/2026-08-28-keep-quality-loops-in-application-ruby.md — current product decision
  • docs/roadmaps/ROADMAP-v5.md — next cross-domain evidence plan

@pftg pftg changed the title Add local blog workflow example ✨ feat: add reliable multi-agent blog workflow Aug 27, 2026
pftg and others added 19 commits August 27, 2026 22:46
Make the ADR self-contained for a cold-eye reviewer: verified blog
accounting (19 asks by role, 0 parallel, 0 tools, ~77 gate lines),
upstream evaluator-loop and semaphore evidence, option D, objections
section, kill switches, and complete remote/internal source tables.

🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
Add the upstream evaluator-loop and Async::Semaphore snippets to §4.2
so the Option A rejection rationale is verifiable in the document
itself, and record the change in the ADR changelog.

🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
Keep .claude, .claude-flow, and .junie harness state out of git so it
cannot be staged accidentally.

🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
@pftg pftg changed the title ✨ feat: add reliable multi-agent blog workflow ✨ feat: add artifact-centered team collaboration Aug 27, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant