Skip to content
/
Phase B Batch 20

Batch 20 Notes

10 frameworks in this batch

Batch 20 Notes

Roster

# slug display_name stars distribution_type tier key_differentiator
1 apm-agentic-project-mgmt APM 106 markdown-scaffold A Handoff Protocol for fresh-agent-instance transfer; Implementation Plan → Task hierarchy; Manager/Sub-Agent/Human roles; MIT
2 aigon Aigon 10 npm-cli A Fleet mode = parallel agent racing in separate git worktrees; supports 7 AI CLI tools (cc/gg/cx/cu/op/km/am); AutoConductor Pro tier; port 4100 dashboard
3 tessl Tessl 38 tile-registry A Versioned "tile" methodology bundles; tessl eval run . evaluates agent compliance against 9 scenarios; founded by Guy Podjarny (Snyk); MIT
4 flow-free Flow (free) N/A reddit-only C No public repo or material; UNCONFIRMED in reddit-mentions-verified.md
5 multitable MultiTable N/A reddit-only C No public repo or material; UNCONFIRMED in reddit-mentions-verified.md
6 claudeautopilot claudeAutoPilot 0 node-scripts A 12 Claude Code slash commands + 10 Node.js scripts; HTML kanban board at localhost:3003; PRD-driven 4-column kanban; Claude Code only
7 saflib-workflows saflib-workflows 0 npm-playground A TypeScript defineWorkflow+step DSL; makeWorkflowMachine for sub-workflow composition; saf-workflow CLI; playground-state (no license)
8 worca WORCA 26 pip+npm A 9-stage pipeline; 8 named agents (Planner/Coordinator/Implementer/Tester/Reviewer/Guardian/Learner); 10 Python hooks using sys.exit() for physical blocking; multi-model (Opus strategic + Sonnet execution); WebSocket dashboard
9 docbrain DocBrain N/A docker+helm A Shift-left documentation platform; 13+ capture sources; 3-layer quality scoring (structural×0.4 + style×0.3 + semantic×0.3); DBSCAN clustering; confidence routing; BSL 1.1 pre-source release
10 patchwork-os Patchwork OS 1558 pip A CI/CD-first automated code patching; 11 patchflows; 46+ reusable Steps; AGPL-3.0 framework + Apache-2.0 steps; creates GitHub PRs autonomously

Intra-Batch Patterns

This batch is the most heterogeneous in the corpus: it spans six distinct archetypes across ten frameworks. Three items (APM, claudeAutoPilot, saflib-workflows) are pure developer-workflow scaffolds with no LLM-specific runtime; three (WORCA, Aigon, tessl) add genuine multi-agent or evaluation runtime; two (DocBrain, Patchwork OS) are not coding-methodology frameworks at all — they are infrastructure tools that happen to use LLMs.

The two Reddit-only entries (flow-free, multitable) were UNCONFIRMED in the reddit-mentions-verified.md index. Both received Tier C treatment: 00-summary.md + METRICS.yaml only. This is the first batch where the Tier C escape has been needed more than once.

WORCA's governance model is the most mechanically rigorous in this batch: 10 Python hook scripts call sys.exit() to physically prevent dangerous operations at the OS process level — a design choice that no seed framework matches. WORCA also has the most complete multi-model assignment strategy, explicitly routing Opus to strategic planning and Sonnet to execution, with model assignments configurable per agent role.

Patchwork OS (1558 stars) is the highest-starred framework in this batch but is the furthest from spec-driven development. Its inclusion appears to be a Reddit context error — users describing automated code maintenance pipelines as "spec-driven" loosely. Patchwork is a CI/CD automation tool, not an agent development methodology.

DocBrain is in a similar position: it is an organizational knowledge platform that coding agents query, not a framework for building agent workflows. Its closest seed is ccmemory, but at enterprise scale — 13+ connectors, confidence-scored routing, DBSCAN clustering, and multi-stage human review are all absent from any seed.

Most Interesting Finds

WORCA — Physical OS-level governance hooks: The combination of pre_tool_use.py + post_tool_use.py calling sys.exit() creates a guardrail that operates below the LLM layer. The AI cannot "reason its way around" a sys.exit() call — the process terminates. This is architecturally distinct from prompt-level constraints (which can be overridden by sufficiently clever model behavior) and represents the strongest safety enforcement mechanism in the batch.

Aigon — Fleet mode as parallel spec validation: Running identical specs in parallel across multiple AI CLI tools in separate git worktrees, then selecting the fastest successful completion, is a novel strategy for both speed and implicit cross-model validation. The model that completes first wins; failures in one worker don't block others.

Tessl — Eval-driven methodology validation: tessl eval run . running 9 predefined agent compliance scenarios against a checked-out tile turns methodology validation into a repeatable test suite. No other framework in this batch (or the known seed frameworks) has a built-in evaluation harness for the methodology itself rather than for the code it produces.

Tier C Items

slug reason
flow-free UNCONFIRMED in reddit-mentions-verified.md; no GitHub repo, no public material found
multitable UNCONFIRMED in reddit-mentions-verified.md; no GitHub repo, no public material found

Cross-References

  • APM Handoff Protocol is architecturally similar to the context-compaction strategy in ccmemory (seed), but APM's handoff is triggered by context-window pressure rather than by session end; APM formalizes it as a named protocol with a structured transfer document
  • claudeAutoPilot is the closest implementation to taskmaster-ai (seed) in this batch: both use Claude Code slash commands to drive a kanban board-style task workflow; claudeAutoPilot is simpler (no AI-generated task decomposition, manual PRD entry)
  • saflib-workflows TypeScript DSL is architecturally unique in this corpus — the only framework using a compiled state-machine approach (makeWorkflowMachine) for workflow composition; no seed or other batch framework uses a formal state-machine model
  • DocBrain docbrain_ask + docbrain_incident MCP tools are directly consumable by any framework that does RAG-style knowledge lookup; DocBrain could serve as the knowledge backend for any MCP-compatible framework in this corpus
  • Patchwork OS CreatePR + CommitChanges + ScanSemgrep steps are reusable under Apache-2.0; frameworks that need CI integration could embed these steps without AGPL obligations
  • tessl tile registry (tessl.io) is the only package-registry-based methodology distribution in the corpus; comparable in structure to npm packages but for agent behavior bundles — a potential distribution model for other frameworks