Phase D Batch 32 — Notes

Agent: batch-32
Date: 2026-05-26
Frameworks assigned: 11
Frameworks completed: 11 (9 full analysis + 1 Tier-C stub + 1 dual-repo combined)

Framework Roster

Intra-Batch Patterns

1. Spectrum of Abstraction Layers

This batch spans the full stack from low-level primitives to full platform:

• Language layer (compose-lang): compiles a 3-keyword DSL to LLM calls — the most primitive
• Coordination layer (asynkor): MCP server providing file leases and agent briefings
• Skill/behavioral layer (flokay, opsx-feature-dev): CLAUDE.md + skills, no infrastructure
• CLI orchestration layer (shep-cli): full SDLC lifecycle via YAML specs + git worktrees
• Desktop workspace (codeg): Tauri app managing multiple concurrent agents across sessions
• Control plane (openase): Go backend with Postgres, Harnesses, ticket lifecycle, adapter pattern
• BaaS platform (insforge): MCP server + web dashboard + vendored skills; developer experience focus

2. Execution Mode Distribution

3. Isolation Mechanisms

4. Multi-Model Usage

Only two frameworks in this batch use non-Claude models:

• codeg — supports Claude Code, Codex CLI, OpenCode, Gemini CLI, OpenClaw, Cline (agent-agnostic platform)
• cc-command-center — Qwen via Dashscope API as secondary advisory LLM (only case of Claude calling a non-Claude model from within a session)

5. DSL / Language-Level Innovation

compose-lang stands alone in this batch (and in the full corpus as of this batch) as the only framework that defines a new programming language for agent behavior. Rather than prompts or YAML config, it compiles .clang source files through a full compiler pipeline (Lexer → Parser → Semantic Analyzer → IR → LLM backend). This is architecturally closer to Terraform/HCL than to CLAUDE.md.

6. Platform Governance vs. Repository Governance

A recurring tension in this batch: where do behavioral rules live?

• Repository governance: flokay, opsx-feature-dev, compose-lang — rules in SKILL.md files and CLAUDE.md committed to repos
• Platform governance: openase, asynkor — rules in Postgres/cloud dashboard, injected at runtime; repos are behavioral-free
• Hybrid: shep-cli, cueapi — YAML spec files in repo, but platform manages state transitions

OpenASE's "Harness as versioned managed asset" (not committed to repo) is the most explicit statement of platform governance in the batch.

7. Observability Gap

None of the prescriptive frameworks in this batch provide observability into what their agents actually do. The cc-command-center combined entry is the only pure-observability tool in the batch — and appears to be the only one in the entire corpus. This suggests a systematic gap: frameworks tell agents how to behave but provide no retroactive visibility into behavior.

Resolution Notes for Reddit References

auto-claude

Searched GitHub exhaustively using: "auto-claude", "auto claude spec-kitty", "auto claude vibe-kanban", author searches. No public repository found. The Reddit hint "Auto Claude spec-kitty vibe kanban" most likely describes a user's personal workflow that combines spec-kitty (analyzed in batch ?) and vibe-kanban (analyzed in batch ?) rather than a standalone framework. Filed as Tier-C.

cc-command-center

Reddit hint "Claude Code command center Qwen advisor" matched two independent repositories:

• amahpour/claude-code-command-center (5 stars): FastAPI web dashboard indexing Claude Code sessions via hooks
• AshlynYang-dev/claude-code-qwen-skill (4 stars): Shell script skill routing to Alibaba Qwen models

Neither repo references the other. Analyzed together as a combined entry because together they describe the "command center + Qwen advisor" concept from the Reddit hint. Documented as dual-repo in METRICS.yaml.

Anomalies

• flokay display name mismatch: The pacaplan/flokay repo's actual display name (from README) is "Codagent Agent Skills," not "Flokay." The slug retains flokay to match the GitHub repo name and assignment table.
• insforge stars: 10,636 stars is an outlier in this batch. This is a developer-tools BaaS with active marketing and a managed service offering — different adoption context from the other frameworks.
• compose-lang completeness: The repo is real and functional (Lexer/Parser source code confirmed) but clearly in early-stage development (3 stars, no npm publish, .vsix not listed on VSCode marketplace). Analyzed on technical merit regardless of adoption.
• cueapi auto-merge: The CLAUDE.md explicitly permits Claude Code to open PRs and merge them once CI passes with no human reviewer. This is the most explicit auto-merge authorization found in this batch and one of the most explicit in the corpus.
• openase Go 1.26+: Requires Go 1.26+, which was not released at time of corpus collection (2025). Likely means the repo was written targeting a pre-release or the version field is aspirational.