Skip to content
/
Phase B Batch 2

Batch 02 — Spec-driven (Spec Kit + Kiro derivatives)

Batch 02 — Spec-driven (Spec Kit + Kiro derivatives)

Roster (10)

slug stars distribution cli_binary local_ui orchestration multi_model tier
cc-sdd 3394 npm-package cc-sdd none task-decomposition-tree no A
claude-code-spec-workflow 3744 npm-package claude-code-spec-workflow web-dashboard (8247) + Tauri sequential no A
spec-workflow-mcp 4199 mcp-server spec-workflow-mcp web-dashboard (5000) + VSCode ext sequential no A
spec-kit-aider 0 cli-tool (py) specify none sequential no A
spec-kit-command-cursor 182 command-pack none none parallel-fan-out no A
spec-kit-command-cursor-foxgod 3 command-pack none none sequential no A
kiro-jasonkneen 665 claude-plugin none none sequential no A
spec-driven-kiro-project 3 standalone-repo none none none no C
kiro-spec-agent-gist unknown methodology-doc none none sequential no C
maxim-saplin-kiro-sdd unknown methodology-doc none none sequential no C

Intra-batch patterns

All 10 frameworks in this batch share the same three-phase spec workflow (requirements → design → tasks) with EARS notation, and all use .kiro/ or equivalent directories for spec storage — a clear genetic lineage from Amazon's Kiro IDE. The batch splits cleanly into three tiers by architectural ambition: (1) full harnesses with subagent orchestration and autonomous implementation (cc-sdd, spec-kit-command-cursor), (2) portal/dashboard frameworks that wrap the workflow in a real-time UI (claude-code-spec-workflow, spec-workflow-mcp), and (3) minimal redistributions of the core spec-phase pattern with no tooling beyond the prompt files themselves (kiro-jasonkneen, the gists, the demo project). The dominant execution model is interactive-loop with three human approval gates (one per spec phase), inherited from Kiro's original design. No framework in the batch uses multi-model routing or worktrees by default. The spec-workflow-mcpclaude-code-spec-workflow successor relationship is the only explicit evolution lineage in the batch.

Most interesting find

spec-kit-command-cursor (madebyaris): The file-conflict detection mechanism — where sdd.touchedFiles declarations on tasks are used to compute safe parallel batches before dispatching background implementer subagents — is the most architecturally novel idea in the entire batch. No other framework in the full corpus tracks file-level parallelism safety. The checkpoint-resume (execution-checkpoint.json) and downstream-propagation (/evolve) features compound this: together they constitute a micro-workflow engine inside a Cursor IDE plugin.

spec-workflow-mcp (Pimzino): The approvals MCP tool that creates dashboard-backed blocking gates — where the agent literally cannot continue until the browser/VSCode sidebar shows an explicit user action — is the strongest human-in-the-loop gate in the corpus. It survives session boundaries and is the only framework where "phase approval" is a separately persisted state rather than just a conversational exchange.

Items written as Tier C

slug reason
spec-driven-kiro-project Demonstration project, not a framework; ships only spec artifacts and app code, no redistributable tooling
kiro-spec-agent-gist Single-file gist (CLAUDE.md); complete spec workflow encoded in one file, not a framework
maxim-saplin-kiro-sdd 5-file gist; dual-format (Cursor + Copilot) command files only, no unique architectural contribution

Cross-references discovered

  • spec-workflow-mcp is the explicit successor to claude-code-spec-workflow (same author: Pimzino); README in predecessor states "development focus has shifted to the MCP version"
  • cc-sdd explicitly acknowledges Kiro IDE as inspiration ("Kiro-inspired") and states "existing Kiro specs remain compatible and portable"
  • kiro-jasonkneen ships the verbatim Kiro IDE system prompt from .kiro/system/complete-instructions.md
  • spec-kit-aider is a direct fork of GitHub's spec-kit (same philosophy text copied verbatim from spec-driven.md)
  • spec-kit-command-cursor (foxgod183) appears to derive from spec-kit-command-cursor (madebyaris) without attribution (same command names, directory structure, feat-{N}-{name} pattern)
  • kiro-spec-agent-gist (notdp) and maxim-saplin-kiro-sdd independently implement nearly identical workflows — both are minimal single/multi-file ports of the Kiro spec workflow to non-Kiro agents
  • spec-kit-command-cursor (madebyaris) cites Cursor 3.2 as a key dependency and references the Cursor-specific is_background: true agent feature not available in other frameworks