| Specsafe |
Issue/spec safety layer |
Issue-driven enhancements |
Not primary |
GitHub issues |
Issue/spec references |
Integration-specific |
Not primary |
Spec safety implied |
Issue review |
IDE/CLI integration oriented |
Traceability and integration roadmap |
IDE-first lock-in |
| BMAD |
Agile AI development method |
Structured lifecycle workflows |
Analyst, PM, Architect, Dev, QA, UX, etc. |
Project/module files |
Workflow artifacts |
Agents/modules |
Not core |
Workflow reviews |
Multiple approvals |
Works through AI IDE installs |
Role separation and lifecycle stages |
Letting BMAD own the ontology |
| superpowers |
Skills + coding methodology |
Brainstorm → worktree → plan → TDD → review → finish |
Skill-driven subagents/reviewers |
Skill docs, plans, branches |
Design/plan docs |
Markdown skills |
Git worktrees |
Tests + reviews |
Design approval, reviews, merge choice |
Multiple agent hosts |
Mandatory skill discovery and review loops |
Single host/plugin assumptions |
| GSD v1 (Get Shit Done) |
Context-engineered command loop |
New project → discuss → plan → execute → verify |
Fresh subagents |
Project context and phase docs |
Requirements/roadmaps/plans |
Commands/prompts |
Fresh contexts, not primarily FS |
Verification loops |
Phase gates |
Multi-tool but Claude Code-centered |
Context-rot mitigation and phase sizing |
Permission model assumptions |
GSD 2 (gsd-pi) |
TypeScript agent CLI built on Pi SDK |
DB state machine: plan → execute → verify → reassess → validate-milestone |
Capability-routed model selection (ADR-004); subagent telemetry with random names |
SQLite gsd.db authoritative; .gsd/*.md projections; memories table for decisions/patterns |
M###-CONTEXT.md / M###-ROADMAP.md / STATE.md / DECISIONS.md / KNOWLEDGE.md |
Extensions framework (@gsd-extensions/<name>); per-unit-type skill manifest resolver |
Worktree-mode (milestone/<MID> branch) or branchless (ADR-001) |
Three-verdict validate-milestone (pass / needs-attention / needs-remediation); auto-fix retries; drift detection (ADR-017) |
/gsd verdict manual override; step-mode wizard between units; escape-to-pause |
20+ LLM providers; provider-agnostic by design; headless mode with JSON query |
Three-verdict validation + manual override, drift registry under uh validate, uh status --json, canonical VISION.md, pre-inlined dispatch contract |
SQLite-authoritative state, branchless worktree, external state dir, bundled VS Code / web visualizer in core, $GSD token coupling |
| matt-pocock/skills |
Small engineering skills |
Clarify → shared language → implement/debug/review |
Skill-specific |
CONTEXT.md and docs |
Docs/tickets |
Small reusable skills |
Not primary |
Engineering feedback loops |
Clarification checkpoints |
Agent-agnostic |
Small practical skills and shared language |
Over-process or verbose ceremonies |
| oh-my-openagent |
Opinionated agent harness |
Packaged workflows + hooks + routing |
Multiple configured agents |
Config, hooks, sessions |
Harness config/docs |
Agents, hooks, MCP tools |
Runtime-dependent |
Hooks/reviews |
Workflow discipline |
OpenCode-first, multi-model |
Batteries-included defaults and hooks |
Becoming a plugin distro |
| OpenSpec |
Change artifact folder |
Propose → apply → archive |
Assistant commands |
openspec/changes/... |
Proposal, specs, design, tasks |
Commands/skills |
Not primary |
Verify command and tasks |
Propose/archive approvals |
25+ assistants |
Artifact-guided lifecycle |
Spec-only worldview |
| Pi / oh-my-pi |
Terminal agent harness |
Interactive/JSON/RPC/SDK; sessions and branches |
Subagents |
Session trees, config |
Prompt/session artifacts |
Skills/extensions/packages |
Worktrees, FUSE overlay, ProjFS in oh-my-pi |
Review command, LSP/tests |
User steering and review |
Multi-provider, embeddable |
Adapter target and possible engine |
Premature engine commitment |
| AgentFS |
Agent filesystem |
Init/mount/exec/run/sync |
Not role-based |
DB-backed filesystem |
Filesystem state |
MCP FS tools |
Copy-on-write, overlay, sandbox |
Inspect deltas/sync |
Promotion by inspection |
CLI/MCP |
Sandbox backend design |
Equating with git branches |
| oh-my-claudecode (OMC) |
Multi-agent orchestration over Claude Code |
Staged team pipeline (plan → prd → exec → verify → fix) |
19 named agents across build/review/domain lanes with model-tier routing |
.claude/ + skill/agent files |
Markdown agents + skills |
36 skills + /-commands |
Native Claude Code teams; tmux for cross-CLI workers |
Verifier agent + critic + code-reviewer + security-reviewer |
Deep-interview gate; team-verify stage |
Claude Code plugin + npm CLI; spawns gemini/codex workers in tmux |
Staged pipeline as workflow profile family, team N:role mission shape, model-tier routing as a per-role hint |
tmux as canonical worker host; Claude-Code-plugin lock-in; swarm/ralph as core vocabulary |
| oh-my-codex (OMX) |
Workflow layer for OpenAI Codex CLI (Rust) |
$deep-interview → $ralplan → $ralph / $team with staged pipeline |
Specialist roles per lane (core dev, QA, domain, product strategy) |
.omx/ (plans, logs, memory, mode, ultragoal, integration-report) |
AGENTS.md, DESIGN.md, .omx/ultragoal artifacts |
33 prompts + 36 $-skills |
Per-worker git worktrees by default (.omx/team/<name>/worktrees/worker-N); detached branches; leader-merge integration |
$ultraqa adversarial profile (hostile scenarios, prompt injection, interrupt/resume, cleanup evidence) |
$ralplan consensus + $ultraqa verdict |
npm shim + Codex plugin layout; Hermes MCP Bridge for external coordination |
Adversarial QA profile, per-worker worktree auto-isolation, Hermes MCP Bridge contract, design.md companion artifact |
Rust as the host language; --yolo/--madmax slang flags; single-multiplexer worker contract |
