feat: add hivespec plugin (phase-based delivery lifecycle for agent swarms)

[christso]

Summary

Add a new agentic-workflows plugin to plugins/agentic-workflows/ that provides a phase-based delivery lifecycle for AI coding agents. Replaces superpowers plugin with a unified, opinionated workflow: claim → explore → design → plan → implement → verify → ship.

Motivation

Current agent workflows rely on 14+ scattered superpowers skills plus local repo skills. Agents need human intervention to remember worktree discipline, e2e verification, and cross-repo coordination. This plugin consolidates everything into 8 phase-based skills that map to how work actually flows.

Skills

Skill	Phase	Replaces
aw-using-agentic-workflows	Skill discovery + enforcement	superpowers:using-superpowers
aw-claim	Claim issue, worktree, branch, draft PR	using-git-worktrees, ship
aw-explore	Codebase research, issue analysis	agentic-eval-research (generic)
aw-design	Brainstorm, write spec	brainstorming
aw-plan	Write implementation plan	writing-plans
aw-implement	TDD, subagent dispatch, debugging	TDD, subagent-driven-dev, executing-plans, systematic-debugging
aw-verify	E2E red/green, code review, verification	verification-before-completion, requesting/receiving-code-review
aw-ship	Finish branch, merge/PR, cross-repo	finishing-a-development-branch, agentevals-develop

Design Decisions

• Phase-based, not concern-based — skills map to delivery phases, each embeds relevant discipline (TDD inside implement, verification inside verify)
• Convention over configuration — sensible defaults (worktrees at ../<repo>.worktrees/, conventional commits) but repo CLAUDE.md/AGENTS.md overrides all defaults
• Client-agnostic — aw- prefix prevents collisions in non-plugin environments
• Plans on branches — .agents/plans/ on the worktree branch, not main repo tree
• No hooks initially — relies on skill frontmatter injection; SessionStart hook is a future option if evals show frontmatter isn't reliable
• No private context — all content is public, no proprietary references

Companion Plugin

Pairs with agentic-engineering (already in plugins/):

• agentic-engineering = "here are the workflow patterns" (design-time)
• agentic-workflows = "here's the one you use" (execution-time)

Acceptance Signals

• All 8 skills implemented with SKILL.md files
• Plugin follows existing plugins/ structure conventions
• No references to private/proprietary context
• Skills work both as plugin (agentic-workflows:aw-claim) and standalone (.claude/skills/aw-claim/)
• Spec at agentevals-research/docs/superpowers/specs/2026-03-28-agentic-workflows-plugin-design.md

Non-Goals

• Research orchestration (future plugin)
• Writing skills / plugin-dev (stays in plugin-dev)
• Agent architecture design (stays in agentic-engineering)
• Client-specific adaptations (handled by allagents sync)

Design Spec

Full spec: agentevals-research/docs/superpowers/specs/2026-03-28-agentic-workflows-plugin-design.md

Evals

Eval suite at evals/agentic-workflows/ following the pattern of evals/agentic-engineering/:

• Workspace templates with planted scenarios (shared utils with multiple consumers, partial prior art, multiple execution modes, strict lint config)
• Tests that agents invoke correct skills and follow the lifecycle
• Coverage: aw-claim, aw-explore, aw-design, aw-verify, aw-ship

[christso]

Plugin implemented at EntityProcess/hivespec. Tracking: #832