Skip to content

feat(plugin): add hivespec plugin (phase-based delivery lifecycle for agent swarms)#824

Closed
christso wants to merge 9 commits intomainfrom
feat/823-agentic-workflows-plugin
Closed

feat(plugin): add hivespec plugin (phase-based delivery lifecycle for agent swarms)#824
christso wants to merge 9 commits intomainfrom
feat/823-agentic-workflows-plugin

Conversation

@christso
Copy link
Copy Markdown
Collaborator

@christso christso commented Mar 28, 2026

Summary

  • Adds autopilot-dev plugin with 8 phase-based skills covering the full delivery lifecycle: claim → explore → design → plan → implement → verify → ship
  • Replaces superpowers plugin with a unified, opinionated workflow where each phase embeds relevant discipline (TDD in implement, e2e red/green in verify, blast radius checks in ship)
  • Includes eval suite with 5 eval files and a workspace template containing planted scenarios (shared utility with 3 consumers, partial prior art, multiple execution modes, strict biome lint config)
  • Workspace setup copies skills to .claude/skills/, .agents/skills/, and .pi/skills/ for cross-provider skill discovery

Plugin Structure

Skill Phase Key Discipline
ad-using-autopilot-dev Entry point 1% rule, red flags, phase skip rules
ad-claim Workspace setup Read guidelines first, worktree + draft PR
ad-explore Codebase research Find prior art, all consumers, structured summary
ad-design Spec writing Multiple approaches, section-by-section approval
ad-plan Task breakdown No placeholders, enumerate all consumers
ad-implement TDD execution Red/green, subagent dispatch, systematic debugging
ad-verify E2E verification Red/green protocol, blast radius, all modes tested
ad-ship PR merge Final blast radius, risk classification, cleanup

Eval Results

Eval Tests Claude CLI Pi CLI
ad-explore 3 3/3 PASS (1.000) 3/3 PASS (1.000)
ad-verify 4 4/4 PASS (1.000) 3/4 (0.833)
ad-design 3 3/3 PASS (1.000) 0/3 (0.417)
ad-claim 3 3/3 PASS (1.000) 2/3 (0.833)
ad-ship 4 4/4 PASS (1.000) 1/4 (0.708)
Total 17 17/17 PASS 9/17 pass

Design Decisions

  • Phase-based (not concern-based) — each skill is a delivery phase, not an isolated concern
  • Convention over configuration — sensible defaults, repo CLAUDE.md/AGENTS.md overrides all
  • Plans on branches — .agents/plans/ on worktree branch, not main
  • No hooks initially — relies on skill frontmatter injection
  • No private context — all content is public

Closes #823

🤖 Generated with Claude Code

christso and others added 2 commits March 28, 2026 22:24
…lifecycle

8 skills covering the full delivery lifecycle: claim → explore → design →
plan → implement → verify → ship. Replaces superpowers with a unified,
opinionated workflow. Includes eval suite with workspace template containing
planted scenarios (shared utility with 3 consumers, partial prior art,
multiple execution modes, strict lint config).

Closes #823

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@cloudflare-workers-and-pages
Copy link
Copy Markdown

cloudflare-workers-and-pages bot commented Mar 28, 2026

Deploying agentv with  Cloudflare Pages  Cloudflare Pages

Latest commit: acb0f8c
Status: ✅  Deploy successful!
Preview URL: https://56edda47.agentv.pages.dev
Branch Preview URL: https://feat-823-agentic-workflows-p.agentv.pages.dev

View logs

christso and others added 2 commits March 28, 2026 23:28
…l discovery

Claude CLI discovers skills from .claude/skills/, not .agents/skills/.
Without this, skill-trigger assertions fail for claude-cli targets.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rename plugin from agentic-workflows to autopilot-dev for clarity:
- Plugin dir: plugins/autopilot-dev/
- Skill prefix: ad- (was aw-)
- Eval dir: evals/autopilot-dev/
- Entry point: ad-using-autopilot-dev (was aw-using-agentic-workflows)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@christso christso changed the title feat(plugin): add agentic-workflows plugin (phase-based delivery lifecycle) feat(plugin): add autopilot-dev plugin (phase-based delivery lifecycle) Mar 28, 2026
christso and others added 4 commits March 29, 2026 00:02
- Include SKILL.md as file input for design, ship, verify, and claim evals
  so agents see skill discipline before responding
- Add README.md (with planted typo for scope assessment test), package.json,
  and tsconfig.json to workspace template
- Initialize git repo with feat branch in setup.mjs for ship/claim tests
- Improve prompts: provide verification context, reference specific files,
  avoid "skip the design" framing that agents comply with literally

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Claim tests: provide issue body directly in prompt, remove GitHub dependency
- Ship tests: setup script adds a commit on feat branch so agent sees changes
- Claim reads-full-issue-body: explicit "no need to fetch from GitHub"
- Claim assesses-scope: ask for scope assessment explicitly

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Ship tests were failing because agents checked git diff and found the
simulated changes didn't match prompt descriptions. Rewrite to test
the skill's decision-making discipline directly: provide change
descriptions in prompts and ask "should I auto-merge or get review?"
instead of simulating actual git operations.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ap- is the natural abbreviation of autopilot. ad- reads as "advertisement."

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
HiveSpec reflects the multi-agent swarm coordination model where agents
claim issues from a shared task board (GitHub, Linear, etc.) and execute
the delivery lifecycle independently. The plugin is the individual agent's
flight plan; the hive emerges from many agents running it concurrently.

- Plugin: plugins/hivespec/
- Prefix: hs- (hs-claim, hs-explore, hs-design, etc.)
- Evals: evals/hivespec/

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@christso christso changed the title feat(plugin): add autopilot-dev plugin (phase-based delivery lifecycle) feat(plugin): add hivespec plugin (phase-based delivery lifecycle for agent swarms) Mar 29, 2026
@christso
Copy link
Copy Markdown
Collaborator Author

Closing — plugin source moved to standalone repo EntityProcess/hivespec. Eval results at EntityProcess/hivespec-evals.

The evals in evals/hivespec/ on this branch can be submitted as a separate PR that references the external plugin, or kept as-is once the plugin is installable.

Tracking: #832

@christso christso closed this Mar 29, 2026
@christso christso deleted the feat/823-agentic-workflows-plugin branch March 29, 2026 06:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

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

1 participant