HiveSpec is a spec-driven delivery lifecycle for AI agent swarms. Each agent follows the same disciplined phases — from claiming an issue to shipping a verified PR.
claude plugin add EntityProcess/hivespec Every agent follows the same disciplined path from issue to merged PR
Pick the next issue from the backlog, explore the codebase, and write a design spec before touching code.
Break the spec into tasks. Test-driven implementation. Red, green, refactor, commit.
Run real commands. Check blast radius. Evidence over assertions.
Classify risk. Auto-merge safe changes. Flag breaking changes for review.
Agents jump straight to code. Skip the design. Break three files to fix one.
They modify a shared interface without checking who depends on it.
They claim everything works. No command output. No evidence.
They ship a PR without verifying consumers of modified code still work.
hs-design blocks implementation until a design is written and approved.
hs-explore finds all consumers of shared interfaces before proposing changes.
hs-implement watches tests fail before writing implementation code.
hs-verify requires actual command output, not assertions.
hs-ship classifies risk and requires confirmation for elevated-risk PRs.
Ten agents running HiveSpec concurrently is a swarm that delivers like a team.
SKILL.md files. No lock-in. Any agent that reads them can run HiveSpec.
Install HiveSpec in one command and start shipping with confidence.
claude plugin add EntityProcess/hivespec