Skip to content

Releases: rajkaria/boil-the-ocean

v1.2.0 — one-click plugin install

Choose a tag to compare

@rajkaria rajkaria released this 13 Jul 07:22

One-click plugin install — no shell, no clone.

boil-the-ocean is now its own one-plugin marketplace. From inside Claude Code:

/plugin marketplace add rajkaria/boil-the-ocean
/plugin install boil-the-ocean@boil-the-ocean

That wires the skill, the /ocean* commands, and both hooks — updated in place on re-install, no install.sh needed.

Added

  • .claude-plugin/marketplace.json — the repo is its own one-plugin marketplace.
  • Cross-agent plugin install — the same manifest installs on every agent that speaks the Claude Code plugin-marketplace protocol: Antigravity (agy plugin install <url>), Factory Droid, GitHub Copilot CLI.
  • Marketplace test coverage — 4 new tests (86 total): marketplace.json is valid JSON, its name and plugin entry match the plugin, and its source: "./" resolves to a real plugin.json.

Notes

  • The plugin carries no separate version — its entry inherits from plugin.json, so the release checklist stays VERSION + plugin.json in lockstep.

The shell installer (curl … | bash, install.sh auto, ocean upgrade) is unchanged and still works for headless / non-Claude agents.

Full changelog: https://github.com/rajkaria/boil-the-ocean/blob/main/CHANGELOG.md

v1.1.0 — installation overhaul

Choose a tag to compare

@rajkaria rajkaria released this 13 Jul 06:58

One-liner curl install, install.sh auto agent detection, ocean version --check + ocean upgrade self-updating, doctor PATH/update checks. 82/82 offline tests. Full notes: CHANGELOG.md

v1.0.0 — the first boil

Choose a tag to compare

@rajkaria rajkaria released this 13 Jul 06:23

Hand your AI agent an entire spec and walk away.

Highlights

  • Autonomous multi-sprint execution protocol (SKILL.md): plan once, one commit per sprint, tests green before done, every decision journaled, checkpoint before every exit
  • Scheduler (ocean-daemon) that relaunches fresh workers past context/usage limits: reset-timestamp backoff, stall detection, failure caps, iteration budget, independent verify gate, launchd/cron/systemd support
  • Multi-agent: Claude Code, OpenAI Codex CLI, Gemini CLI, and a custom adapter for any headless agent — unified safe/standard/yolo permission model
  • 54 offline tests (mock agent, zero tokens), CI on macOS + Ubuntu
  • Full docs: architecture, configuration recipes, state schema, per-agent install guides, troubleshooting

Install

git clone https://github.com/rajkaria/boil-the-ocean.git && cd boil-the-ocean
./install.sh        # claude | codex | gemini | bin | all