v1.4.0

What's new?

• Removed plugin components that aren't supported by OpenCode's plugin runtime — deleted the dependency-detective, impact-investigator, and source-scout agents, all seven constellation-* skills, and the unused CLAUDE.md/GEMINI.md files
• Streamlined src/plugins/constellation.ts and rewrote README.md and AGENTS.md to reflect the leaner plugin surface
• Added .npmignore to keep development files out of the published package

v1.3.0

What's new?

• Updated plugin description to "Upgrade OpenCode from text search to code understanding." for clearer messaging
• Enhanced constellation-troubleshooting skill with guidance on inspecting error.guidance[0] from MCP responses to diagnose AUTH_ERROR failures and verify CONSTELLATION_ACCESS_KEY state

v1.2.2

What's new?

• Replaced the seven /constellation-* slash commands with on-demand skills loaded via OpenCode's skill tool — agents now invoke constellation-status, constellation-diagnose, constellation-impact, constellation-deps, constellation-unused, constellation-architecture, and constellation-troubleshooting automatically based on user intent rather than explicit / invocation
• Folded the standalone constellation-troubleshoot command's full error-code reference into the existing constellation-troubleshooting skill (single source of truth)
• Renamed prompts/ → agents/ to clarify the folder's purpose and switched opencode.json to use {file:...} syntax for agent prompt references
• Updated README.md and AGENTS.md to document the skill-based UX (natural-language triggers) instead of slash commands

v1.2.1

What's new?

• Resolve CONSTELLATION_ACCESS_KEY at runtime via process.env so the MCP plugin picks up the key from the active shell instead of the value captured at plugin load

v1.2.0

What's new?

• Replaced erroneous reference to MCP binary name with npx package command

v1.1.0

What's new?

• Slash commands are now namespaced with a constellation- prefix: /constellation-status, /constellation-diagnose, /constellation-impact, /constellation-deps, /constellation-unused, /constellation-architecture, /constellation-troubleshoot
• Replaced the noisy tool.execute.before console log with a tool.execute.after hook that appends an LLM-visible code_intel reminder after grep / glob — the user's terminal stays quiet
• Clarified code_intel scope and usage guidance in the agent instructions
• Updated the Haiku model reference to claude-haiku-4-5-20251001
• Abstracted AI-agent instructions to AGENTS.md; harness-specific files (CLAUDE.md, GEMINI.md) are now symlinks to it

v1.0.1

What's new?

• Constellation plugin for OpenCode release!
• Customized packages details and homepage URL
• Testing automated deployment and trusted publishing to NPM from GitHub Actions