Implement an interactive CLI tool to browse and install workflow
packs. Consolidate decentralized manifests into a single root
registry for easier maintenance.
Add registry.json manifests to agent packs and commands to support
structured discovery. Include a CLI for project initialization, a
registry generator script, and a plugin version locker. Add a new
learn command for session documentation.
Update command-creator documentation with critical placeholder rules,
XML block structures, and direct address terminology. Add the /learn
command for documenting session findings and workarounds.
Introduce a specialized agent pack for repository documentation.
Includes a unified `/init` command with routing for AI navigation or
user assistance, alongside SOP skills for mapping codebases and
onboarding users.
- Add repo-navigator agent and pack documentation
- Implement `/init` command for automated AGENTS.md generation
- Add agent-navigation-sop and user-onboarding-sop skills
- Update root README and AGENTS.md guides
- Enhance skill-creator with bundled reference documents
Introduce the Principle of Least Privilege to agent templates by
enforcing scoped skill permissions. This update adds five specialized
Convex skills to the VRTCT stack and updates documentation to reflect
the new architecture and folder structure.
Apply structured tags (<overview>, <rules>, <context>) across the
Vite-Convex stack. Update the database expert agent permissions and
modernize guidance for Convex Auth Beta and Tailwind v4.
Implement comprehensive skills for Convex auth, core, deployment, and
runtime. Add detailed component references for Agent, RAG, Workflow,
and Workpool patterns. Update the database expert agent with new
schema and security guidelines.
Rename repo-navigator-creator to repo-navigator and add support for
dual workflows (/init and /howto). Integrate explore subagent logic
for handling complex repositories and update the role definition and
quality checklist accordingly.
Add standardized blocks for core mission, safety standards, tool
discipline, LSP management, and engineering workflow to all agent
definitions. Include resumption protocols where applicable.
Introduce a fast/smart agent split to replace the legacy general
subagent. Rename improve commands to remove colons. Update the audit
script to support multi-line frontmatter and stricter example
validation.
- Implement internal maintenance suite: repo-maintainer agent and automated audit/sync scripts.
- Add "Opencode Configurator" suite: agent-architect, command-creator, and skill-creator.
- Refactor core commands (/improve, /refactor, /rmslop) to RFC 2119 + XML standards.
- Standardize all agent metadata and structural tagging across the repository.
- Introduce utility commands: /npm for package management and /create-pack for bundling.
- Overhaul README.md and documentation to reflect new workflow patterns.
Introduce a new agent and comprehensive skill set for professional React
component engineering. This includes specialized commands for component
review and creation based on a formal specification.
Key additions:
- New component-engineer agent and associated commands
- Deep knowledge-base skill covering accessibility and composition
- Integration of the specification into the VRTCT stack agents
Reorder permission blocks in agents to ensure the wildcard denial is
listed first, which is required for correct skill evaluation.
Update the permissions-update command documentation to clarify this
requirement.
Remove the gpt-oauth-prompt-enhancer plugin and disable the
gpt-oss-120b-medium model in the proxy configuration.
The Ralph Wiggum Loop is an autonomous orchestrator using stateful
subagent sessions and dynamic checklists for task verification.
Documentation now includes warnings that the skill permission system is
non-functional in OpenCode 1.1.2.
- Add parallel-PRD agent collection for multi-perspective planning
- Migrate agents and skills from legacy tools to permission schema
- Add permissions-update command to automate configuration modernization
- Update documentation to reflect new agent collections and schema
changes
Add instructions for enabling skills via YAML frontmatter. Clarify that
skills are disabled by default to prevent context pollution and specify
default skill sets for the Plugin Creator, Configurator, and Security
Reviewer agents.
Apply research-backed formatting across all agents, commands, and
plugins to improve LLM precision. This update also replaces the
commands-md-updater agent with a new /sync-commands workflow.
- Add Security Reviewer with 10 new security skills.
- Split Vite-React expert into orchestrator and brain roles.
- Add /init and /refactor-rfc-xml commands.
- Update model references to GPT-5.2 and remove deprecated agents.
Apply a default deny policy for skills across all agents. Refine
frontmatter descriptions for clarity and restructure the
commands-md-updater agent using semantic tags. Update the TypeScript
expert to include test writing capabilities.
Remove CLAUDE.md and mark Commands2Skills as partially deprecated to
reflect OpenCode's native Skills support. Add VRTCT-orchestrator agent
and an init command for AGENTS.md. Improve template documentation.
Convert markdown files to use XML-like tags for structure and adopt RFC
2119 keywords. Delete redundant generic agent definitions and update
coding guidelines to enforce stricter modularity, file size limits, and
barrel exports.
OpenCode Agent