Version History

This document records all version updates of the OpenCode Agent Skills plugin. Through the version history, you can understand the evolution path of features, fixed issues, and architecture improvements.

Current Version

The latest stable version is v0.6.4 (2026-01-20)

Version Timeline

Version	Release Date	Type	Main Content
0.6.4	2026-01-20	Fix	YAML parsing, Claude v2 support
0.6.3	2025-12-16	Improvement	Optimize skill recommendation prompts
0.6.2	2025-12-15	Fix	Separate skill name from directory name
0.6.1	2025-12-13	Improvement	Avoid recommending already loaded skills
0.6.0	2025-12-12	New Feature	Semantic matching, embedding precomputation
0.5.0	2025-12-11	Improvement	Fuzzy matching suggestions, refactoring
0.4.1	2025-12-08	Improvement	Simplify installation method
0.4.0	2025-12-05	Improvement	Recursive script search
0.3.3	2025-12-02	Fix	Symlink handling
0.3.2	2025-11-30	Fix	Preserve agent mode
0.3.1	2025-11-28	Fix	Model switching issue
0.3.0	2025-11-27	New Feature	File listing functionality
0.2.0	2025-11-26	New Feature	Superpowers mode
0.1.0	2025-11-26	Initial	4 tools, multi-location discovery

Detailed Change Log

v0.6.4 (2026-01-20)

Fixes:

• Fixed YAML frontmatter parsing for skill multi-line descriptions (supporting | and > block scalar syntax) by replacing custom parser with yaml library
• Added support for Claude plugin v2 format, installed_plugins.json now uses plugin installation array instead of single object

Improvements:

• Claude Code plugin cache discovery now supports new nested directory structure (cache/<marketplace>/<plugin>/<version>/skills/)

v0.6.3 (2025-12-16)

Improvements:

• Optimized skill evaluation prompts to prevent models from sending "no skill needed" messages to users (users don't see hidden evaluation prompts)

v0.6.2 (2025-12-15)

Fixes:

• Skill validation now allows directory names to differ from the name in SKILL.md frontmatter. The name in frontmatter is the canonical identifier, while directory names are only for organization. This aligns with the Anthropic Agent Skills specification.

v0.6.1 (2025-12-13)

Improvements:

• Dynamic skill recommendations now track loaded skills per session to avoid recommending already loaded skills, reducing redundant prompts and context usage

v0.6.0 (2025-12-12)

Additions:

• Semantic Skill Matching: After initial skill list injection, subsequent messages use local embeddings to match with skill descriptions
• Added @huggingface/transformers dependency for local embedding generation (quantized all-MiniLM-L6-v2)
• When messages match available skills, inject 3-step evaluation prompt (EVALUATE → DECIDE → ACTIVATE) to encourage skill loading (inspired by @spences10's blog post)
• Cache embeddings on disk for low-latency matching (~/.cache/opencode-agent-skills/)
• Clean up sessions on session.deleted event

v0.5.0 (2025-12-11)

Additions:

• Added "Did you mean..." fuzzy matching suggestions in all tools (use_skill, read_skill_file, run_skill_script, get_available_skills)

Improvements:

• Breaking Change: Renamed find_skills tool to get_available_skills for clearer intent
• Internal: Reorganized codebase into separate modules (claude.ts, skills.ts, tools.ts, utils.ts, superpowers.ts) for improved maintainability
• Internal: Improved code quality by removing AI-generated comments and unnecessary code

v0.4.1 (2025-12-08)

Improvements:

• Installation method now uses npm package via OpenCode config instead of git clone + symlink

Removals:

• Removed INSTALL.md (no longer needed after simplification)

v0.4.0 (2025-12-05)

Improvements:

• Script discovery now recursively searches the entire skill directory (max depth 10) instead of only root directory and scripts/ subdirectory
• Scripts are now identified by relative paths (e.g., tools/build.sh) instead of base names
• Renamed skill_name parameter to skill in read_skill_file, run_skill_script, and use_skill tools
• Renamed script_name parameter to script in run_skill_script tool

v0.3.3 (2025-12-02)

Fixes:

• Fixed file and directory detection by using fs.stat to correctly handle symbolic links

v0.3.2 (2025-11-30)

Fixes:

• Preserved agent mode when injecting synthetic messages at session start

v0.3.1 (2025-11-28)

Fixes:

• Fixed unintended model switching when using skill tools by explicitly passing current model in noReply operation (temporary workaround for opencode issue #4475)

v0.3.0 (2025-11-27)

Additions:

• Added file list in use_skill output

v0.2.0 (2025-11-26)

Additions:

• Added Superpowers mode support
• Added release proof

v0.1.0 (2025-11-26)

Additions:

• Added use_skill tool to load skill content into context
• Added read_skill_file tool to read supporting files in skill directory
• Added run_skill_script tool to execute scripts from skill directory
• Added find_skills tool to search and list available skills
• Added multi-location skill discovery (project-level, user-level, and Claude-compatible locations)
• Added frontmatter validation compliant with Anthropic Agent Skills Spec v1.0
• Added automatic skill list injection at session start and after context compression

New Contributors:

• Josh Thomas [email protected] (Maintainer)

Feature Evolution Overview

Feature	Introduced Version	Evolution Path
4 Basic Tools	v0.1.0	v0.5.0 added fuzzy matching
Multi-location Skill Discovery	v0.1.0	v0.4.1 simplified installation, v0.6.4 added Claude v2 support
Auto Context Injection	v0.1.0	v0.3.0 added file list, v0.6.1 avoided duplicate recommendations
Superpowers Mode	v0.2.0	Continuously stable
Recursive Script Search	v0.4.0	v0.3.3 fixed symlinks
Semantic Matching Recommendation	v0.6.0	v0.6.1 avoided duplicates, v0.6.3 optimized prompts
Fuzzy Matching Suggestions	v0.5.0	Continuously stable

Breaking Changes

v0.6.0: Semantic Matching Feature

Introduced local embedding-based semantic matching capability, allowing AI to automatically recommend relevant skills based on user message content without users needing to memorize skill names.

• Technical Details: Using HuggingFace's all-MiniLM-L6-v2 model (q8 quantized)
• Caching Mechanism: Embedding results cached to ~/.cache/opencode-agent-skills/, improving subsequent matching speed
• Matching Process: User message → embedding → calculate cosine similarity with skill descriptions → Top 5 recommendations (threshold 0.35)

v0.5.0: Refactoring and Tool Renaming

Code architecture refactored to modular design with clearer tool naming:

• find_skills → get_available_skills
• skill_name → skill
• script_name → script

v0.4.0: Script Discovery Mechanism Upgrade

Script discovery upgraded from "root directory + scripts/ only" to "recursive search of entire skill directory" (max depth 10), supporting more flexible script organization.

v0.2.0: Superpowers Integration

Added support for Superpowers workflow mode, requiring installation of using-superpowers skill and setting environment variable OPENCODE_AGENT_SKILLS_SUPERPOWERS_MODE=true.

---

Appendix: Source Code Reference

Click to expand source code locations

Updated: 2026-01-24

Feature	File Path	Line
Current Version	package.json	3
Version History	CHANGELOG.md	19-152

Key Version Information:

• v0.6.4: Current version (2026-01-20)
• v0.6.0: Semantic matching introduced (2025-12-12)
• v0.5.0: Refactoring version (2025-12-11)
• v0.1.0: Initial version (2025-11-26)