skills

CocoSearch Skills

Reusable AI coding assistant skills for CocoSearch — semantic and symbol-aware search, dependency graphs, cross-repo search. Fewer tokens, less hallucination.

Available Skills

Skill	Description	Interactive?
cocosearch-quickstart	First-time setup: infrastructure check, indexing, and verification	Yes
cocosearch-onboarding	Guided tour of a new codebase: architecture, layers, patterns	Yes
cocosearch-explore	Codebase exploration with two modes: autonomous (subagent/plan mode) and interactive ("how does X work?")	Both
cocosearch-debugging	Root cause analysis: symptom parsing, call tracing, fix suggestions	Yes
cocosearch-deps	Dependency graph exploration: impact analysis, connection tracing, hub identification	Yes
cocosearch-refactoring	Safe refactoring: full impact analysis, dependency mapping, step-by-step execution	Yes
cocosearch-new-feature	Add new functionality: find patterns, match conventions, integrate	Yes
cocosearch-add-language	Add language support: handlers, symbol extraction, context expansion	Yes
cocosearch-add-grammar	Add grammar handler: domain-specific formats within a base language	Yes
cocosearch-add-extractor	Add dependency extractor: enables deps tree, deps impact, and dependency-enriched search	Yes
cocosearch-review-pr	Review GitHub PRs / GitLab MRs with blast radius and dependency analysis; optionally push findings back as inline comments	Yes
cocosearch-commit	Smart commit messages: analyzes diffs with semantic search and dependency impact	Yes

Installation

Automated (Recommended)

cocosearch init

The init command interactively offers to install skills to either project-local (.opencode/skills/) or global (~/.config/opencode/skills/) directories. It also handles MCP registration and CLAUDE.md/AGENTS.md setup.

Claude Code Plugin

claude plugin marketplace add VioletCranberry/coco-search
claude plugin install cocosearch@cocosearch

This automatically configures the MCP server and all 12 skills. No symlinks or manual setup needed.

Claude Code (project-local)

Symlink skills from a cloned CocoSearch repo into your project:

mkdir -p .claude/skills
for skill in cocosearch-onboarding cocosearch-refactoring cocosearch-debugging cocosearch-deps cocosearch-quickstart cocosearch-explore cocosearch-new-feature cocosearch-add-language cocosearch-add-grammar cocosearch-add-extractor cocosearch-review-pr cocosearch-commit; do
    ln -sfn "../../skills/$skill" ".claude/skills/$skill"
done

Claude Code (global)

Copy skills to your global Claude config:

for skill in cocosearch-onboarding cocosearch-refactoring cocosearch-debugging cocosearch-deps cocosearch-quickstart cocosearch-explore cocosearch-new-feature cocosearch-add-language cocosearch-add-grammar cocosearch-add-extractor cocosearch-review-pr cocosearch-commit; do
    mkdir -p ~/.claude/skills/$skill
    cp skills/$skill/SKILL.md ~/.claude/skills/$skill/SKILL.md
done

OpenCode

OpenCode discovers skills from .opencode/skills/, .claude/skills/, and .agents/skills/ in your project root, and from ~/.config/opencode/skills/ globally.

Project-local (symlink from cloned CocoSearch repo):

mkdir -p .opencode/skills
for skill in cocosearch-onboarding cocosearch-refactoring cocosearch-debugging cocosearch-deps cocosearch-quickstart cocosearch-explore cocosearch-new-feature cocosearch-add-language cocosearch-add-grammar cocosearch-add-extractor cocosearch-review-pr cocosearch-commit; do
    ln -sfn "../../skills/$skill" ".opencode/skills/$skill"
done

Global:

for skill in cocosearch-onboarding cocosearch-refactoring cocosearch-debugging cocosearch-deps cocosearch-quickstart cocosearch-explore cocosearch-new-feature cocosearch-add-language cocosearch-add-grammar cocosearch-add-extractor cocosearch-review-pr cocosearch-commit; do
    mkdir -p ~/.config/opencode/skills/$skill
    cp skills/$skill/SKILL.md ~/.config/opencode/skills/$skill/SKILL.md
done

Note: OpenCode has built-in Claude Code compatibility — it also reads skills from .claude/skills/, so existing Claude Code skill installations work without changes.

After installation, restart your AI coding assistant or run the skill activation command for your platform.

Common Search Tips

These tips apply across all CocoSearch skills:

• Always use use_hybrid_search=True -- combines semantic understanding with keyword precision via RRF fusion. Essential for both concept discovery and identifier lookup.
• Always use smart_context=True -- expands results to full function/class boundaries using Tree-sitter. Gives you complete code units, not truncated snippets.
• Use include_deps=True for dependency context -- attaches direct dependencies and dependents to each search result. Useful when you need to understand a file's connections alongside the code.
• Use symbol_name for precision -- when you know the identifier, use symbol_name="<name>*" with glob patterns to catch variants (e.g., User* finds User, UserService, UserProfile).
• Use symbol_type for structural searches -- filter to "function", "class", "method", or "interface" to reduce noise when looking for specific code structures.
• Use language for polyglot codebases -- add language="python" (or any supported language) to scope results when debugging language-specific issues.
• Be specific in queries -- "search query embedding generation" finds more relevant results than "how search works".
• Follow identifiers across hops -- when a function body references another function, search for it by name using symbol_name for precision. Use semantic queries for intent-based discovery.

Dependency Tools

When the project has a dependency index (created with cocosearch index . --deps or cocosearch deps extract .), these MCP tools provide instant dependency analysis:

• get_file_dependencies(file, depth) -- what does a file depend on? depth=1 for direct imports, depth>1 for transitive trees.
• get_file_impact(file, depth) -- what would be affected if a file changes? Returns the transitive impact tree.

These tools are especially powerful for refactoring (impact analysis), debugging (tracing call chains), and exploration (understanding module connections). They complement search by providing structural dependency data that search alone can't guarantee.