feat: migrate commands to skills with hot-reload support

[ojowwalker77]

Summary

• Migrate all 10 Matrix slash commands to the new skills format with hot-reload support
• Add proper frontmatter (user-invocable, context, agent, allowed-tools)
• Use haiku for simple operations, sonnet for complex ones
• Use context: fork for review and deep-research (isolated execution)
• Add progressive disclosure with references/ for complex skills

Skills Migration

Skill	Agent	Context	Notes
doctor	haiku	main	Single MCP call
reindex	haiku	main	Single MCP call
list	haiku	main	Data display
export	haiku	main	Data formatting
warn	haiku	main	CRUD operations
skill-candidates	haiku	main	List filtering
repomix	sonnet	main	Two-phase flow
create-skill	sonnet	main	Template generation
deep-research	sonnet	fork	Multi-source synthesis
review	sonnet	fork	5-phase analysis

Benefits

• Hot-reload: Edit skills without restarting Claude Code
• Cost savings: ~40-50% reduction from haiku for simple operations
• Isolation: review/deep-research get fresh context via fork
• Tool safety: Explicit allowed-tools prevents unintended access
• Progressive disclosure: Keep SKILL.md lean, load references as needed

Test plan

• Verify skills appear in /skills menu
• Test trigger phrases load correct skill
• Confirm haiku skills run on haiku model
• Confirm fork context works for review/deep-research
• Test hot-reload by modifying a SKILL.md

[greptile-apps]

Greptile Summary

• Migrates all 10 Matrix slash commands from the legacy commands/ directory to the new skills-based architecture in skills/ directories with hot-reload support
• Introduces intelligent agent selection using haiku for simple operations (doctor, reindex, list, export, warn, skill-candidates) and sonnet for complex operations (repomix, create-skill, deep-research, review)
• Implements context isolation with fork mode for review and deep-research skills to provide fresh execution environments for unbiased analysis

Important Files Changed

Filename	Overview
skills/deep-research/SKILL.md	New skill with sonnet agent and fork context; implements 5-phase research pipeline with multi-source synthesis
skills/review/SKILL.md	New skill with sonnet agent and fork context; comprehensive 5-phase code review with matrix integration
skills/*/SKILL.md (8 files)	Migrated skills with haiku agents and explicit tool permissions for cost optimization and security

Confidence score: 4/5

• This PR is well-structured with clear benefits but requires testing to ensure the new architecture works as expected
• Score reflects good architectural design and proper frontmatter configuration, but uncertainty about runtime behavior of the skills system and potential migration issues with complex skills
• Pay close attention to skills/deep-research/SKILL.md and skills/review/SKILL.md as they use fork context and complex multi-tool workflows

Sequence Diagram

sequenceDiagram
    participant User
    participant Claude as "Claude Code"
    participant HookSystem as "Matrix Hook System"
    participant SkillLoader as "Skill Loader"
    participant MatrixMCP as "Matrix MCP Server"
    participant Database as "Matrix Database"

    User->>Claude: "Submit prompt with skill trigger"
    Claude->>HookSystem: "Trigger UserPromptSubmit hook"
    HookSystem->>HookSystem: "Parse prompt for skill patterns"
    HookSystem->>SkillLoader: "Load skill frontmatter and content"
    SkillLoader->>SkillLoader: "Parse YAML frontmatter (agent, context, tools)"
    HookSystem->>HookSystem: "Check allowed-tools permissions"
    HookSystem->>Claude: "Inject skill context and instructions"
    Claude->>MatrixMCP: "Execute allowed MCP tools"
    MatrixMCP->>Database: "Query/store data as needed"
    Database-->>MatrixMCP: "Return results"
    MatrixMCP-->>Claude: "Tool execution results"
    Claude->>User: "Display skill execution output"

Loading

[greptile-apps]

_{14 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[github-actions]

Version files synced to 2.1.0 from CHANGELOG.md

[ojowwalker77]

Addressed Greptile review comments:

• Added v2.1.0 changelog entry (fixes failing check)
• Replaced unclear $ARGUMENTS references with descriptive text: "Parse user arguments from the skill invocation (text after the trigger phrase)"