claude-all-hands 1.0.0 → 1.0.1

package/.claude/skills/git-ops/SKILL.md

@@ -3,14 +3,45 @@ name: git-ops
 description: Use when performing git operations (commits, PRs, conflicts, branch mgmt). Planner auto-derives from plan context. Main agent (direct mode) may prompt user.
 ---
 
-# Git Operations Skill
+<objective>
+Standardize git operations across planning and direct modes. Ensures consistent commit messages, PR bodies, conflict resolution, and branch management while enforcing safety rules for destructive operations.
+</objective>
+
+<quick_start>
+- **Commit**: `git diff --cached` then generate `<type>(<scope>): <description>`
+- **PR**: Use `gh pr create -a @me` with template below
+- **Conflict**: Analyze with `git show :1/:2/:3:<file>`, resolve, stage
+- **Branch**: Always use prefix (feat/, fix/, chore/, etc.)
+</quick_start>
+
+<success_criteria>
+- Commits follow Conventional Commits format
+- PR bodies include Summary, Changes with inline links, Plan Reference
+- Conflicts resolved with intent preserved from commit history
+- No destructive operations without explicit user confirmation
+</success_criteria>
+
+<constraints>
+**ALWAYS AskUserQuestion before:**
+- Force push (`--force`, `-f`)
+- Hard reset (`reset --hard`)
+- Branch deletion (`branch -D`)
+- Rebase of shared branches
 
-## Usage Context
+**NEVER:**
+- Force push main/master without explicit confirmation
+- Skip pre-commit hooks unless requested
+- Amend commits not authored by current user
+</constraints>
+
+<workflow name="usage-context">
 
 **Planner Agent (plan mode)**: Auto-generate all commit/PR content from plan context
 **Main Agent (direct mode)**: Prompt user if insufficient context
 
-## Commit Message Generation
+</workflow>
+
+<workflow name="commit-message">
 
 Conventional Commits: `<type>(<scope>): <description>`
 
@@ -26,7 +57,9 @@ Types: feat, fix, chore, refactor, docs, test, style, perf, ci, build
 2. If clear context: auto-generate
 3. If ambiguous: AskUserQuestion for message
 
-## PR Creation
+</workflow>
+
+<workflow name="pr-creation">
 
 ### Pre-PR Checklist
 1. Check if README/docs need updates for this change
@@ -65,7 +98,9 @@ EOF
 - If plan exists: derive from plan
 - If no plan: AskUserQuestion for PR details
 
-## Merge Conflict Resolution
+</workflow>
+
+<workflow name="conflict-resolution">
 
 ### Detection
 ```bash
@@ -88,7 +123,9 @@ git show :3:<file>  # theirs
 3. Propose resolution with explanation
 4. Apply fix, stage, continue rebase/merge
 
-## Branch Management
+</workflow>
+
+<workflow name="branch-management">
 
 ### Prefixes (enforced)
 feat/, fix/, chore/, refactor/, exp/, docs/, quick/
@@ -106,20 +143,11 @@ git branch --merged main  # list candidates
 git branch -d <branch>
 ```
 
-## Safety Rules
-
-**ALWAYS AskUserQuestion before:**
-- Force push (`--force`, `-f`)
-- Hard reset (`reset --hard`)
-- Branch deletion (`branch -D`)
-- Rebase of shared branches
+</workflow>
 
-**NEVER:**
-- Force push main/master without explicit confirmation
-- Skip pre-commit hooks unless requested
-- Amend commits not authored by current user
+<workflow name="checkpoint-integration">
 
-## Checkpoint Integration (Planner Only)
+**Planner Only**
 
 ### Non-Final Checkpoint
 After user testing approval:
@@ -135,3 +163,5 @@ git add -A
 git commit -m "<type>(<scope>): complete <plan name>"
 gh pr create --title "<plan name>" --body "<auto-derived>"
 ```
+
+</workflow>

package/.claude/skills/hook-development/SKILL.md

@@ -3,11 +3,33 @@ name: hook-development
 description: Use when creating hooks, implementing PreToolUse/PostToolUse/Stop validation, setting up event-driven automation, or working with hook events and exit codes.
 ---
 
-# Hook Development
+<objective>
+Enable event-driven automation in Claude Code's lifecycle. Hooks provide deterministic control - ensuring certain actions always happen rather than relying on LLM choice.
+</objective>
 
-Hooks are event-driven automation that execute at specific points in Claude Code's lifecycle. Use for deterministic control - ensuring certain actions always happen rather than relying on LLM choice.
+<quick_start>
+1. Identify event (PreToolUse, Stop, etc.)
+2. Choose type: prompt (flexible) vs command (deterministic)
+3. Add hook config to `.claude/settings.json`
+4. For command hooks, create script in `.claude/hooks/scripts/`
+5. Test with `claude --debug`
+</quick_start>
+
+<success_criteria>
+- Hook registered (verify with `/hooks`)
+- Event triggers at correct lifecycle point
+- Exit codes handled properly (0=success, 2=blocking)
+- JSON output parsed correctly if used
+</success_criteria>
+
+<constraints>
+- Prompt hooks have 30s default timeout
+- Command hooks have 60s default timeout
+- Exit code 2 blocks operation (stderr fed to Claude)
+- Other non-zero exits are non-blocking (stderr logged)
+</constraints>
 
-## Quick Reference
+## Event Reference
 
 | Event | When | Control |
 |-------|------|---------|
@@ -85,15 +107,7 @@ Bash/Python scripts for deterministic checks:
 - `$CLAUDE_PROJECT_DIR` - Project root (always available)
 - `$CLAUDE_ENV_FILE` - Persist env vars (SessionStart only)
 
-## Deep References
-
-- `references/event-reference.md` - All events with input/output specs
-- `examples/prompt-hooks.md` - Prompt-based hook examples
-- `examples/command-hooks.md` - Bash/Python script examples
-- `~/.claude-code-docs/docs/hooks.md` - Official reference
-- `~/.claude-code-docs/docs/hooks-guide.md` - Official guide
-
-## Implementation Workflow
+<workflow name="implementation">
 
 1. Identify event (PreToolUse, Stop, etc.)
 2. Choose type: prompt (flexible) vs command (deterministic)
@@ -101,3 +115,13 @@ Bash/Python scripts for deterministic checks:
 4. For command hooks, create script in `.claude/hooks/scripts/`
 5. Test with `claude --debug`
 6. Use `/hooks` to verify registration
+
+</workflow>
+
+## Deep References
+
+- `references/event-reference.md` - All events with input/output specs
+- `examples/prompt-hooks.md` - Prompt-based hook examples
+- `examples/command-hooks.md` - Bash/Python script examples
+- `~/.claude-code-docs/docs/hooks.md` - Official reference
+- `~/.claude-code-docs/docs/hooks-guide.md` - Official guide

package/.claude/skills/orchestration-idols/SKILL.md

@@ -3,21 +3,51 @@ name: orchestration-idols
 description: Use when designing agent orchestration, memory systems, or multi-agent coordination. Contains patterns from wshobson/agents and claude-flow repositories with doc URLs for deeper research.
 ---
 
-# Idol Repository Patterns
-
-Patterns from production agent orchestration systems. Use doc URLs below for deeper research when implementing specific features that should be inspired by these patterns.
-
-## wshobson/agents
+<objective>
+Patterns from production agent orchestration systems. Use doc URLs for deeper research when implementing features inspired by these patterns.
+</objective>
+
+<quick_start>
+1. Identify pattern category needed (registry, memory, coordination)
+2. Review summary patterns below
+3. Use research-tools skill with doc URLs for implementation details
+</quick_start>
+
+<success_criteria>
+- Appropriate pattern selected for use case
+- Doc URLs consulted for implementation specifics
+- Pattern adapted to Claude Code context
+</success_criteria>
+
+<workflow>
+### Pattern Selection
+
+| Need | Pattern | Source |
+|------|---------|--------|
+| Dynamic agent loading | Registry Pattern | wshobson/agents |
+| Decentralized coordination | Hive-Mind Consensus | claude-flow |
+| Fast semantic retrieval | Hybrid Memory (Vector) | claude-flow |
+| Session persistence | Context Recovery | claude-flow |
+| Load distribution | Work-Stealing | claude-flow |
+
+### When to Research Deeper
+Use research-tools skill with doc URLs when:
+- Implementing specific patterns (need exact structures)
+- Adapting memory/reasoning systems
+- Building new agent types
+- Debugging coordination issues
+</workflow>
 
+<examples>
+### wshobson/agents
 Simpler agent coordination patterns.
 
 **Documentation:**
 - Architecture: https://github.com/wshobson/agents/blob/main/docs/architecture.md
 - README: https://github.com/wshobson/agents/blob/main/README.md
 
-## claude-flow
-
-Enterprise AI orchestration with extensive tooling. Achieves Claude Code goals via non-Claude-Code-API tooling - useful for understanding what's possible and adapting patterns.
+### claude-flow
+Enterprise AI orchestration with extensive tooling.
 
 **Core Documentation:**
 - README: https://github.com/ruvnet/claude-flow/blob/main/README.md
@@ -39,30 +69,18 @@ Enterprise AI orchestration with extensive tooling. Achieves Claude Code goals v
 - Reasoning: https://github.com/ruvnet/claude-flow/blob/main/docs/reasoning/README.md
 - AgentDB Integration: https://github.com/ruvnet/claude-flow/blob/main/docs/agentdb/AGENTDB_INTEGRATION_PLAN.md
 
-## Key Patterns Summary
+### Key Patterns Summary
 
-### Registry Pattern
-Dynamic agent loading via central registry. Agents discovered by config, not hardcoded.
+**Registry Pattern**: Dynamic agent loading via central registry. Agents discovered by config, not hardcoded.
 
-### Hive-Mind Consensus
-Decentralized coordination, not top-down control. Shared memory for state alignment.
+**Hive-Mind Consensus**: Decentralized coordination, not top-down control. Shared memory for state alignment.
 
-### Hybrid Memory
+**Hybrid Memory**:
 - Vector search (AgentDB/HNSW): Fast semantic retrieval
 - SQLite fallback (ReasoningBank): Reliable, no API keys
 - Auto-consolidates successful patterns
 
-### Context Management
-- 5-exchange conversation limit per task
-- Context recovery files for persistence across sessions
+**Context Management**: 5-exchange conversation limit per task. Context recovery files for persistence.
 
-### Work-Stealing Load Balancing
-Dynamic task redistribution from overloaded to underutilized agents.
-
-## When to Research Deeper
-
-Use research-tools skill with these URLs when:
-- Implementing specific patterns (need exact structures)
-- Adapting memory/reasoning systems
-- Building new agent types
-- Debugging coordination issues
+**Work-Stealing Load Balancing**: Dynamic task redistribution from overloaded to underutilized agents.
+</examples>

package/.claude/skills/repomix-extraction/SKILL.md

@@ -1,75 +1,69 @@
 ---
 name: repomix-extraction
-description: Directory extraction using repomix for pattern discovery. Use when skills/directives fall short and you need to understand codebase patterns in a specific directory.
+description: Directory extraction using repomix. Use for ANY directory exploration or multi-file read - always more efficient than individual Read calls.
 ---
 
-# Repomix Extraction
-
-Pack directory contents into AI-friendly format for comprehensive pattern analysis.
-
-## When to Use
-
-- Skills/directives don't cover the directory you need to understand
-- Need to discover patterns in unfamiliar codebase areas
-- Creating specialist agents that need repo pattern knowledge
-- Analyzing code structure before making changes
-
-## Quick Reference
+<objective>
+Pack directory contents into AI-friendly format for comprehensive pattern analysis. ALWAYS prefer over individual Read calls when exploring directories or reading multiple files.
+</objective>
 
+<quick_start>
 ```bash
-# Pack directory (outputs repomix-output.xml)
-npx repomix@latest path/to/directory
+# Pack directory to stdout (NO FILES CREATED)
+npx repomix@latest --stdout path/to/directory
 
 # Include specific patterns
-npx repomix@latest --include "**/*.ts,**/*.md" path/to/directory
-
-# Exclude noise
-npx repomix@latest --ignore "**/*.log,node_modules/" path/to/directory
+npx repomix@latest --stdout --include "**/*.ts,**/*.md" path/to/directory
 
-# Compress for large directories (tree-sitter extraction)
-npx repomix@latest --compress path/to/directory
-
-# Output to stdout (for piping)
-npx repomix@latest --stdout path/to/directory
+# Compress for large directories
+npx repomix@latest --stdout --compress path/to/directory
 ```
+</quick_start>
+
+<success_criteria>
+- Output streams to stdout (no file created)
+- All relevant files included in single output
+- Patterns identified from combined content
+</success_criteria>
 
-## Extraction Process
+<constraints>
+- NEVER use repomix without `--stdout` (creates files in project root)
+- Always stream to stdout, never create output files
+</constraints>
 
-### 1. Pack Directory
+<workflow>
+### 1. Pack Directory to Stdout
 ```bash
-npx repomix@latest path/to/directory
+npx repomix@latest --stdout path/to/directory
 ```
-Output: `repomix-output.xml` in current directory
 
-### 2. Read Output
-Read the XML file to analyze:
+### 2. Analyze Output
+From stdout output, identify:
 - File structure conventions
 - Directory organization patterns
-- Naming conventions (files, functions, variables)
+- Naming conventions
 - Code style/idioms
 - Import/export patterns
-- Error handling approaches
 
 ### 3. Synthesize Patterns
 Extract actionable patterns:
 - Recurring code patterns
 - Domain-specific conventions
 - Anti-patterns to avoid
-- Guidelines for system prompts or implementation
+- Guidelines for implementation
+</workflow>
 
-## Options Reference
+<examples>
+### Options Reference
 
 | Flag | Use |
 |------|-----|
+| `--stdout` | REQUIRED - Output to stdout, no file |
 | `--include "glob"` | Include specific file patterns |
 | `--ignore "glob"` | Exclude patterns |
 | `--compress` | Tree-sitter extraction (large dirs) |
-| `--stdout` | Output to stdout instead of file |
-| `--remote URL` | Pack remote repo (GitHub shorthand) |
-
-## Output Format
 
-Return extracted patterns as:
+### Output Format
 ```markdown
 ## Directory Analysis: [path]
 
@@ -87,3 +81,11 @@ Return extracted patterns as:
 ### Recommendations
 [Actionable guidance based on patterns]
 ```
+</examples>
+
+<anti_patterns>
+- Using individual Read calls when exploring directories
+- Running repomix without `--stdout` flag
+- Creating output files in project root
+- Reading files one by one when batch extraction available
+</anti_patterns>

package/.claude/skills/research-tools/SKILL.md

@@ -3,30 +3,42 @@ name: research-tools
 description: Use when need to "search the web", "research [topic]", "extract URL content", or "find sources". Provides web search (Tavily), deep research (Perplexity), and X/Twitter insights (Grok). Only for curator/researcher agents.
 ---
 
-# Research Tools
-
-External research capability for curator and researcher agents only.
-
-## Quick Reference
+<objective>
+External research capability for curator and researcher agents. Provides web search, deep research with citations, and real-time social signals via envoy commands.
+</objective>
 
+<quick_start>
 ```bash
-# Deep research with citations (Perplexity) - synthesized findings
+# Deep research (Perplexity)
 envoy perplexity research "query"
-envoy perplexity research "query" --grok-challenge  # validate via X search
-
-# X/Twitter search (Grok) - real-time social signals, tech community insights
-envoy xai search "query"
-envoy xai search "query" --results-to-challenge "findings"  # challenger mode
 
-# Web search (Tavily) - find sources, get URLs (includes LLM answer by default)
+# Web search (Tavily)
 envoy tavily search "query"
-envoy tavily search "query" --max-results 10  # optional, API default is 5
 
-# Extract content from URLs (Tavily) - full page content
+# Extract URL content
 envoy tavily extract "url1" "url2"
 ```
+</quick_start>
+
+<success_criteria>
+- JSON response with `status: success` and `data` containing findings
+- Sources/citations included for synthesis queries
+- Content extracted for known URLs
+</success_criteria>
+
+<constraints>
+- Only curator/researcher agents may use these tools
+- GitHub content: use `gh` CLI instead of extract
+- All commands return JSON - parse `data.content` or `data.results`
+</constraints>
 
-## When to Use What
+<workflow>
+### 1. Scope the Query
+- What specifically needs to be learned?
+- What will the findings be used for?
+- What level of depth is needed?
+
+### 2. Choose Approach
 
 | Need | Tool | Cost |
 |------|------|------|
@@ -35,91 +47,49 @@ envoy tavily extract "url1" "url2"
 | X/Twitter community insights | `xai search` | Medium |
 | Find sources, discover URLs | `tavily search` | Medium |
 | Get full content from known URL | `tavily extract` | Low |
-| Agentic: search then deep-dive | search → extract | Medium |
-
-## Decision Tree
-
-```
-Need information?
-├─ Know the exact URL? → tavily extract (or delegate to specialist agent)
-├─ Need to find sources? → tavily search → extract promising URLs
-├─ Tech research for planning? → perplexity research --grok-challenge (default)
-└─ Quick answer, no validation? → perplexity research
-```
-
-## GitHub Content
 
-For GitHub repositories, files, issues, PRs - use `gh` CLI instead of extract:
+### 3. Process Results
+All tools return JSON. Parse `data.content` or `data.results` for findings.
+</workflow>
 
+<examples>
+### Command Reference
 ```bash
-# Get file content
-gh api repos/{owner}/{repo}/contents/{path}
-
-# Get issue/PR
-gh issue view {number} --repo {owner}/{repo}
-gh pr view {number} --repo {owner}/{repo}
-
-# Search code
-gh search code "query" --repo {owner}/{repo}
-```
-
-## Research Workflow
-
-### 1. Scope the Query
-Before calling any tool, clarify:
-- What specifically needs to be learned?
-- What will the findings be used for?
-- What level of depth is needed?
-
-### 2. Choose Approach
+# Deep research with citations (Perplexity)
+envoy perplexity research "query"
+envoy perplexity research "query" --grok-challenge  # validate via X search
 
-**Option A: Pre-synthesized (faster)**
-```bash
-envoy perplexity research "[topic] best practices 2025"
-```
-Returns synthesized findings with citations. Good for broad questions.
+# X/Twitter search (Grok)
+envoy xai search "query"
+envoy xai search "query" --results-to-challenge "findings"  # challenger mode
 
-**Option B: Agentic (more control)**
-```bash
-# Find sources
-envoy tavily search "[topic]"
+# Web search (Tavily)
+envoy tavily search "query"
+envoy tavily search "query" --max-results 10
 
-# Extract full content from promising URLs
+# Extract content from URLs (Tavily)
 envoy tavily extract "url1" "url2"
-```
-Returns raw content for you to process. Good when you need specific details.
-
-### 3. Process Results
 
-All tools return JSON. Parse `data.content` or `data.results` for findings.
-
-## Query Tips
-
-**For best practices:**
-```
-"[topic] best practices 2025"
-```
-
-**For implementation:**
-```
-"how to implement [specific thing] in [context]"
+# GitHub content (use gh CLI)
+gh api repos/{owner}/{repo}/contents/{path}
+gh issue view {number} --repo {owner}/{repo}
 ```
 
-**For comparison:**
+### Decision Tree
 ```
-"[option A] vs [option B] for [use case]"
+Need information?
+├─ Know the exact URL? → tavily extract
+├─ Need to find sources? → tavily search → extract promising URLs
+├─ Tech research for planning? → perplexity research --grok-challenge
+└─ Quick answer, no validation? → perplexity research
 ```
 
-## Output Format
-
-After research, return to parent agent:
-
+### Output Format
 ```markdown
 ## Research: [Topic]
 
 ### Key Findings
 - Finding with context
-- Finding with context
 
 ### Sources
 - [Source URL] - what was relevant
@@ -127,3 +97,11 @@ After research, return to parent agent:
 ### Recommendations
 What to do based on findings.
 ```
+</examples>
+
+<anti_patterns>
+- Using extract for GitHub content (use `gh` CLI instead)
+- Skipping scope clarification before research
+- Not parsing JSON response properly
+- Using research tools from non-curator/researcher agents
+</anti_patterns>