Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Interactive workspace discovery - learn what tools, workflows, agents, and hooks are available
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 80% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 712% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 285% | 0% |
Guide users through the capabilities of this workspace setup.
/help # Interactive guided discovery
/help workflows # Workflow orchestration skills
/help agents # Specialist agents catalog
/help tools # CLI tools (tldr, prove, recall)
/help hooks # Active hooks and what they do
/help advanced # MCP, frontmatter, customization
/help <name> # Deep dive on specific skill/agentUse AskUserQuestion to guide the user:
question: "What are you trying to do?"
header: "Goal"
options:
- label: "Explore/understand a codebase"
description: "Find patterns, architecture, conventions"
- label: "Fix a bug"
description: "Investigate, diagnose, implement fix"
- label: "Build a feature"
description: "Plan, implement, test new functionality"
- label: "Prove something mathematically"
description: "Formal verification with Lean 4"Based on response, show relevant tools:
| Goal | Show | |------|------| | Explore codebase | scout agent, tldr CLI, /explore workflow | | Fix a bug | /fix workflow, sleuth agent, debug-agent | | Build feature | /build workflow, architect agent, kraken agent | | Prove math | /prove skill, lean4 skill, Godel-Prover | | Research docs | oracle agent, nia-docs, perplexity | | Configure workspace | hooks, rules, settings, frontmatter |
Display workflow meta-skills:
markdown## Workflow Skills Orchestrate multi-agent pipelines for complex tasks. | Workflow | Purpose | Agents Used | |----------|---------|-------------| | /fix | Bug investigation → diagnosis → implementation | sleuth → kraken → arbiter | | /build | Feature planning → implementation → testing | architect → kraken → arbiter | | /debug | Deep investigation of issues | debug-agent, sleuth | | /tdd | Test-driven development cycle | arbiter → kraken → arbiter | | /refactor | Code transformation with safety | phoenix → kraken → judge | | /review | Code review and feedback | critic, judge | | /security | Vulnerability analysis | aegis | | /explore | Codebase discovery | scout | | /test | Test execution and validation | arbiter, atlas | | /release | Version bumps, changelog | herald | | /migrate | Framework/infrastructure changes | pioneer, phoenix | **Usage**: Just describe your goal. Claude routes to the right workflow.
Display agent catalog:
markdown## Specialist Agents Spawn via Task tool with subagent_type. ### Exploration & Research | Agent | Purpose | Model | |-------|---------|-------| | scout | Codebase exploration, pattern finding | sonnet | | oracle | External research (web, docs, APIs) | sonnet | | pathfinder | External repository analysis | sonnet | ### Planning & Architecture | Agent | Purpose | Model | |-------|---------|-------| | architect | Feature planning, design docs | sonnet | | plan-agent | Create implementation plans | sonnet | | phoenix | Refactoring & migration planning | sonnet | ### Implementation | Agent | Purpose | Model | |-------|---------|-------| | kraken | TDD implementation, refactoring | sonnet | | spark | Quick fixes, lightweight changes | haiku | ### Review & Validation | Agent | Purpose | Model | |-------|---------|-------| | arbiter | Test execution, validation | sonnet | | critic | Code review | sonnet | | judge | Refactoring review | sonnet | ### Investigation | Agent | Purpose | Model | |-------|---------|-------| | sleuth | Bug investigation, root cause | sonnet | | debug-agent | Issue investigation with logs | sonnet | | profiler | Performance, race conditions | sonnet | ### Documentation & Handoff | Agent | Purpose | Model | |-------|---------|-------| | scribe | Documentation, session summaries | sonnet | | chronicler | Session analysis, learning extraction | sonnet |
Display CLI tools and capabilities:
markdown## Built-in Tools ### TLDR Code Analysis Token-efficient code exploration (95% savings vs reading raw files).
tldr tree src/ # File tree tldr structure src/ --lang python # Code structure (codemaps) tldr search "pattern" src/ # Search files tldr cfg file.py func # Control flow graph tldr dfg file.py func # Data flow graph tldr impact func src/ # Reverse call graph (who calls this?) tldr dead src/ # Find dead code tldr arch src/ # Detect architectural layers
### /prove - Formal Verification
Machine-verified proofs without learning Lean syntax.
/prove every group homomorphism preserves identity /prove continuous functions on compact sets are uniformly continuous
Requires: LM Studio running Godel-Prover model locally.
### Memory System
Store and recall learnings across sessions.
(cd opc && uv run python scripts/core/recall_learnings.py --query "hook patterns")
/remember "Hook X works by..."
### Premortem Risk Analysis
Identify failure modes before they occur.
/premortem plan-file] # Analyze implementation plan for risks
Display active hooks:
markdown## Active Hooks Hooks extend Claude's behavior at key lifecycle points. ### Session Lifecycle | Hook | Event | Purpose | |------|-------|---------| | session-register | SessionStart | Register session in coordination DB | | session-start-recall | SessionStart | Auto-inject relevant learnings | | session-end-cleanup | SessionEnd | Cleanup temp files | | session-outcome | SessionEnd | Prompt for session outcome | ### User Prompt Processing | Hook | Event | Purpose | |------|-------|---------| | skill-activation-prompt | UserPromptSubmit | Suggest relevant skills | | premortem-suggest | UserPromptSubmit | Suggest risk analysis for implementations | ### Tool Interception | Hook | Event | Purpose | |------|-------|---------| | tldr-read-enforcer | PreToolUse:Read | Suggest tldr for large files | | smart-search-router | PreToolUse:Grep | Route to ast-grep for structural search | | file-claims | PreToolUse:Edit | Track which sessions edit which files | | signature-helper | PreToolUse:Edit | Inject function signatures | | import-validator | PostToolUse:Edit | Validate imports after edits | ### Validation | Hook | Event | Purpose | |------|-------|---------| | typescript-preflight | PreToolUse:Bash | Type-check before running | | compiler-in-the-loop | Stop | Run Lean compiler for /prove | ### Subagent Coordination | Hook | Event | Purpose | |------|-------|---------| | subagent-start | SubagentStart | Initialize agent context | | subagent-stop | SubagentStop | Extract learnings from agents |
Display advanced customization:
markdown## Advanced: Customization & Extension ### Skill Frontmatter Skills use YAML frontmatter for metadata and tool restrictions:
name: my-skill description: What it does triggers: "keyword1", "keyword2"] allowed-tools: Bash, Read, Edit] priority: high skills: other-skill] # Auto-load dependencies
### Agent Frontmatter
Agents declare their capabilities:
name: my-agent description: Specialist for X model: inherit # Claude adapter example; Codex uses luna_worker tools: Read, Grep, Glob, Bash]
The `model` field above is Claude-adapter metadata. Codex does not select a model from
skill or agent examples; the bounded `luna_worker` contract remains authoritative.
### MCP Servers
External tool integrations:
| Server | Purpose |
|--------|---------|
| ast-grep | Structural code search/refactoring |
| firecrawl | Web scraping |
| github-search | Search GitHub code/issues |
| morph | Fast file editing (10k tokens/sec) |
| nia | Documentation search |
| perplexity | AI-powered web research |
### Rules (.claude/rules/*.md)
Always-on instructions injected into context:
- `claim-verification.md` - Verify before asserting
- `dynamic-recall.md` - How to use memory system
- `tldr-cli.md` - TLDR usage reference
### Creating Custom Skillsmkdir -p .claude/skills/my-skill
### Creating Custom AgentsFor specific skill/agent lookup:
.claude/skills/<name>/SKILL.md.claude/agents/<name>.mdTo keep help current, dynamically read frontmatter:
bash# List all skills with descriptions for f in .claude/skills/*/SKILL.md; do head -10 "$f" | grep -E "^(name|description):" done # List all agents for f in .claude/agents/*.md; do head -10 "$f" | grep -E "^(name|description):" done
User: /help
Claude:
What are you trying to do?
[ ] Explore/understand a codebase
Find patterns, architecture, conventions
[ ] Fix a bug
Investigate, diagnose, implement fix
[ ] Build a feature
Plan, implement, test new functionality
[ ] Prove something mathematically
Formal verification with Lean 4
[Other: describe your goal]User selects: "Fix a bug"
Claude:
## Bug Fixing Tools
### Quick Path
Just describe the bug. I'll use the /fix workflow automatically.
### Manual Control
- **sleuth** agent: Deep investigation, root cause analysis
- **debug-agent**: Log analysis, state inspection
- **kraken** agent: TDD implementation of fix
### Commands
- `/fix` - Full workflow (investigate → diagnose → implement → test)
- `/debug` - Investigation only
### Tips
- Provide error messages, logs, reproduction steps
- The premortem hook will suggest risk analysis before implementation| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,849 | 3,326 | -31% | 1 | 1 | 0% | 847 | 2,805 | +231% | 0 | 0 | — |
case-02 | fail→pass | 12,240 | 3,504 | -71% | 1 | 1 | 0% | 1,688 | 3,033 | +80% | 0 | 0 | — |
case-03 | fail→pass | 11,804 | 6,189 | -48% | 1 | 1 | 0% | 2,090 | 3,662 | +75% | 0 | 0 | — |
case-04 | fail→pass | 12,462 | 3,728 | -70% | 1 | 1 | 0% | 1,938 | 3,080 | +59% | 0 | 0 | — |
case-05 | fail→pass | 3,138 | 6,866 | +119% | 1 | 1 | 0% | 467 | 3,791 | +712% | 0 | 0 | — |
case-06 | fail→pass | 5,970 | 4,680 | -22% | 1 | 1 | 0% | 873 | 3,358 | +285% | 0 | 0 | — |
case-07 | fail→pass | 12,614 | 4,458 | -65% | 1 | 1 | 0% | 2,147 | 3,233 | +51% | 0 | 0 | — |
case-08 | fail→pass | 13,511 | 2,099 | -84% | 1 | 1 | 0% | 2,435 | 2,837 | +17% | 0 | 0 | — |
case-09 | fail→pass | 9,155 | 1,751 | -81% | 1 | 1 | 0% | 1,561 | 2,743 | +76% | 0 | 0 | — |
case-10 | fail→pass | 10,578 | 2,818 | -73% | 1 | 1 | 0% | 1,609 | 2,943 | +83% | 0 | 0 | — |
case-11 | fail→pass | 14,748 | 2,775 | -81% | 1 | 1 | 0% | 2,439 | 2,903 | +19% | 0 | 0 | — |
case-12 | fail→pass | 14,536 | 3,003 | -79% | 1 | 1 | 0% | 2,287 | 2,933 | +28% | 0 | 0 | — |
case-13 | pass→pass | 13,810 | 3,725 | -73% | 1 | 1 | 0% | 2,378 | 3,032 | +28% | 0 | 0 | — |
case-14 | pass→pass | 14,589 | 4,939 | -66% | 1 | 1 | 0% | 2,394 | 3,360 | +40% | 0 | 0 | — |
case-15 | fail→pass | 17,748 | 6,453 | -64% | 1 | 1 | 0% | 3,070 | 3,584 | +17% | 0 | 0 | — |
case-16 | fail→pass | 10,085 | 2,735 | -73% | 1 | 1 | 0% | 1,635 | 2,922 | +79% | 0 | 0 | — |
case-17 | fail→pass | 10,108 | 3,693 | -63% | 1 | 1 | 0% | 1,667 | 3,090 | +85% | 0 | 0 | — |
case-18 | fail→pass | 12,119 | 2,327 | -81% | 1 | 1 | 0% | 2,132 | 2,984 | +40% | 0 | 0 | — |
case-19 | fail→pass | 9,232 | 2,285 | -75% | 1 | 1 | 0% | 1,674 | 2,880 | +72% | 0 | 0 | — |
case-20 | fail→pass | 12,078 | 2,437 | -80% | 1 | 1 | 0% | 2,100 | 2,946 | +40% | 0 | 0 | — |
case-21 | fail→fail | 7,162 | 2,434 | -66% | 1 | 1 | 0% | 1,172 | 2,876 | +145% | 0 | 0 | — |
case-22 | fail→pass | 14,248 | 5,736 | -60% | 1 | 1 | 0% | 2,166 | 3,469 | +60% | 0 | 0 | — |
case-23 | fail→pass | 12,960 | 4,699 | -64% | 1 | 1 | 0% | 2,133 | 3,328 | +56% | 0 | 0 | — |
case-24 | pass→pass | 2,415 | 1,332 | -45% | 1 | 1 | 0% | 374 | 2,678 | +616% | 0 | 0 | — |
case-25 | fail→pass | 1,943 | 1,799 | -7% | 1 | 1 | 0% | 380 | 2,805 | +638% | 0 | 0 | — |
case-26 | pass→pass | 2,358 | 2,279 | -3% | 1 | 1 | 0% | 442 | 2,872 | +550% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 26 cases were attempted. The headline lift of +77 percentage points is the difference between those two pass rates over the 26 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/29/2026 | +55% |
Other measured skills in the registry, with their headline benchmark lift.