anistark/sutras/sutras

sutras bench <target>
    Benchmark a skill across models with Claude Code.
    --models: Comma-separated model IDs or aliases, optionally with @effort
    --baseline: Model to compare against (default: bench.baseline or first)
    --runs: Runs per case per model
    --cases: Comma-separated subset of case names
    --max-cost: Hard cap in USD
    --yes/-y (flag): Skip the approval prompt (requires --max-cost)
    --dry-run (flag): Show models, plan, and estimate without running
    --pilot (flag): Run one case per model first to calibrate the estimate
    --record (flag): Write the latest complete run into sutras.yaml
    --report: Also write a Markdown report (JSON history is always saved)
    --history (flag): List previous bench runs
    --parallel: Concurrent runs
    --keep-sandbox (flag): Keep run workspaces for debugging
    --path: Custom skills directory to search for NAME
 
sutras build <name>
    Build a distributable package for a skill.
    --output/-o: Output directory for the package (default: ./dist)
    --no-validate (flag): Skip validation before building
 
sutras docs <name>
    Generate documentation for a skill.
    --output/-o: Output file or directory (default: print to stdout)
    --no-supporting (flag): Exclude supporting file contents (examples.md, etc.)
 
sutras eval <name>
    Evaluate a skill using configured metrics.
    --verbose/-v (flag): Enable verbose evaluation output
    --no-history (flag): Don't save evaluation results to history
    --show-history (flag): Show evaluation history for this skill
 
sutras info <name>
    Show detailed information about a skill.
 
sutras install <source>
    Install a skill from various sources.
    --version/-v: Specific version (for registry installs)
    --registry/-r: Registry to install from (for registry installs)
 
sutras list
    List available skills.
    --local (flag): Include project skills from .claude/skills/
    --global (flag): Include global skills from ~/.claude/skills/
 
sutras new [name]
    Create a new skill with proper structure.
    --description/-d: Skill description (what it does and when to use it)
    --author/-a: Skill author name
    --template/-t: Template to use (run with --list-templates to see options)
    --list-templates (flag): List available skill templates
    --global (flag): Create in global skills directory (~/.claude/skills/)
 
sutras publish [skill_path]
    Publish a skill to a registry.
    --registry/-r: Registry to publish to (default: default registry)
    --pr (flag): Use pull request workflow instead of direct push
    --build-dir/-b: Custom build directory
 
sutras registry add <name> <url>
    Add a new registry.
    --namespace/-n: Default namespace for this registry
    --auth-token/-t: Authentication token
    --priority/-p: Registry priority (higher = checked first)
    --default (flag): Set as default registry
 
sutras registry build-index <registry_path>
    Generate index.yaml for a local registry.
    --output/-o: Output path for index.yaml (default: <registry_path>/index.yaml)
 
sutras registry list
    List configured registries.
 
sutras registry remove <name>
    Remove a registry.
 
sutras registry update [name]
    Update cached registry indexes.
    --all (flag): Update all registries
 
sutras setup
    Install the sutras skill into Claude Code's global skills directory.
    --check (flag): Show what would be installed without making changes
    --uninstall (flag): Remove the sutras skill from Claude Code
 
sutras test <name>
    Run tests for a skill.
    --verbose/-v (flag): Enable verbose test output
    --fail-fast/-x (flag): Stop on first test failure
 
sutras uninstall <skill_name>
    Uninstall a skill.
    --version/-v: Specific version to uninstall (default: all versions)
 
sutras update
    Check for updates and upgrade sutras to the latest version.
    --version/-v: Update to a specific version (default: latest)
    --check (flag): Only check if an update is available, don't install
    --skip-pi (flag): Skip updating the pi extension
    --skip-skill (flag): Skip refreshing the global skill (~/.claude/skills/sutras/)
 
sutras validate <name>
    Validate a skill's structure and metadata.
    --strict (flag): Enable strict validation (warnings become errors)

---
name: my-skill
description: >
  What it does and when the agent should use it.
  Be specific — the description drives skill discovery.
allowed-tools: Read, Write, Bash
---

version: "1.0.0"
author: "Your Name"
license: "MIT"
 
capabilities:
  tools: [Read, Write, Bash]
  dependencies:
    - name: "@utils/common"
      version: "^1.0.0"
 
tests:
  cases:
    - name: "basic-test"
      inputs: {example: "value"}
      expected: {status: "success"}
 
eval:
  framework: "ragas"
  metrics: ["faithfulness", "answer_relevancy"]
  threshold: 0.7
 
distribution:
  tags: ["automation", "pdf"]
  category: "document-processing"