Portable packages
Install skills for your chosen agent in this project. Host hooks are a separate integration.
npx skills add i-9-ai/skills --yesBetter skills.By design.
Create, evaluate and evolve agent skills. One responsibility. One reviewable output.
- 24 focused meta-skills
- Apache-2.0
- Portable core
Start anywhere.Improve deliberately.
Keep what works. Change one responsibility at a time.
- 01
Create
Find the job. Research the domain. Build a complete package.
skill-authoring - 02
Verify
Evaluate behavior, security and declared compatibility.
skill-evaluator - 03
Improve
Audit what exists. Keep evidence. Evolve the smallest useful change.
skills-audit - 04
Distribute
Choose an approved package and a deliberate distribution path.
skill-publication
Find the skillfor the job.
Focused packages for the full life of an agent skill.
24 skills
skill-authoring
Use to author or revise a focused skill package from a goal and design, coordinating discovery, synthesis, and evaluation through companion skills. Use skills-discovery for search alone.
Create
skill-design
Use to resolve a skill's responsibility, interface, and design tradeoffs from a brief or merge plan before authoring. It does not discover sources or implement the package.
Create
skill-domain-research
Use to research one proposed skill's subject, reconciling process-owner facts, reviewed skill claims, and current authoritative web sources into a bounded evidence dossier.
Create
skill-evaluator
Use to evaluate a skill package against frozen behavioral cases and a baseline, returning evidence and a readiness verdict. It does not author fixes or publish the skill.
Evaluate
skill-evidence-collection
Use to collect, normalize, and preserve bounded evidence about one skill or skill collection for a later evaluation, lifecycle, or evolution decision. It does not judge or change the target.
Evolve
skill-evolution
Use to update one existing skill from new evidence while preserving its responsibility, provenance, compatibility, and a reviewable change record.
Evolve
skill-icon-design
Use to design or refresh a distinct, accessible SVG icon for one skill package from its responsibility and visual constraints. It does not redesign other package content.
Create
skill-installation
Use to install one approved skill package at an immutable source revision into a selected agent environment and return an auditable installation receipt.
Distribute
skill-lifecycle-review
Use to determine the lifecycle state and next justified transition for one skill from current evidence without editing or publishing the package.
Evaluate
skill-migration
Use to migrate one skill package between collections while preserving behavior, provenance, history, compatibility, and an auditable handoff.
Distribute
skill-naming
Use to select or review a skill's canonical name, grouping related skills by domain and checking scope, cardinality, and known collisions. Return a naming decision without renaming files.
Create
skill-optimization
Use to improve one skill through scored rollouts, bounded candidate edits, and held-out validation. It does not discover sources, author unrelated skills, or publish changes.
Evolve
skill-publication
Use to publish one approved skill package through an authorized distribution channel after exact-revision release gates have passed.
Distribute
skill-routing
Use to decide whether a task should activate one skill, an ordered set of skills, or no skill from supplied catalogs. It recommends a route but does not invoke, install, or modify skills.
Manage
skill-security-review
Use to review one skill package for secrets, private data, unsafe instructions, dangerous scripts, dependencies, permissions, and disclosure risk.
Evaluate
skills-audit
Use to audit a collection of installed or authored skills for inventory, provenance, validation, policy, and drift, producing one findings report.
Manage
skills-catalog
Use to create, synchronize, or validate a repository or global catalog from installed SKILL.md metadata. It maintains inventory and summaries but does not choose a skill for a task.
Manage
skills-catalog-index
Use to combine explicit skill catalogs into a local searchable index, synchronize their change history, or record evidenced evolution operations. It does not maintain source catalogs or choose a skill.
Manage
skills-discovery
Use to discover and qualify existing skills for a specific task, returning pinned candidates and evidence. It does not install, merge, or author a skill.
Create
skills-host-compatibility
Use to verify declared repository-local aliases for a canonical skill collection and return a bounded structural compatibility report.
Evaluate
skills-maintenance-scheduling
Use to turn an approved recurring maintenance goal for a bounded skill collection into an approval-ready schedule proposal. It does not configure a scheduler or execute maintenance.
Manage
skills-refactoring
Use to turn an audit of an existing skill collection into a bounded decomposition and reorganization plan without editing the packages.
Manage
skills-snapshot
Create, verify, list, restore, and explicitly prune deterministic snapshots of a caller-selected skill collection or package. Use before risky skill maintenance; do not use to install or refactor skills.
Manage
skills-synthesis
Use to synthesize useful contributions from two or more reviewed skills into one coherent synthesis plan. It does not search for candidates, write the final package, or certify its quality.
Create
No skills match this search.
Try another responsibility or clear the filters.
Your agent.Your install path.
Portable skills, a host plugin, or the toolkit CLI. Choose what you need.
Codex plugin
Skills, session hook and catalog MCP. Requires Codex CLI and Node.js 24+.
codex plugin marketplace add i-9-ai/skills --ref main
codex plugin add i9-skills@i9-skills
codex plugin listToolkit CLI
Run catalog and skill management commands with Node.js 24+.
npx --yes @i-9.ai/skills catalog overviewPublished registry package; running the CLI is separate from installing skills.
Command referenceEvidence before confidence.
Portable core
Agent Skills Markdown and bundled resources.
Explicit limits
Structural checks and provider pilots cover different things.
Still experimental
No universal compatibility or production-quality claim.
Codex and Claude plugin pilots exercise selected native paths. Copilot CLI evidence covers ephemeral legacy-plugin MCP discovery and direct calls, without a model turn or persistent installation. Icons and hooks remain host-dependent.