XIIGen AI Agent Skills - reusable Claude, Codex, Cursor and AI coding agent skills

Open-source AI agent skills for safer Claude, Codex, Cursor and AI coding workflows. XIIGen AI Agent Skills gives coding agents reusable planning, review, execution-gate, testing, documentation, and no-false-completion workflows that can be adapted to real projects.

License: Apache-2.0 Skills Safety

What This Is

XIIGen AI Agent Skills is a reusable skill pack for AI coding assistants:

  • Agent Skills-style folders in agent-skills/.
  • Claude-compatible skills in claude/skills/.
  • Codex-compatible skills in codex/skills/.
  • Cursor rules in cursor/rules/*.mdc.
  • Canonical universal skills in skills/universal/.
  • Architecture templates in guides/architecture/.
  • Stack-skill factory guides in guides/stack-skill-factories/.
  • Adaptation map and runbook for generating target-project rules.
  • npm and VS Code/Open VSX scaffolds for future distribution.

The ready-to-use SKILL.md files are universal-only. Architecture-dependent and stack-dependent material stays as templates or factory guidance that must be adapted from the target project.

Searchable Use Cases

  • Claude Code skills for planning, review, testing and safer completion.
  • Codex skills for project-rule loading, execution gates and evidence checks.
  • Cursor rules for AI coding workflows and no-false-completion behavior.
  • Agent Skills-compatible folders for cross-agent reuse.
  • AI coding agent review skills, debugging skills and documentation sync skills.
  • MCP-compatible guidance skills without pretending to be an MCP server.

Quick Install From A Clone

node cli/xiigen-agent-skills.js init
node cli/xiigen-agent-skills.js add cursor-rules
node cli/xiigen-agent-skills.js add claude-skills
node cli/xiigen-agent-skills.js add codex-skills

After npm publication, the same flow is intended to work as:

npx @xiigen/agent-skills init
npx @xiigen/agent-skills add planning
npx @xiigen/agent-skills add cursor-rules

Install Manually

Target Copy from Copy to
Agent Skills-style tools agent-skills/ target project agent-skills/
Claude claude/skills/ target project .claude/skills/
Codex codex/skills/ target project .agents/skills/
Cursor cursor/rules/*.mdc target project .cursor/rules/

Review copied files before committing them to a target project.

Adapt The Full Kit

  1. Copy adaptation/project-profile.template.yaml into the target project as project-profile.yaml.
  2. Fill stack, architecture, commands, CI, project rules, forbidden actions and evidence expectations.
  3. Read adaptation/ADAPTATION-RUNBOOK.md.
  4. Run the render script or ask Claude, Codex or Cursor to follow the runbook.
  5. Review ADAPTATION_REPORT.md, selected mappings, blockers and smoke tests.

The render script produces neutral adapted artifacts first. Tool-specific Claude, Codex and Cursor files are created by following the selected adapter instructions after review.

Repository Map

  • skills/universal/: canonical universal skill source.
  • agent-skills/: Agent Skills-style distribution copy.
  • claude/: Claude install instructions and skill copy.
  • codex/: Codex install instructions and skill copy.
  • cursor/: Cursor install instructions and .mdc rules pack.
  • guides/architecture/: architecture templates for target-project adaptation.
  • guides/stack-skill-factories/: stack-specific skill factory guides.
  • docs/compatibility.md: Claude, Codex, Cursor and Agent Skills compatibility.
  • docs/skill-authoring-guide.md: how to add or adapt skills safely.
  • docs/safety.md: safety boundary for public and target-project use.
  • docs/mcp-compatible-guidance.md: MCP-compatible guidance without server claims.
  • examples/before-after/: prompt-only vs skill-based workflow examples.
  • launch/: Show HN, Reddit, Dev.to/Hashnode and awesome-list drafts.

Key Skills

  • authority-chain: resolve conflicting instructions.
  • project-rules-first: load target project rules before acting.
  • plan-before-execution: plan before risky or multi-file work.
  • evidence-before-final: require proof before completion claims.
  • executor-cannot-self-finish: separate execution from approval.
  • no-secrets: keep secrets, private prompts and raw evidence out.
  • code-execution: run commands intentionally and safely.
  • test-integrity: keep tests meaningful.
  • generated-code-review: review generated code before trusting it.
  • goal-delivery-completeness: verify delivered work against the goal.

Boundary Rule

Universal skills must not include project-specific stack commands, private paths, raw logs, internal prompts, copied worktrees, generated bundles, customer data, model artifacts or source-specific architecture contracts.

Architecture and stack items are guidance for adaptation. Their commands and constraints must come from the target project profile, CI, docs and maintainers.

Safety And Responsibility

These skills and templates are developer-assistance materials. They do not guarantee correct AI outputs, secure code, complete plans or production-ready results. Users remain responsible for review, testing, validation and approval.

This repository provides general software-development guidance only. It is not legal, financial, medical, safety, security certification, compliance or professional advice.

Release Readiness

This repository uses Apache-2.0. Public release readiness is tracked in docs/release-gates.md and includes sanitizer review, provenance review, secret/path/archive scans, examples review, git history/export review, security contact setup and legal review.