0xwilliamortiz
Featured
ponytail-improved
Ponytail is a plugin for AI coding agents that enforces a disciplined ladder of reuse before writing code, reducing code volume by roughly 54% while preserving safety.
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It prioritizes context economy and role separation, producing only the agents strictly justified by project evidence, and provides clear validation levels (L0–L5) to prevent speculative agent bloat.
Who it is for
- Developers using OpenAI Codex or Claude Code who want structured agent packs
- Project maintainers seeking to reduce token waste and maintenance overhead
- Teams needing to audit existing agent packs for routing, role separation, and validation claims
- AI tooling integrators evaluating minimal agent designs for new repositories
Use cases
- Designing a minimal agent pack for a new project from its README and repository structure
- Auditing an existing mature agent pack to identify outdated agents, routing issues, or missing validation
- Generating validation-level reports (customer-facing or internal) to document agent pack quality
- Creating starter templates for both Codex (through a skill) and Claude Code (through file-based templates)
Strengths
- Enforces a minimal agent design principle – only agents with current project evidence are included, avoiding speculative overhead
- Provides a clear validation level framework (L0–L5) with explicit claim rules, preventing overclaiming of runtime readiness
- Supports both Codex (as an installable skill) and Claude Code (via file-based starter packs) from a single repository
- Includes a mature audit profile (`validate_pack.py --profile mature-existing-pack`) that checks routing, role separation, and token economy without requiring the starter-pack shape
Considerations
- Manual copy installation for Codex (future installer only if needed), no package manager integration
- Generated-pack validation only reaches L2 (structural consistency); higher levels (L3–L5) require separate recorded trials that are project-specific
- Claude Code support is file-based and not a native install, requiring users to copy starter templates manually
README quick start
agent-pack-designer
Design minimal sufficient, validation-driven AI agent packs for Codex and Claude Code from project descriptions, README files, technical specs, or existing repository evidence.
agent-pack-designer is a portable Codex skill for creating or auditing project-specific agent packs that are intentionally scoped to the repository's real workflows, risks, and validation evidence. It designs the smallest role set that can handle the work responsibly: context routing, primary workflow ownership, risk or domain review, validation review, and optional trigger-only roles such as architecture, red-team, security, data quality, or release review when the project evidence justifies them.
It focuses on context economy, role separation, reviewer ownership, validation levels, and evidence boundaries, while avoiding speculative agents that add maintenance cost without improving reliability.
It helps produce:
AGENTS.mdfor Codex project rules.CLAUDE.mdfor Claude Code project memory when Claude Code is in scope.agents/context_router.mdfor choosing the minimum useful context.- one primary project-specific workflow agent.
- one risk or domain reviewer only when a concrete recurring risk or review workflow justifies it.
agents/validation_reviewer.mdfor evidence and validation claims.agents/task_spec_short.mdfor non-trivial tasks.- optional extra agents only when current project evidence justifies them.
How This Differs
agent-pack-designer is a design and audit skill, not a large agent marketplace or runtime synchronizer.
| Nearby project type | What it usually does | How agent-pack-designer differs |
|---|---|---|
Config and pack synchronizers such as anywhere-agents | Keep shared rules, skills, hooks, and pack selections in sync across tools and repositories | Designs or audits the smallest project-specific pack from repository evidence before anything needs to be synchronized |
Broad methodology frameworks such as [superpowers](https://github.com/obr |
Description
Design minimal AI agent packs for Codex and Claude from project descriptions.
Related repositories
Similar projects matched by category, topics, and programming language.
powerycy
Featured
goutoujunshi
An AI relationship coach that provides emotional support, relationship science, and actionable strategies through a multidisciplinary knowledge base.
makecindy
Featured
cindy
Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.