Summary This repository provides a suite of skills designed to systematically improve AI agent reliability and output quality by addressing common failure modes. It employs dual feedback loops (inner-session and cross-session) and leverages critical "peak context" moments in agent execution to apply structured workflows, benefiting developers seeking more robust AI integrations.

How It Works The core architecture features two feedback loops: an "inner loop" for real-time error detection, verification, and recovery within a session, and an "outer loop" for cross-session learning, pattern encoding, and regression testing. Skills are philosophy-driven, targeting specific agent failure modes like scope creep (intent-framed-agent), context degradation (context-surfing), compilation errors (verify-gate), mid-task failures (self-healing), and repeated mistakes (self-improvement). These skills exploit "peak context" moments—after planning, mid-execution, or upon completion—to apply structured workflows, ensuring agents internalize improvements over time.

Quick Start & Requirements Installation is flexible:

• Plugins: Integrate with Claude Code, Codex, and GitHub Copilot CLI via their respective plugin marketplaces.
• CLI Tools: Utilize gh skill (requires v2.90.0+) for GitHub-supported agents or the Agent Skills CLI (npx skills add).
• Manual: Clone the repository and manually copy/symlink desired skill directories. No specific non-default prerequisites beyond tool versions are mentioned.

Highlighted Details

• Dual Feedback Loops: Implements distinct inner (session-based) and outer (cross-session) loops for systematic agent improvement.
• Philosophy-Driven Skills: Each skill targets a specific AI agent failure mode with a guiding principle, translating abstract concepts into actionable workflows.
• CI Variants: Offers headless CI versions (-ci skills) for automated workflows like scanning, reporting, and gating Pull Requests.
• Pipeline Depth: Skills can be used standalone or combined into pipelines, with depth adjusted based on task complexity.

Maintenance & Community Described as a "personal skill testing ground," the repository lacks explicit details on maintainers, community channels, or project roadmaps.

Licensing & Compatibility The README does not specify a software license, which may impact commercial use or integration into proprietary systems.

Limitations & Caveats The plan-interview skill requires explicit invocation (/plan-interview). CI variants are noted as beta. The project's nature as a personal testing ground may influence its maintenance cadence and support structure. The absence of a license is a significant adoption blocker.