Back to directory

HERO — Anti-OverDefense

A short block of text pasted into a coding agent's configuration file that names four shapes of unnecessary defensive behavior, hashing, edge-case handling, rubric-chasing and overbuilding, so a developer can call out which one just happened instead of arguing about vibes.

HERO — Anti-OverDefense

HERO stands for Hashing, Edge cases, Rubrics and Overbuild, four repeated patterns the project's author observed in coding-agent responses where an agent builds elaborate safeguards around a feature instead of finishing the feature itself. Rather than a tool to install, HERO is a short contract, provided in English and Chinese, meant to be pasted into an agent's own configuration file, alongside per-host notes for where to paste it in Claude Code, Codex, Antigravity, Cursor, GitHub Copilot, Windsurf and Gemini CLI.

A cases folder documents specific observed instances, what was asked, what the agent did, and why it was disproportionate, meant to be quoted back at an agent that argues its own behavior was reasonable rather than pasted into its configuration. An examples folder collects unrelated AGENTS.md and CLAUDE.md files that other contributors adapted for their own projects, shared to show the approach rather than as ready-made HERO variants.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.