HaluCatch
Paste the following prompt into your AI chat to install this skill:
Please install @user_48539314/halucatch into my AI assistant using https://skillhub.cn/install/skillhub.md.
About this skill
Problem: Why AI can fail while executing a Skill
A SKILL.md usually describes steps in natural language, but an AI may misread columns, change thresholds, skip edge cases, or present correlation as causation. HaluCatch reviews whether a Skill folder gives the AI enough constraints to run a task stably and reproducibly.
How it works
It passes the target directory to the local script halucatch_core.py, which completes L1 file scanning, L2 multi-dimension evaluation, and L3 report generation in one run:
- Code-engineering Skills: checks .py foundations, hard-coded paths, input validation, --validate mode, and risks of AI rewriting code.
- Methodology-only Skills: checks step inputs and outputs, exception branches, examples, output format, and self-verification.
- Reports: generates standard, professional, and AI-action versions in reports/; repair suggestions require user confirmation before application.
Boundaries
HaluCatch performs offline reliability review only. It does not cover SQL injection, GDPR compliance, business correctness, or performance optimization. It reads the target Skill directory and writes reports, so it is not strictly read-only; large files, binaries, batch directories, and cross-path access are not supported.
Use Cases
- Before release, audit a Skill folder with SKILL.md and .py files to verify pipeline, code logic, and guardrails.
- After editing SKILL.md, rerun the local audit and compare risks against the previous report.
- Check whether a methodology-only Skill defines steps, exception branches, and output format clearly enough for consistent AI execution.
- Quickly scan a Skill directory for missing SKILL.md, hard-coded paths, or brittle exception handling.
Best For
- Skill authors with .py pipelines who need to confirm script parameters, column checks, and error branches before release.
- Docs owners writing methodology-only Skills who want consistent AI output format and boundary rules.
- Engineering leads gating AI agent quality who need quick re-audits and risk diffs after Skill edits.
- Data analysts turning spreadsheet workflows into Skills who need clear statistical definitions and interpretation guardrails.
Related Skills
Turn vague requests into structured prompts with role, task, context, and constraints, outputting only usable prompts without executing the task.
A skill for game NPC and enemy AI that separates decision, steering, and pathfinding, covering FSMs, behavior trees, A*, path caching, and pitfalls.
Provides multi-turn emotional companion chat through a third-party Agent API and supports continuing sessions via session_id.
An evaluation-driven workflow for diagnosing, optimizing, validating, and logging Agent Skills based on design patterns and anti-patterns.