skill-forge

To structure a Claude Code skill package, you need to create a SKILL.md file with platform-aware frontmatter and define clear trigger boundaries. Skill Forge drafts these files, refactors existing skills, and generates Codex-compatible agents/openai.yaml metadata to ensure proper validation and packaging.

Testing AI skill triggers to avoid false positives involves running evaluation workflows and comparing results against baselines. Skill Forge automates trigger eval design, grades outputs, aggregates benchmarks, and provides a viewer to review results and tune descriptions for accurate activation.

Yes, you can use Skill Forge to generate and validate Codex metadata. It creates and validates the agents/openai.yaml file required for Codex-compatible skills, ensuring that your skill packaging aligns with platform-specific metadata layouts and passes validation checks.

Packaging Claude Code and Codex skills requires validating SKILL.md files, organizing references, and keeping helper assets consistent. Skill Forge provides validation helpers and packaging workflows that check frontmatter parsing, verify trigger descriptions, and bundle skill files into a deployable format.

Yes, Skill Forge requires the pyyaml dependency to parse frontmatter and execute its validation workflows. This library allows the skill to read YAML metadata from draft files, validate trigger boundaries, and aggregate benchmark reports during the eval iteration process.

AI skills trigger on unrelated requests when description boundaries are too broad or poorly defined. Skill Forge refines trigger descriptions by running automated evals against baseline prompts, grading the activation accuracy, and tuning the wording to eliminate nearby false positives.