Interactive skill authoring bench — create, test, and refine Claude Code skills through conversation
Agent
Use when validating a skill draft through multirun testing. Takes a test case library, runs the skill-tester agent N times per case, summarizes consistency across runs, collects user pass/fail judgments with annotations, analyzes failure patterns across cases, proposes targeted skill edits, and manages the re-run loop…
Agent
Use when you need to browse, list, or summarize existing skill drafts and their test history. Scans the drafts directory, reads frontmatter, and reports on draft status.
Agent
Use when a skill draft has been tested against multiple cases and needs refinement based on test results and user feedback. Takes skill content, N test results with thinking traces, and user annotations identifying which results failed and why. Analyzes failure patterns across runs and proposes targeted skill edits.
Agent
Use when testing a skill draft via simulated execution. Takes a skill's SKILL.md content + sample input, reasons through what the skill would produce, and returns a structured evaluation with pass/fail status, issues, thinking trace, and suggested next test cases.