Tests interactive lesson skills by running self-play transcripts and publishing them to a static page.
npx skills add https://github.com/exiao/meta-skills --skill skill-tester-exiao
Please help me install this Agent Skill. Skill: skill-tester Source: https://github.com/exiao/meta-skills/tree/main/skill-tester Command: npx skills add https://github.com/exiao/meta-skills --skill skill-tester-exiao
scripts/build_transcripts.py converts the transcript tree into one collapsible HTML page grouped by arc, with speaker-turn bubbles and a SYNTHETIC warning header, ready for static hosting. - Deterministic post-run audit: Greps instructor-attributed lines for failure patterns (e.g. leaked stage directions) instead of trusting the subagents' self-reported verdicts. - Use Case: After rewriting 20 lessons of an AI course, run this Skill to generate full self-play transcripts for every lesson, deploy them to a static page, and audit that a stage-direction fix actually landed across all transcripts. ## Quick Start Run the skill-tester on my course lessons: self-play every lesson with a calibrated student persona, then build and publish the raw transcripts page.High-intent search queries and answers about installing and using this skill.
Run each lesson via self-play: one agent follows the lesson file as instructor while another plays a calibrated student persona, capturing the full transcript. This Skill dispatches those runs in parallel and publishes every raw transcript to a static HTML page.
Define a calibrated persona matching the real audience, such as a ChatGPT-literate PM who can run setup commands, and pass it as the student's first message. Avoid a false-floor beginner persona, which over-drives fixes and patronizes the actual audience.
Findings-based approaches produce ranked issues plus a few sample transcripts. skill-tester captures the complete turn-by-turn transcript of every lesson and publishes them all raw, so nothing is cherry-picked.
Subagents launched through a framework inherit its system prompt, which does not reflect how a lesson behaves when pasted into a bare chat. Omitting the system parameter entirely mirrors that real-world usage; an empty string is rejected with a 400 error.
No. Self-play subagents reliably self-report clean runs even when bugs exist. Run a deterministic audit afterward: grep instructor-attributed lines for failure patterns and confirm positive changes landed, filtering benign matches.
The script resolves the test-output directory relative to its own location, not the working directory. A zero-lesson result means the sections config or transcript tree is misplaced, so it prints a warning instead of silently writing an empty page.
minicooheiValidate lesson-ids and load lesson files from .cursor/commands/lesson and courses/aiagent.
bonjohenProcess inbound lesson files into researched markdown outputs via pipeline.yaml.
NebulavenusReview UI lesson code for correctness, cleanup, and documentation issues.
bikeindexAutomates end-to-end browser UI testing for Bike Index web applications using Capybara/Selenium.