How do I install the Pbench skill?

Run npx skillmds@latest add yaphet2015/pbench in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

What does the Pbench skill do?

Use when a coding-agent outcome is wrong or incomplete, or when the user asks to capture, run, replay, compare, report, or operate PBench. It is listed under AI & ML on SkillMD.

Is Pbench safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Pbench?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Pbench free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Pbench?

Yaphet2015 (@yaphet2015) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Pbench by yaphet2015 — Agent Skill

Name: Pbench
Author: Yaphet2015

File contents

PBench

Outcome mismatch

Stop claiming completion.
Ask for user approval to snapshot the failure before repair.
Select the registered source that produced the transcript: codex or claude. Run yk pbench capture --source <registered-source> --yes from the subject repository. If the transcript source is not registered, stop and report that direct capture is unsupported.
Repair the user task.
Read the generated private/authoring-checklist.md and follow the single printed recovery or finalization action.

The snapshot records dirty tracked and untracked state as unresolved private candidates. Do not silently treat those edits as replay input.

Explicit benchmark operation

Headless: yk pbench run --case <case> --agent <agent> --profile <label>.
Manual: yk pbench run --case <case> --manual --profile <label>.
Compare: yk pbench report --profile <label>.

Current built-in runners are codex and claude. Use stable profile labels for comparisons.

Invariants

Private material never enters replay input or benchmark-agent context.
Finalization must pass fresh strict validation of the exact case bundle.

Frequently asked questions

Pbench