Skip to main content
skill CX Codex

replay-regression — Claude Code skill: PoC replay

The replay-regression Claude Code skill independently tests a supplied PoC against fresh builds from before and after a change. It runs the same input on S0 and S1, then records what happened.

CC-BY-SA-4.0 668 installs ★ 8,341 updated 2026-10-24

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add trailofbits/skills --skill replay-regression -a claude-code
Open on GitHub

What replay-regression does

The skill checks a complete reproduction package and runs the PoC in a context separate from the original reproducer. It builds and runs actual S0 and S1 inside its own attempt directory, then executes the unchanged PoC with the same input on both. Every build directory goes into build_dirs.

It tests whether the claimed user or attacker can reach the real interface, saving setup details, versions, commands, observations, and separate logs. It then writes and seals replay.json, linking it to the reproduction record and the original PoC and input. The attempt needs frozen source records, a complete sealed reproduction package, and a registry assignment; failed prerequisites stop execution. The skill does not write or repair a PoC. Any changed PoC needs a new reproduction and replay.

What it helps with

  • independently replay a supplied PoC
  • compare S0 and S1 using identical input
  • test interface access with the claimed permissions
  • record blockers or divergent results in replay.json

Example prompts for the agent

  • Replay this PoC independently against S0 and S1 from the reproduction package.
  • Run the supplied PoC in a separate attempt and record the result in replay.json.
  • Check interface access with the claimed user permissions and save the observations.

How to install replay-regression

1Skills CLI (Vercel)

npx skills add trailofbits/skills --skill replay-regression -a claude-code

2Manually: put the files in place

~/.claude/skills/replay-regression/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/replay-regression/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about replay-regression

How do I install replay-regression in Claude Code?

Run npx skills add trailofbits/skills --skill replay-regression -a claude-code. The skill file is at ~/.claude/skills/replay-regression/SKILL.md.

What does replay-regression do?

It independently runs a supplied PoC against S0 and S1, checks whether the claimed failure appears, and saves a sealed replay.json record.

Can replay-regression modify a PoC?

No. It tests the supplied PoC unchanged; edits require a new reproduction and replay.

Similar skills

All skills for Testing and QA →

From the same set95

The whole set →
prove-patch prove-patch — a skill for Claude Code validates a local security patch with scoped cases and a three-arm comparison. It runs when a patch commit needs independent proof. CX Codex +1 676installs hunt-regressions hunt-regressions — a skill for Claude Code that discovers potential regressions before reproduction. Runs when analyzing code changes and their callers. CX Codex +1 675installs reproduce-regression The reproduce-regression Claude Code skill reproduces a regression candidate on live systems before and after a change. It builds a PoC and compares both versions using the same input. CX Codex +1 671installs report-regressions report-regressions — a skill for Claude Code builds regression reports from saved campaign decisions and links findings to evidence. It runs when a campaign report is created or refreshed. CX Codex +1 670installs prepare-regression prepare-regression — a skill for Claude Code that pins source revisions and gathers commit or PR context for a regression investigation. It runs during investigation setup. CX Codex +1 670installs adjudicate-regression adjudicate-regression — a skill for Claude Code and Codex that reviews reproduction evidence to confirm whether a flaw is a true regression or an exclusion. It evaluates live artifacts before logging issues. CX Codex +1 669installs cvss-scoring The cvss-scoring Claude Code skill turns a vulnerability description into a CVSS v4.0 vector. It activates for requests to score a finding or estimate its severity. CX Codex +1 304installs taint-trace The taint-trace Claude Code skill adjusts a project's static taint-analysis models, then runs its analyzer. It addresses missed data paths, unresolved framework calls, and false positives caused by imprecise models. CX Codex +1 130installs

Project sponsors

Companies that keep this analytics open