Skip to content
SkillModel-led399 tokens in use

hunt-regressions

The discovery stage of regression-hunt: read a pinned change and trace it through callers to list possible regressions for later proof.

8,319 stars on the repository CC-BY-SA-4.0trailofbits/skills · plugins/regression-hunt/skills/hunt-regressions
/plugin marketplace add trailofbits/skills
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing hunt-regressions costs

21tokens idle
399tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Model-led

Script-ledMixedModel-led

Mostly instructions, so how well it works depends on the model that follows them.

This describes the skill's structure, how much of it is scripts. It does not say how well it works.

Touches

Nothing detected: no shell, network, file writes, git or credentials in its files.

Declared tags come from the skill's frontmatter; detected ones from its scripts, its code blocks and the commands it names in inline code. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Good
  • Frontmatter12/15
  • Structure16/25
  • Examples3/15
  • Output12/15
  • Size10/15
  • Resources10/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Excellent

No findings.

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity100
  • Maintenance52
  • Community64
  • Maturity32
  • Momentum40

Hunt-regressions takes a sealed preparation, the frozen before and after sources, change context and a registry assignment, and looks for defects the change may have introduced. It reads commit messages, PR discussion and security documentation for intent and supported use.

The agent reviews the whole diff, traces changed behavior through callers to real interfaces, and records each lead with its changed code, previously supported behavior, trigger, proposed defect, false-positive checks and the proof needed. It cannot confirm findings, and an empty list is valid.

What You Get

  • Candidate files that each describe one lead and what proof it needs
  • A sealed hunt.json with IDs, file mappings, coverage and limitations
  • Leads that suggest old bugs or incomplete fixes kept for adjudication
  • A short report back with status, record path, candidate count and coverage gaps

Common Use Cases

  • Reviewing a pull request for behavior it may have broken
  • Tracing a refactor through its callers to supported interfaces
  • Recording partial coverage when a prerequisite blocks the review

What Makes It Unique

The hunter stays blind to later fixes and other reviews and must show a supported interface that reaches a suspect helper. Leads are reserved by a stable hypothesis key, so a retry reuses the same candidate IDs, and building the PoC is left to a separate reproduction stage.

Under The Hood

The agent reads SKILL.md, a regression definition and an evidence file contract, verifies the sealed preparation, and reads the frozen sources. It runs a bundled Python registry script from the plugin to reserve candidate IDs and writes candidate files and a sealed JSON record into the campaign directory. It does not build or run target code in this stage.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers