Files
Michael D'Angelo 775231e9bf feat(scanner-imports): retain validation evidence for individual Continuum findings (#2625)
Retain validation evidence associated with the selected finding during collection, import, and publication.
2026-11-17 14:28:23 -08:00
..

Codex Security skills

With the plugin enabled, you can ask Codex to "check my uncommitted changes for security bugs." To pick the skill yourself, include its name in the request, as in $security-diff-scan. Setup and CLI examples

Choosing a scan

  • security-scan audits a repo or folder in one pass. Start here for a general review.
  • security-diff-scan looks for vulnerabilities introduced or exposed by a PR, commit, branch diff, or local edits.
  • deep-security-scan makes several passes over the same code and combines the findings in one report.

Before you merge or install something

  • assess-patch-risk takes a patch file, PR diff, or commit range. It explains what could break, whether the tests would catch it, and whether the patch is ready to merge.
  • check-skills reviews a downloaded agent skill before you install it. Open Codex in a separate, trusted workspace and point it to the skill folder. It reads the instructions and scripts without running them.

Security policy and threat models

  • define-security-policy writes or updates SECURITY.md in the repo root or a component folder. Use it to set the rules for future reviews, including what counts as a finding and what's out of scope.
  • threat-model maps what the current code protects and where an attacker could get in. Look for threatmodel.md in the scan output directory, or in the plugin's artifact storage for that repo if you run the skill on its own.
  • threat-model-visualize turns an existing model into a diagram you can explore.

Investigating a finding

  • triage-finding checks a reported bug against your code and security policy. Give it the alert or ticket.
  • resolve-dependency-alert traces what pulled a flagged package into your checkout or image and checks your upgrade options.
  • finding-discovery looks for possible vulnerabilities in a path or diff and flags what still needs proof.
  • validation checks whether a suspected bug is real by tracing the code or trying to reproduce it.
  • attack-path-analysis traces how an attacker could trigger the bug and what they'd gain.

Fixing a bug

  • fix-finding patches a vulnerability and checks that the fix stops the attack and preserves normal use.
  • verify-fix checks an existing fix. Give it the original finding and the current checkout to see whether the fix worked or what evidence is missing.
  • propose-security-hardening compares design changes that could keep the same kind of bug from turning up again.

Reports and tickets

  • vulnerability-writeup turns notes or a proof of concept into a report explaining the bug, its impact, and the evidence.
  • publish-findings previews alerts from another scanner and imports them into the local workbench. Some formats also support import into Codex Security Cloud.
  • cloud-findings looks up a finding, report, or scan status in your connected Cloud account. You can paste a finding link.
  • track-findings takes validated findings from a completed scan directory and checks for existing tickets. You review any new Linear, Jira, or GitHub issues before it files them. It can also draft a GitHub security advisory.