Skip to content

docs(security): add GitHub Actions security audit methodology#22

Merged
smartwatermelon merged 1 commit into
mainfrom
claude/security-audit-methodology-doc-20260429
Apr 29, 2026
Merged

docs(security): add GitHub Actions security audit methodology#22
smartwatermelon merged 1 commit into
mainfrom
claude/security-audit-methodology-doc-20260429

Conversation

@smartwatermelon
Copy link
Copy Markdown
Owner

Summary

Documents the methodology used in the 2026-04-29 inaugural GitHub Actions security audit (the manual run that produced 9 PRs across the smartwatermelon/* and nightowlstudiollc/* repos) and codifies it for re-use by:

  1. Future ad-hoc human audits.
  2. The recurring quarterly remote agent (routine trig_01JaKYSFQhPJoc3jADyQPBgM), which fires next on 2026-07-01T17:00Z and reads this doc as its source of truth.

Contents

  • Source: link to the Nesbitt article that catalogs the nine attack patterns
  • In-scope filter: gh repo list commands with the isArchived / isFork gates
  • The 9 vulnerability patterns with detection commands + remediation guidance for each
  • Severity rubric (blast radius × exploitability)
  • Mitigation framework (Tier 1/2/3 by cost-to-fix and exposure — the same triage I used in April)
  • Tooling pointers (zizmor, pinact, ratchet)
  • Recurring agent reference (routine ID, cadence, deliverables)
  • Audit history table with the 2026-04-29 row pre-filled

Recurring-agent contract

The agent's prompt is hard-coded to read this doc. Methodology changes propagate automatically on the next run. The agent is instructed NOT to modify this doc directly — if it discovers something the methodology should cover, it files an issue labeled audit-methodology-update and a human (you) decides what to merge.

Test plan

  • Doc renders cleanly on GitHub
  • Manual smoke-test: re-run the inaugural audit's enumeration commands from the in-scope filter section, confirm the 29-repo count
  • First quarterly run on 2026-07-01 produces an issue and follows the structure described

🤖 Generated with Claude Code

Documents the methodology used in the 2026-04-29 inaugural audit and
intended for use by both ad-hoc human audits and the recurring quarterly
remote agent (routine trig_01JaKYSFQhPJoc3jADyQPBgM).

Contents:
- Source: link to nesbitt.io 2026-04-28 article
- In-scope filter: gh repo list with isArchived/isFork gates
- The 9 vulnerability patterns: detection commands + remediation guidance
  for each pattern catalogued in the article
- Severity rubric (blast radius × exploitability)
- Mitigation framework (Tier 1/2/3 by cost-to-fix and exposure)
- Tooling pointers (zizmor, pinact, ratchet)
- Recurring audit reference (routine ID, cadence, deliverables)
- Audit history table (2026-04-29 row pre-filled)

The recurring agent's prompt instructs it to read this doc as the
canonical methodology, so changes here propagate automatically. The
agent is instructed NOT to modify this doc itself; methodology drift
proposals from the agent come back as issues labeled
"audit-methodology-update".

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@smartwatermelon smartwatermelon merged commit 6557d76 into main Apr 29, 2026
2 checks passed
@smartwatermelon smartwatermelon deleted the claude/security-audit-methodology-doc-20260429 branch April 29, 2026 22:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant