Changelog

View raw markdownEdit on GitHub ↗

User-facing changes to the repo, the site, and the advisory corpus. Per-sweep details live in .claude/skills/vibe-security-update/runs.log.md.

The format is loosely based on Keep a Changelog. Newest entries first.

[Unreleased]

Added

  • LLM-friendly outputs (Mintlify / Anthropic pattern):
  • Per-page .md mirrors alongside every .html page — replace .html with .md in any URL to get the raw markdown source without re-parsing HTML.
  • Per-section llms.txt at advisories/llms.txt, playbooks/llms.txt, prevention/llms.txt, sources/llms.txt, tools/llms.txt.
  • Compact llms-ctx.txt — alerts + per-advisory TL;DRs + "am I affected?" only (compact per-advisory digest, ~70KB at ~50 advisories) for mid-context-window ingestion.
  • feed.xml — Atom feed of advisories for RSS readers.
  • advisory-schema.json — JSON Schema (draft 2020-12) for advisory frontmatter.
  • Versioned API at api/v1/advisories.json and api/v1/index.json.
  • JSON-LD ItemList schema on the advisories index page.
  • Sitemap hints — changefreq + priority per URL.
  • Page actions — "View raw markdown" + "Edit on GitHub" on every page.
  • <link rel="alternate" type="text/markdown"> on every page so tools can discover the markdown variant.
  • CITATION.cff — academic citation metadata.
  • SECURITY.md (renamed from lowercase security.md for the standard GitHub convention).
  • CHANGELOG.md — this file.
  • Comprehensive pytest test suite under tests/ — frontmatter schema, link integrity, no-secrets, determinism, accessibility, llms.txt format, advisory ID uniqueness, etc. Wired into the deploy workflow.
  • Prevention guides:
  • prevention/supply-chain-attack-surface.md — a map of every pathway by which external code/data reaches you (package managers, lockfiles, CI/GitHub Actions, browser CDN, MCP & agent tooling, editor/browser extensions, scraped data, datasets, global installs, hosted builds), each with the one defense that matters and links to the deep guide.
  • prevention/ci-cd-hardening.md — GitHub Actions / CI supply-chain hardening: SHA-pin actions, least-privilege permissions:, the third-party-action-with-write-token anti-pattern, dangerous triggers, script injection, and OIDC publishing.

Changed

  • Site title displayed as "Vibe Coding · Security Issue Tracking" (was "vibe-coding-security"). Repo URL slug unchanged.
  • Footer now exposes the full set of LLM/feed format links.
  • robots.txt explicit comment: AI/LLM training allowed.

Security

  • Hardened the deploy workflow (.github/workflows/deploy-site.yml): pinned all GitHub Actions to full commit SHAs (were floating @vN tags) and scoped permissions: to least privilege per job — build keeps only contents: read, the elevated pages: write + id-token: write now live solely on the deploy job. The project now practices the CI hardening documented in prevention/ci-cd-hardening.md.

2026-05-17

Added

  • 3 new advisories from the 2026-05-17 sweep:
  • Windsurf zero-click MCP RCE (CVE-2026-30615)
  • "Comment and Control" PR prompt injection (CVSS 9.4) — Claude Code Security Review, Gemini CLI Action, Copilot Agent
  • Systemic MCP stdio RCE class — OX Security, ~200k servers exposed
  • Authoritative-source citations added to every prevention doc and playbook (OWASP, Anthropic, npm Docs, AWS Customer Playbook Framework, GitHub Docs, MCP spec, OpenSSF, CISA, Karpathy).

Changed

  • node-ipc advisory: corrected to 822K weekly downloads (was 10M+); added specific bad versions (9.1.6, 9.2.3, 12.0.1), DNS C2 IOCs, shasum, forensic markers.
  • TanStack Mini Shai-Hulud advisory: broadened to full 172-package / 403-version wave including Mistral, UiPath, OpenSearch namespaces; attributed to TeamPCP; noted first valid SLSA-provenance abuse.
  • Claude Code InversePrompt advisory: added 4 more CVEs (CVE-2025-52882, CVE-2025-59536, CVE-2026-21852, CVE-2026-33068) + TrustFall reference.

2026-05-16

Added

  • Initial repo: 13 advisories backfilled, 6 playbooks, 5 prevention docs, sources + tools indexes.
  • vibe-security-update skill at .claude/skills/vibe-security-update/ with tiered sweep procedure (deep 24h / medium 3d / shallow 7d) and self-learning source-priority list.
  • Static site generator (site/build.py), template, CSS, validator.
  • GitHub Actions deploy workflow.
  • Initial 78-source priority list seeded from first-run findings.