Skip to content

v15.9.0 "Security Boundaries and Multimodal Workflows"

Latest
Latest

Choose a tag to compare

View all tags
@sickn33 sickn33 released this 04 Aug 18:49
v15.9.0
6bfc62a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

[15.9.0] - 2026-08-04 - "Security Boundaries and Multimodal Workflows"

Hardened repository and runtime trust boundaries while adding governed Gemini media generation and evidence-first Shopify review triage.

This release gives Claude Code, Cursor, Codex CLI, Gemini CLI, Antigravity, and related AI coding assistants a safer 2,003-skill catalog. It closes the August Codex Security sweep, clears the remaining local dependency alerts, and adds two human-gated workflows for multimodal generation and public-review triage.

Start here:

Added

  • Added generate-nanobanana for routing image and video generation across Nano Banana and Gemini Omni models, with reference-image support, explicit video-cost approval, and prompt/seed sidecar logs.
  • Added shopify-review-triage for turning supplied public Shopify App Store reviews into a source-linked P0-P3 first-pass brief, with private-data stop conditions, human-verification labels, and no outbound messaging.

Changed

  • Updated unified-ai-gateway to the reviewed v0.4.0 release, documenting its ninth governed MCP tool and the pinned Linux arm64 image-review path while preserving provider-free, no-network, and separate-approval boundaries.
  • Regenerated the canonical catalog, offline AAS Core data, marketplaces, bundles, and Codex/Claude plugin distributions for 2,003 skills.

Security

  • Closed the August Codex Security sweep by making image review symlink-safe, treating project scripts and credentials as untrusted, pinning the GitHub attachment helper, confining Loki cleanup, loading PR-policy dependencies only from protected main, scoping oversized MCP requests to bounded metadata, and removing the public Star History token.
  • Pinned and bundled Ajv's fast-uri dependency on the patched 3.1.x line, resolving GHSA-7p8r-x3mc-p8w7 in the production dependency tree.
  • Updated the web app's PostCSS and transitive brace-expansion and ip-address tooling to patched releases, clearing the remaining local npm audit findings.
  • Refreshed the bundled loki-mode todo example lockfiles to patched ip-address and PostCSS releases, clearing the remaining Dependabot findings in the canonical skill and generated Claude mirror.

Who should care

  • Claude Code, Cursor, Codex CLI, Gemini CLI, and Antigravity users who install skills from the canonical or plugin distributions.
  • Security-conscious teams relying on bounded MCP framing, protected CI dependencies, credential handling, or container-image review guidance.
  • Creators using Gemini image or video generation who need explicit cost approval and reproducible prompt metadata.
  • Shopify app teams triaging low-star public reviews without treating unverified reports as confirmed incidents.

Validation

  • Resolved all 22 findings from the August Codex Security sweep: 19 verified fixes and 3 documented false positives; the refreshed open-findings view is empty.
  • Passed canonical skill and reference validation, documentation security, warning-budget enforcement, the complete 106-group repository suite, the 150-test AAS Core suite, web build and prerender, and package dry-run checks.
  • Confirmed root and web-app npm audits report zero vulnerabilities and protected main CI and CodeQL are green before release preparation.

Credits

Assets 2
Loading