AI readiness toolkit

v1.6.3 Breaking

This release includes 1 breaking change for platform teams planning a safe upgrade.

Published 10d Developer Productivity

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Summary

AI summary

Layer 0 renamed to Agent Instructions and /huddle now bundles hat methodologies correctly.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Low	Adds `bundle_files` config field to copy arbitrary repo files into a skill's standalone ZIP, with YAML frontmatter stripping for markdown bundles. Adds `bundle_files` config field to copy arbitrary repo files into a skill's standalone ZIP, with YAML frontmatter stripping for markdown bundles. Source: llm_adapter@2026-05-25 Confidence: high	—
Feature	Low	Adds path‑traversal guard on `bundle_files` destinations to prevent unsafe file writes. Adds path‑traversal guard on `bundle_files` destinations to prevent unsafe file writes. Source: llm_adapter@2026-05-25 Confidence: high	—
Feature	Low	Adds test `test_huddle_plugin_source_retains_team_mode_infrastructure` to enforce that plugin installs keep team mode while standalone strips it. Adds test `test_huddle_plugin_source_retains_team_mode_infrastructure` to enforce that plugin installs keep team mode while standalone strips it. Source: llm_adapter@2026-05-25 Confidence: high	—
Bugfix	Medium	/huddle now works correctly in standalone installs by bundling hat methodologies and using deterministic mode picker. /huddle now works correctly in standalone installs by bundling hat methodologies and using deterministic mode picker. Source: llm_adapter@2026-05-25 Confidence: high	—
Bugfix	Low	Improves `strip_frontmatter` to tolerate CRLF line endings and EOF‑closed fences in markdown files. Improves `strip_frontmatter` to tolerate CRLF line endings and EOF‑closed fences in markdown files. Source: llm_adapter@2026-05-25 Confidence: high	—
Refactor	Low	Renames Layer 0 from "Breadcrumbs" to "Agent Instructions" in the /assess narrative documentation. Renames Layer 0 from "Breadcrumbs" to "Agent Instructions" in the /assess narrative documentation. Source: granite4.1:30b@2026-05-25-audit Confidence: low	—

Full changelog

Highlight: /huddle now works correctly in standalone installs (Claude Desktop, claude.ai web, Cowork). Hat methodologies are bundled inside the ZIP so phased sub-agent mode no longer silently degrades to "improvise the hat." Driven by real-use feedback from running /huddle end-to-end in Cowork - thanks to Claude Desktop for the substantive retro that surfaced three defects and pushed back on my initial triage twice.

README also got a narrative pass for /assess: previously dry, now explains why a layered contract model matters for an AI-driven codebase and shows the 0-7 score table inline. Layer 0 renamed from "Breadcrumbs" to "Agent Instructions" to match what the ecosystem actually calls these files (CLAUDE.md, AGENTS.md, copilot-instructions, Cursor rules). And the standalone-install section now has a verified UI path (claude.ai/customize/skills) with the real prerequisites.

What's Changed

Fixes

fix(huddle): bundle hat methodologies, deterministic mode picker by @bjcoombs in https://github.com/bjcoombs/ai-native-toolkit/pull/28

Documentation

docs: clarify two install paths and verify standalone ZIP steps by @bjcoombs in https://github.com/bjcoombs/ai-native-toolkit/pull/26
docs: rewrite /assess narrative + rename Layer 0 to Agent Instructions by @bjcoombs in https://github.com/bjcoombs/ai-native-toolkit/pull/27

Pipeline additions

New bundle_files config field copies arbitrary repo files into a skill's standalone ZIP, with YAML frontmatter stripping for markdown bundles. Used by huddle to ship the 6 hat methodologies under hats/. Source of truth stays in agents/ - no drift.
Path-traversal guard on bundle_files destinations.
strip_frontmatter now tolerates CRLF line endings and EOF-closed fences.
New test test_huddle_plugin_source_retains_team_mode_infrastructure pairs with the existing standalone forbidden-string tests to fence the invariant: plugin install keeps team mode, standalone strips it. A PR that accidentally removes team mode from the source will fail loudly.

Full Changelog: https://github.com/bjcoombs/ai-native-toolkit/compare/v1.6.1...v1.6.3

Breaking Changes

Layer 0 renamed from "Breadcrumbs" to "Agent Instructions" in README and related guidance.

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track AI readiness toolkit

Get notified when new releases ship.

About AI readiness toolkit

All releases →