Guard metadata repair and doc sync scripts against symlink targets so
repo maintenance tasks cannot overwrite arbitrary local files.
Replace recursive skill discovery with an iterative walk that skips
symlinked directories, and harden the VideoDB listener to write only
private regular files in the user-owned state directory.
Also fix the broken pr:preflight script entry and make the last30days
skill stop embedding raw user arguments directly in the shell command.
Make the skill filter helper treat the complete bundle as a
pass-through so categories missing from the hardcoded map are
not silently omitted.
Add a regression test to keep complete bundle behavior aligned
with its name.
* feat: Add skill filtering utility to fix Gemini token truncation (Issue #215) Adds skill-filter.js module that provides: - Category-based skill filtering to reduce context overhead - Pre-configured skill bundles (minimal, balanced, complete) - Support for environments with token limits This addresses the "could not convert a single message before hitting truncation" error when all 1,200+ skills are loaded with Gemini. Users can now use the minimal bundle to avoid truncation: - Minimal bundle: ~5k tokens - Balanced bundle: ~15k tokens - Complete bundle: ~50k tokens Fixes#215
This module filters and manages skills by category to address token limit issues in Gemini, providing various skill bundles to optimize performance.
* Fix: Use catalog-backed categories for skill bundles Replaces hardcoded categories (core, architecture, etc) with actual categories present in skills_index.json to ensure getSkillsByBundle returns a usable subset instead of empty set.
Consolidate the repository into clearer apps, tools, and layered docs areas so contributors can navigate and maintain it more reliably. Align validation, metadata sync, and CI around the same canonical workflow to reduce drift across local checks and GitHub Actions.
