v0.8.0 — Triggers frontmatter, language-agnostic code-map, churn-stable doc-sync

[mvoutov]

v0.8.0 — Triggers frontmatter, language-agnostic code-map, churn-stable doc-sync

This is a substantial release. Three things are worth knowing about up front, then a longer list below.

Headline changes

1. Skill triggers moved to YAML frontmatter. Activation rules (file globs, keywords, alwaysActivate) now live in skill frontmatter under a triggers: key instead of a free-form ## Activation section. The activation hook reads frontmatter first; existing ## Activation sections still work via a back-compat fallback so you don't need to migrate by hand. New skills generated by aspens doc init / doc sync emit the frontmatter form.

2. code-map.md is now language-agnostic and churn-stable. Previously this file shuffled on every sync (file counts, edge counts, hub rankings, hotspot churn numbers, "+N more" suffixes — all changing constantly). v0.8.0 removes the noise and restores Domain clusters (cluster name + top files) for any language, not just Next.js repos. Sync diffs stop being dominated by code-map churn.

3. AGENTS.md no longer drops unchanged skills on partial doc sync. When only one skill changed, the codex transform was rebuilding AGENTS.md's Skills section from the in-flight subset and silently dropping every other skill. Fixed.

Upgrade

npm i -g aspens@latest
aspens doc sync .   # regenerates skills with the new triggers format

Existing ## Activation sections continue to work — no forced migration.

What else is in here

• Next.js entry-point detection — App Router (page, layout, route, loading, error, not-found, template, default, global-error), Pages Router, and special top-level files (middleware, instrumentation) surface as Framework entry points in code-map.md.
• Python and TypeScript import parsers — dedicated modules under src/lib/parsers/ for richer extraction (re-exports, barrels), plus path-resolver.js for tsconfig.json alias resolution with extends chain support.
• isNoOpDiff() classifier — doc sync skips the LLM call entirely on lockfile-only or non-code-bearing diffs.
• sanitizePublishedContent() chokepoint — every disk write goes through it; strips forbidden blocks from skill files while preserving them in code-map.md.
• Multi-target parity validator — guards against publishing divergent file sets between Claude and Codex.
• syncSkillsSection is now idempotent — fresh inserts no longer push prose underneath the Skills heading where the next sync would eat it.
• .cjs / .mjs extension parity — full graph coverage for CommonJS-heavy and ESM-explicit repos.
• Path-traversal guard in resolveAliasImport() — a crafted tsconfig.json with paths: { "@/*": ["../../../outside/*"] } no longer creates out-of-repo graph nodes.

Plus a handful of smaller fixes: CLAUDE.md-mentioning lines no longer silently dropped from AGENTS.md, Pages Router walker no longer recurses into .next/ / .vercel/, broken symlinks in .claude/ no longer crash aspens customize agents, and the no-op sync path runs scanRepo once instead of twice.

Test suite grew from ~360 to 439.

Full changelog: https://github.com/aspenkit/aspens/blob/main/CHANGELOG.md
Release: https://github.com/aspenkit/aspens/releases/tag/v0.8.0