First Aggregator Feedstock — 23 Changelogs, 222 Entries, One Data File

Why Care?

The Lossless Group has had a long-running goal: aggregate every changelog across every Lossless repo into one umbrella feed via the GitHub API. Until tonight, the aggregator’s feedstock — a list of where changelog content actually lives, paired with each upstream remote URL — didn’t exist as a single artifact. Now it does, at site/data_site/changelog-locations-index.md.

For the audience: this is the moment Changelog First Development gets a multiplier. When the aggregator ships, every shipping signal across the family becomes visible in one place. Cadence becomes pulse. Pulse becomes signal. Signal becomes trust.

For the team: this is also the first real test of the search-first discipline the pseudomonorepos skill teaches. The discipline says walk the tree, find prior work, surface findings, link rather than duplicate. Tonight: literal find. Same instinct.

What’s New?

A new data file in the lossless-site repo:

site/data_site/changelog-locations-index.md

Frontmatter holds the structured data for machines:

• 23 changelog locations discovered across the lossless-monorepo tree
• 222 total entries (Markdown files) across them
• 18 unique upstream remotes (one external — mem0, vendored for study)
• Drift surfaced and noted: plural changelogs/, legacy context-v/changelog/ nesting, dirname↔remote mismatches, one suspicious git remote URL

Body holds the human-readable summary, a leaderboard, and a regeneration script.

The file is on disk but not committed — lossless-site is the original beast and was deliberately not poked tonight.

The Story

The session had been deep in skill-building all evening. The pivot to the index file felt like a natural rest stop: a concrete artifact that exercises the conventions we’d just shipped.

flowchart LR
    A["find: 41 dirs<br/>called changelog/<br/>or changelogs/"]
    B["filter: cut<br/>build artifacts,<br/>routes, layouts"]
    C["23 real<br/>entry locations"]
    D["git remote per dir<br/>(submodules<br/>auto-resolve)"]
    E["count entries,<br/>note drift,<br/>categorize"]
    F["one .md file,<br/>frontmatter<br/>= the data"]

    A --> B --> C --> D --> E --> F

The trick was what to filter. The naive find -name changelog returned 41 paths. Most of them weren’t entry sources at all:

• .vercel/output/...changelog — build artifacts, rendered HTML
• src/pages/changelog — Astro routes (index.astro, [slug].astro), not entries
• src/layouts/changelog, src/components/changelog — code, not content
• public/changelog — static assets

Filtering those out left 23 directories that actually hold entries. The filter list became part of the file’s frontmatter (filters_applied.excluded_paths) so the regeneration is reproducible.

The other small move was honoring the drift policy the way it was designed to be honored. Tonight’s tree walk surfaced a lot of inconsistency: plural-vs-singular folder names, legacy nesting under context-v/, local directory names that don’t match their upstream repo names (banner-site → emblem-site, perplexed → perplexed-plugin), and one git remote URL that looks like an honest mistake (changelog_matter-site.git). The temptation was to fix some of it on the way past. The discipline — and the reason the policy exists — was to observe, note, surface, but not auto-fix. Every drift item went into drift_note fields in the index, and a “Drift, Surfaced Not Fixed” section in the body lists the cleanup candidates for whenever a future authorized pass happens.

How It Works

The index is generated, not handwritten. The full reproduction recipe is in the data file itself, but the gist:

ROOT="/Users/mpstaton/code/lossless-monorepo"

find "$ROOT" -type d \( -name changelog -o -name changelogs \) \
  -not -path '*/node_modules/*' \
  -not -path '*/.git/*' \
  # ... (full exclusions in the file)
  -not -path '*/src/pages/*' \
  -not -path '*/src/layouts/*' \
  | while read -r dir; do
    rel="${dir#$ROOT/}"
    remote=$(git -C "$dir" remote get-url origin 2>/dev/null)
    count=$(find "$dir" -maxdepth 1 -name '*.md' -type f | wc -l)
    # build YAML entry
done

git -C "$dir" remote get-url origin is the load-bearing line — it walks up from the changelog directory to find the nearest .git, which means it correctly resolves submodule remotes too. So astro-knots/sites/banner-site/changelog returns emblem-site.git (the actual upstream of that submodule), not the astro-knots parent’s remote.

The output went through a quick three-way validation:

• ✓ YAML frontmatter parses cleanly
• ✓ Sum of declared entry_count values matches the live count: 222
• ✓ Number of unique remotes matches the declared count: 18

(One off-by-one caught on the unique-remotes count and corrected. Validation prevented shipping wrong data.)

Top of the Tree

What’s Next

• The file lives on disk but is uncommitted in lossless-site. Tomorrow’s first move is reviewing it before merging — lossless-site is a deliberate beast.
• The index is the feedstock; the aggregator is the next build. That’s a separate project — fetch each remote’s changelog/ (or context-v/changelogs/, etc.) via the GitHub API, normalize, render as one feed.
• Drift cleanup is on the docket as an explicitly-authorized future task. The candidates are pre-sorted in the index file’s body for whenever it gets prioritized.
• A regeneration skill or extension could turn the bash script into a one-command refresh — likely a future chapter of lossless-loop or a small dedicated skill.

Related

• The data file: lossless-monorepo/site/data_site/changelog-locations-index.md (uncommitted)
• changelog-conventions/references/changelog-first-development.md — the methodology this serves
• pseudomonorepos/references/search-first.md — the discipline this exercises
• The drift policy: ~/.pi/agent/AGENTS.md
• Previous entry: Lossless Skills v0.0.0.1