Changelog
What shipped, when, and why
Entry-by-entry notes for lossless-agent-skills, following the
Lossless changelog conventions.
-
Three Skills Shipped, Two Date Families Untangled, and a Repo That Can Keep a Secret
A skill library that publishes everything needs somewhere to put the things it can't. This session built that boundary — gitignored config read by instruction, an audit grep that has to come back empty, and a rule against injecting our conventions into repos we only have READ on — then shipped three skills through it and finally wrote down the frontmatter standard the tree had been running on tacitly for months.
-
Public skills, private config — and the first skill to prove it
Some skills only work if they know who they're working for, which is exactly what a public repo can't hold. The answer is a gitignored config/private.json read at runtime by instruction rather than by templating, an example file that serves as the contract, and an audit grep that has to come back empty before you push. The CRM interface guidelines are the first skill built to that shape.
-
context-vigilance grows from six folders to eight — plus a universal tier and an experimental tier
plans/ and agent-skills/ earned canon the honest way: fourteen and five repos adopted them without anyone declaring them official. The skill now names them, blesses extra/ and sitemap/ as universal utilities (extra/ gitignored by default), and stakes out loops/, handoffs/, decisions/, and contracts/ as the experimental tier still finding its shape.
-
A new SurrealDB verification skill, a lede-length discipline propagated across two skills, and sync-skills-symlinks preps for an agent-skills migration
A skill for querying a SurrealDB canonical layer directly instead of writing a throwaway Node script per check. Plus: the lede field was drifting toward mini-essays, so changelog-conventions and context-vigilance both got a 'subtitle-length, not a paragraph' rule with the same wording. Plus: sync-skills-symlinks.sh now recognizes a context-v/agent-skills/ layout alongside the legacy context-v/skills/ one, ready for a migration that hasn't moved any actual skill directories yet.
-
New skill (gh-cli-projects-tasks-conventions), new credibility-ingest routine, and three skill updates from real-world incidents
A premature-on-purpose new skill for gh project task creation — codifying just two conventions (compose with pseudomonorepos to find the right repo, link to context-v files via clickable GitHub URLs) so they don't drift while the rest emerge. Plus a company-anchored credibility-ingest routine that inverts the crawl-fetch-ingest cascade so a fundraising deck can render its backers legibly. Plus three updates born from real incidents: an Astro Knots prerender-vs-auth rule (calmstorm-decks shipped a silent production bypass before we caught it), an audience-cascade cross-reference in git-conventions, and a full rewrite of the content-rollup reference now that two production splashes have implemented it.
-
Reframe the audience-cascade section in changelog-conventions — marketing up top, engineering as the narrative unfolds
The earlier framing read as 'marketing all the way down,' which fought the four-audience cascade the same section then described. Reframed to say what the cascade actually does: the lede / Why Care? / What's New? are marketing-shaped, the How it Works / Under the Hood / Files Touched are engineering-shaped, and both belong — what differs is sequence.
-
Scaffold the lossless-agent-skills splash page
Repo gets its own GitHub Pages splash — operator/CLI posture, skills catalog as the primary content surface, Pagefind search, llms.txt for LLM ingest, sitemap + robots for search engines.
-
Three New Skills — Crawl-Fetch-Ingest, Maintain DESIGN.md, Generate Consistent OG Images
Codified three workflows the team had been doing freehand: VC firm metadata crawls (crawl-fetch-ingest), agent-readable project visual-identity contracts following Google Stitch's open spec (maintain-design-md), and consistent share-imagery generation via the Ideogram v3 API with WhatsApp/iMessage as the primary target (generate-consistent-og-images). The OG-images skill picked up an empirical refinement this session — empty space has to be declared as a first-class rendered subject, not left as residue — that pushed the BannerTall composition from a chronic 85% subject-fill to a clean 40-65%.
-
Two New Skills — Maintain Splash Pages, Chroma Agent Skills (Submodule)
Added the Maintain Splash Pages skill — codifying the <repo>/splash/ GitHub Pages pattern proven across three reference implementations — and pinned chroma-core/agent-skills as a submodule so agents loading the Lossless skills tree get upstream-authoritative Chroma guidance instead of paraphrased-from-training-data vector-store advice.
-
First Aggregator Feedstock — 23 Changelogs, 222 Entries, One Data File
Walked the lossless-monorepo tree, found 23 changelog directories holding 222 entries across 18 upstream repos, and wrote the first machine-readable feedstock for the future Lossless Changelog umbrella aggregator — without waking the lossless-site beast that hosts it.
-
Lossless Skills v0.0.0.1: From a Question to a Repo
We sat down to ask 'how is Pi different from Claude Code?' and ended the session with a public Lossless org repo of four interlocking Agent Skills, a global agent config, and the first principles of Changelog First Development.