Agent Skills · dididecks-ai

This directory is the shared snapshot of the agent skills that load during work on dididecks-ai. Collaborators each maintain their own copy of these skills wired into whatever coding agent they use (Claude Code at ~/.claude/skills/, Pi at ~/.pi/skills/, Cursor’s rules, etc.). This directory exists so that snapshot is visible — checked into the repo, reviewable in PRs, and a reliable point of comparison when an agent’s behavior drifts.

How to use this directory

Two complementary modes, depending on which way the wind is blowing:

Keep your local skills current with this directory. Periodically diff your ~/.claude/skills/<skill-name>/ (or your equivalent) against the matching context-v/agent-skills/<skill-name>/ here. If the version here is newer or different, refresh your local copy. This is how a collaborator joining the repo gets to the same baseline the rest of the team works against.
If you iterate on or author a new skill, push it back to this directory. Edit your local copy, validate it works in your agent, then copy the updated SKILL.md (and any helper files) back into context-v/agent-skills/<skill-name>/ here and commit. Even if your local management system is the source of truth for you, the repo copy is the source of truth for the team.

Each collaborator will have their own system for managing agent-skills on their machine. Collaborators should make sure their skills are current with dididecks-ai’s agent-skills in context-v. Alternately, if they iterate on or create agent-skills, they should update them in the dididecks-ai repo as best they can.

What’s in here (as of authoring)

Sixteen skills, picked for their direct load-bearing relevance to dididecks-ai work. The set divides roughly:

Tree + process discipline (always relevant)

Skill	What it does
`pseudomonorepos`	Parent-of-children repos, branch-tier alignment, HARD STOP rules for relocation, content roll-up
`context-vigilance`	`context-v/` directory roles, frontmatter, version-stamping, four-mode (prep/reflective/journey/spec) discipline
`astro-knots`	Astro convention family — framework rules, approved deps, hard prohibitions (React/JSX/Angular)
`changelog-conventions`	How `changelog/` entries are structured, frontmatter discipline, four-audience cascade
`git-conventions`	Commit message format (action verbs, effort groupings, paragraph-spaced bodies, “Also included” riders)
`study-repos-first`	The “pin upstream code before designing” discipline for studies-driven decisions

Deck-building (domain-specific to dididecks-ai)

Skill	What it does
`setup-new-dddecks-workspace`	Stub. Scaffold a new `client-sites/<slug>/` end-to-end — submodule, brand import (website-parse OR source-code-copy), DESIGN.md, theme tokens, auth surface, corpus connection, starter Scroll-UI deck. Composes the other skills; first run is The Lossless Group as its own client.
`deck-iteration-workflow`	The phased workflow: single-page narrative → individual slides → componentization → features → full deck → advanced interactivity
`slide-target`	Load the full working context for ONE slide (deck/variant/slot) — section file, slot registry, narrative slot, rank status, live URLs, design tokens — and scope work to it. First of the `slide-*` family (target → improve → rank → decompose); the agent-side “context broker” for slide-by-slide iteration.
`maintain-design-md`	How a per-client `DESIGN.md` is authored (Google Stitch spec + Lossless extensions); the three-mode invariant
`theme-system`	Two-tier token architecture (named `__` + semantic kebab), three-mode contract (light/dark/vibrant), `--fx-*` effect dials

Asset + data ingest

Skill	What it does
`crawl-fetch-ingest`	Crawl a firm’s site / a PDF; ingest people + portfolio metadata as canonical `.md` files; four-checkpoint cascade with confidence flagging
`generate-consistent-og-images`	Ideogram v3 recipe for share imagery; pulls locked tokens from `DESIGN.md`; WhatsApp/iMessage as primary unfurl surface
`overlay-svg-text`	Brand-typed text overlays (Hack Bold gradient h1, thin sans eyebrow, Poor Story handwriting) on top of generated imagery

Splash + content

Skill	What it does
`maintain-splash-pages`	The repo-level `splash/` convention — GitHub Pages target, changelog rollup, marketing-up-top discipline
`lossless-flavored-markdown`	LFM directives (callouts, embeds, citations) and how sites consume them via the `@lossless-group/lfm` package
`open-graph-share-seo-geo`	Share-preview engineering: OG metadata, sitemap.xml, robots.txt, `llms.txt` standard for GEO

Search

Skill	What it does
`search-lossless-corpus`	The four-step agentic-search loop against the local Chroma RAG (collections: `context-vigilance-corpus`, `lossless-changelog`, `claude-code-sessions`, `claude-code-tool-traces`)

Skills NOT included here

Some Lossless skills exist that are deliberately not snapshotted into this repo, either because they’re meta (about managing the source-of-truth skills repo itself) or because they’re specific to a different vertical:

chroma-agent-skills — submodule covering chroma-DB-specific agent tooling, not the dididecks chroma-decks client
AUTHORING.md / CANDIDATES.md — internal to the upstream lossless-skills repo
sync-skills.sh — the upstream’s symlink-management script for getting skills into ~/.claude/skills/ and ~/.pi/skills/

If you find a skill that’s not here and that you wish was here for a reason load-bearing to dididecks-ai work, copy it in and update the table above as part of the same commit.

Skills authoring discipline (one-paragraph version)

A skill is a directory: <skill-name>/SKILL.md is the entry point, optional references/, templates/, routines/, scripts/ sit alongside. The SKILL.md frontmatter has at minimum name: and description: — the description is what an agent reads to decide whether to load the skill, so write it to be triggering: name the verbs, contexts, and user-utterance patterns that should fire it. The body is the discipline itself in prose, written for an agent (terse, code-shaped, table-friendly). See the upstream AUTHORING.md for the full guidance.

dididecks-ai/agent-skills/readme