ABVX Lab

June 29, 2026 · View on GitHub

A static hub for the ABVX stack: the repo and workflow layer around AI-assisted coding. It helps turn a GitHub repo into an AI-ready workspace with repo docs, agent surfaces, repo maps, workflows, and checks. ABVX is not another in-editor autocomplete tool; it is the grounding layer around Copilot, Cursor, Continue, and API-driven agents.

Live: lab.abvx.xyz

About the builder: lab.abvx.xyz/about/

Who this is for

  • Solo developers who want one repo setup path that quickly produces AGENTS.md, llms.txt, docs, maps, and checks.
  • Teams who want multi-repo workflow visibility, proof loops, registry baselines, and CI-driven repo maintenance.
  • Tool builders who want portable human context and reusable agent capabilities that can plug into existing workflows.

Start here

  • SET for one CI entrypoint that keeps repo AI surfaces up to date.
  • agentsgen for AGENTS.md, llms.txt, docs/ai, bundles, and checks.
  • ID when human and operator context must travel across repos and tools.
  • Agent Learning Layer for the ABVX position on what actually learns in an agent: model weights rarely change; practical learning usually lives in context, skills, scripts, gates, and evals.

The live site uses the alt-b production shell: SET is the orchestration entrypoint, ID is the portable profile-and-hook layer, LWP is the lightweight desktop execution protocol, the home page opens with a tracked-repos snapshot plus a quieter supporting-tools directory, and tool pages share the same product-sheet layout.

Home page structure

  • SET, agentsgen, ID, repomap, LWP, and ABVX Agent Skills are featured first as the visible core stack.
  • The control plane is summarized as a tracked-repos ledger with queue state, then followed by a quieter directory of supporting tools and secondary surfaces.
  • Featured tiles and tracked repo rows are clickable and route to internal tool pages or GitHub where no first-class internal surface exists.
  • Detailed read-only surfaces still live below that summary:
  • Tools are grouped into product families instead of a flat card grid.

Cleanup status

  • homebrew-core has been archived after the upstream Homebrew PR was closed.
  • mn7r-showcase, revenue-os, Orbitory, and the old profile repo markoblogo have been removed after review.
  • Temporary upstream-contribution forks are still kept while their PRs remain open.
  • The current portfolio classification lives in docs/repos/repo-matrix.md.

Portfolio groups

  • AI coding tools stack with lab.abvx as the public hub, backed by AGENTS.md_generator, SET, ID, abvx-agent-skills, and homebrew-tap.
  • toki pona / sitelen with sitelen-layer-plugin as the package hub, backed by sitelen-emoji-truth and related language/content repos.
  • commodity systems with index as the public infra hub and mn7r as the internal operating workspace.

Tool groups

Orchestration

  • SET — Thin GitHub Action entrypoint for presets, repo-docs, site-ai flows, registry-driven review, and proof-loop orchestration.

Repo docs & agent context

  • agentsgen — Safe repo docs toolchain for coding agents, now with versioned CLI/MCP contracts and opt-in LLM enhancement.
  • ID — Portable human-AI profile protocol plus repo-local hooks for SET-compatible orchestration flows.
  • LWP — Lightweight Workflow Protocol for desktop-first agent development.
  • ABVX Agent Skills — Portable, validation-gated skills for coding, research, frontend, audits, debugging, token economy, repo onboarding, durable context maintenance, loop engineering, grilling, PRDs, issue slicing, triage, handoffs, browser verification, and standalone HTML artifact delivery, installable from PyPI as abvx-agent-skills or from the ABVX Homebrew tap.
  • Agent Learning Layer — Explainer for routing agent learning into model, context/memory/skills, or harness/workflow layers without pretending every improvement trains the model.
  • Goal Loop Designer — ABVX Agent Skills capability for compiling raw /goal prompts into bounded loop harnesses with rubrics, judge prompts, budgets, YAML, JSON, and Mermaid.
  • Local Inference Tuning — ABVX Agent Skills capability for selecting and tuning local LLM engines, cache policy, KV cache, batching, smoke benchmarks, and OpenAI-compatible endpoints.
  • agentsgen init — Bootstrap .agentsgen.json + AGENTS/RUNBOOK marker sections.
  • agentsgen update — Patch managed marker sections only.
  • agentsgen pack — Generate AI docs bundle with repo/site mode, contract-backed drift checks, and machine-readable manifests.
  • agentsgen snippets — Canonical README snippet extraction with deterministic drift checks.
  • agentsgen presets — Copy-paste setup for common stacks.

Validation & CI

Analysis & LLMO

  • repomap — Token-budgeted repo map + import graph artifacts with relevance ranking and slice modes.
  • agentsgen analyze — Shipped CLI surface for AI-visibility scoring of a public URL.
  • agentsgen meta — Shipped CLI surface for SEO + AI metadata generation.

Decision & strategy protocols

  • DecisionMap — Protocol + prompt toolkit for turning complex business, product, market, and marketing decisions into strategy maps, validated JSON outputs, and cascade-log update loops.

Release & publishing

  • git-tweet — Turn git changes into tweet-sized release notes.

Utilities

Agentsgen family naming

Agentsgen commands are presented here as separate tool pages for discoverability. They still ship together as one package: agentsgen.

Control plane surfaces

  • What to review next — Read-only planning queue with status, priority, workflow-sync hints, operator queue, and richer proof-loop readiness signals from the SET planner.
  • Proof queue — Read-only proof-loop queue for blockers, review-ready tasks, evidence quality, and recommendations.
  • Repo cards — Aggregated view combining registry baselines, latest workflow status, workflow sync, repomap metadata, and proof status.
  • Registry snapshot — Read-only view of repo baselines from the SET central registry.
  • Workflow status snapshot — Read-only latest GitHub Actions run per registered repo plus sync state, operator queue, and proof status.

Maintenance

What's inside

  • Registry snapshot generator: scripts/sync_registry_snapshot.py
  • Workflow status generator: scripts/sync_status_snapshot.py
  • Repo cards generator: scripts/build_repo_cards_snapshot.py
  • Planning snapshot generator: scripts/sync_planning_snapshot.py
  • Proof snapshot generator: scripts/sync_proof_snapshot.py
  • Snapshot outputs:
    • docs/registry/index.html
    • docs/assets/registry-snapshot.json
    • docs/status/index.html
    • docs/assets/status-snapshot.json
    • docs/repos/index.html
    • docs/assets/repo-cards-snapshot.json
    • docs/planning/index.html
    • docs/assets/planning-snapshot.json
    • docs/proof/index.html
    • docs/assets/proof-snapshot.json
  • Home page: docs/index.html
  • Tool pages: docs/tools/<slug>/index.html
  • SEO basics: docs/robots.txt and docs/sitemap.xml
  • Theme assets: docs/assets/asciitheme.css, docs/assets/ascii-theme.js

Snapshot behavior

  • Planning, repo cards, and status surfaces can include workflow sync state and operator queue when planning artifacts are present.
  • Planning, repo cards, and status surfaces can also show compact repomap status, policy modes, active slices, slice source labels, and top ranked files when local repo artifacts are available.
  • Proof queue and related surfaces remain snapshot-based: they reflect the latest local rebuild, not a browser-side live GitHub read.

Tool pages (routing)

Visual system

ABVX Lab currently uses the alt-b production shell:

  • docs/assets/lab-alt-b.css is the live stylesheet for the home page and tool pages
  • Control-plane pages (planning, proof, registry, repos, status) also use lab-alt-b.css through a compatibility layer over their existing snapshot markup
  • The old AsciiTheme assets remain in the repo for older/internal surfaces, but they are no longer the main live shell for the public catalog

How to add a new tool

Use this checklist:

  • Create a new tool page from an existing docs/tools/<slug>/index.html
  • Update the title, one-liner, links, metadata, and canonical URL
  • Add the tool entry to docs/index.html in the right group
  • If it is the newest tool, move it to the top of its group and optionally mark it NEW
  • Add the tool URL to docs/sitemap.xml
  • If the tool has a live site, add its live link on both the home entry and the tool page

Deploy

GitHub Pages publishes this site from /docs on main.

Flow: commit -> push -> wait for Pages.

If you change asset URLs or ship a static asset that browsers may cache aggressively, add or update the cache-busting query suffix in the HTML.