SET
Keeps repo AI surfaces fresh in CI: docs, maps, checks, proof loops, registry workflows, proposal lifecycle contracts, optional review lenses, and local plan seeds.
Open tool infrastructure / ABVX catalog
Turn a GitHub repo into an AI-ready workspace.
Repo docs, context surfaces, workflow loops, and checks for teams doing AI-assisted coding seriously.
Core stack. Real tools. One control-plane view.
Not another assistant. The repo and workflow layer around one.
Core tools
Primary surfaces in the stack
The main public stack: orchestration, repo surfaces, human context, analysis, protocol, reusable capabilities, proposal-first execution discipline, review lenses, and optional local exploration.
agentsgen
Turns a repo into AGENTS.md, llms.txt, docs/ai, checks, CLI-readable contracts, and review-lens exploration seeds.
ID
Keeps human and operator context portable across repos, tools, and local automation.
repomap
Token-budgeted repo map with ranked files, import graph, and focused slice modes.
LWP
Desktop-first execution protocol with clear task briefs and verification loops.
ABVX Agent Skills
Portable skills for product context, local friction reports, proposal-first growth loops, MCP access, CLI execution, proof gates, review lenses, staged skill-quality evolution, and optional doc exploration.
ABVX Skills Knowledge
Portable OKF-style catalog export for grouped skill navigation and downstream consumption.
Start here
Pick the first path that matches your job
These three entry paths are the fastest way to know where to start.
Solo devs
Human: Turn one repo into an AI-ready workspace from one setup path.
Technical: Start with SET or go straight to agentsgen.
Teams
Human: Keep multiple repos aligned with visible workflows and CI-driven updates.
Technical: Use SET as the control-plane entrypoint.
Tool builders
Human: Build agents and automations on top of portable context and reusable capabilities.
Technical: Combine ID with ABVX skills and repo-local hooks.
Stack roles
What each layer is responsible for
ABVX complements editor assistants with repo state, shared context, and verification surfaces.
| Tool | Role in the stack | What the user feels |
|---|---|---|
| SET | CI and workflow orchestrator for repo-docs, proof loops, registry review, and repo hooks. | Your repo AI surfaces stay fresh without hand-running the same maintenance steps. |
| agentsgen | Repo docs and contract generator for AGENTS.md, llms.txt, docs/ai, bundles, and checks. | You get a repo that is easier for humans and agents to read, validate, and update safely. |
| ID | Portable human context layer for operator profiles, hooks, and repo-local workflow metadata. | Your context can travel across repos and tools instead of being trapped in one chat session. |
| repomap | Repo analysis layer with ranked files, import graph, and focused code slices. | You can get to the right files faster instead of handing the whole repo to a model. |
| LWP | Execution protocol for task briefs, proof loops, and verification structure. | Agent work becomes easier to inspect, retry, and review. |
| ABVX Agent Skills | Reusable capabilities layer with validated workflows agents can load on demand. | You get repeatable capabilities instead of rebuilding prompts and checklists every time. |
Hub flow
How the AI coding tools stack fits together
ABVX does not choose between MCP, CLI, Skills, and harnesses. It separates the workflow into access, execution, discipline, and runtime gates so agents can connect, act, prove work, and stay inside explicit safety boundaries.
1. Access
MCP and integration discovery connect agents to external systems, auth boundaries, and service APIs.
2. Execution
agentsgen, SET, CI, and local CLIs run deterministic, reviewable work.
3. Discipline
AGENTS.md and ABVX Agent Skills define when to use MCP, when to use CLI, when to keep work reversible, which proof or review gates are required, and when a skill-quality change needs held-out validation.
4. Harness gates
Agent runtimes enforce loop budgets, tool registries, permissions, sandboxing, observability, scheduling, evals, and inspect/apply/discard settlement for retained outputs.
access: MCP / integrations
execution: agentsgen / SET / CLI / CI
discipline: AGENTS.md / ABVX Agent Skills / proof and review gates
harness: loops / permissions / sandbox / evals / proposal settlementID still supplies human context, while agentsgen creates the repo contract, ABVX Agent Skills turn repeated operating rules into portable capability, and the harness enforces runtime boundaries.
Control plane snapshot
Operational visibility across tracked repos
What you are seeing: the day-to-day view of how SET and agentsgen keep repos in sync and surface drift or proof signals.
AGENTS.md_generator
repo-docs baseline · MCP + LLM shipped
repo-docs
2026-09-11
success · v0.5.0 released
ready-now
ID
repo-local hook layer
repo-docs
2026-09-11
success · v0.5.1 released
ready-now
SET
orchestrator and registry owner
repo-docs
2026-09-08
success · v0.4.0 released
ready-now
lab.abvx
public catalog and control plane
site-ai
2026-09-07
release · v0.2.0
ready-now
Quiet directory