ABVX Lab logo ABVX Lab
tool surface

ABVX tool

agentsgen

Agent contract layer for turning repos into AI-readable, checkable surfaces with reversible-work rules, readiness reports, safe fixes, fleet scans, optional Rabbithole seeds, and review-lens handoffs.

GitHub Docs Live Discovery JSON

Core outputs

  • AGENTS.md and repo docs with safe marker updates.
  • Generated workflow sections tell agents to use proposal/worktree/draft-branch output for risky work before inspect -> select/apply/discard.
  • check --report readiness scoring and conservative fix --all remediation.
  • fleet scan reports for read-only team rollout across many repositories.
  • llms.txt, docs/ai/, snippets, and versioned contracts for CLI/MCP consumers.
  • rabbithole-seed creates docs/ai/rabbithole.seed.md with suggested review lenses for local human-in-the-loop doc exploration.

Agent discovery

The product publishes a small discovery layer for agents: /llms.txt, /.well-known/integrations.json, and /.well-known/agent-card.json.

The local MCP server remains stdio-only via agentsgen mcp, so the public declaration points agents to the CLI instead of pretending a remote MCP endpoint exists.

MCP, CLI, Skills

AGENTS.md does not replace MCP or CLI. It binds them to repo rules: MCP for external access, CLI for deterministic execution, and skills for reusable gates.

Use agentsgen when the missing layer is the repo contract that tells agents which tools are allowed and which checks prove the result.

Quickstart

Primary path

agentsgen init . --defaults --autodetect
agentsgen check . --all --report
agentsgen fix . --all
agentsgen rabbithole-seed .
agentsgen fleet scan ~/code --max-depth 2

Use init to establish the repo surface, check --report to score readiness, fix --all to apply safe remediation, rabbithole-seed for optional local exploration, and fleet scan for team rollout planning.

Reflection outputs

Producer side of the reflection loop

Commands

agentsgen reflect sessions .
agentsgen reflect skills .

Artifacts

  • docs/ai/agent-signals.json
  • docs/ai/session-patterns.md
  • docs/ai/skill-usage.json
  • docs/ai/skill-effectiveness.md

Current downstream consumer: ABVX Agent Skills reads these files through session-retrospective and skill-effectiveness-audit. In the broader stack, ID provides human bootstrap, agentsgen emits repo-local traces, and skills turn those traces into reusable behavior.