Public source-available • npm latest

AXF packages local repo know-how as agent capabilities.

Every workspace has ceremony: how to build, test, search, diagnose, scaffold, promote, and recover. AXF turns that ceremony into small, self-describing capabilities an agent can inspect and run.

It is not a universal command catalog. It is a workspace-owned exoskeleton for the agent operating inside that workspace.

One MCP tool
axf doctor
axf list
axf inspect echo say
axf run echo say --message hello

# MCP entrypoint
axf mcp

When AXF helps

Repeated repo discovery

Agents keep asking which test command, build command, or diagnostic command applies.

Local safety rules

Important rules live in prose, convention, or human memory instead of inspectable capability metadata.

Brittle shell chains

Common chores are rebuilt turn after turn instead of promoted into stable capability contracts.

Current surface

Scout

Inspect a workspace and identify candidate capabilities or missing local bearings.

Inspect

Ask what a capability is, what arguments it expects, and how AXF would launch it.

Run

Execute known capabilities through AXF's adapter path instead of raw shell fallback.

Scaffold

Create draft capabilities and adapters that can be reviewed before becoming active.

Promote

Move capabilities through lifecycle states when they are ready for agents to rely on.

MCP

Expose one axf MCP tool that routes into the same capability surface.

Project and execution roots

AXF separates the root used to discover capability manifests from the root where execution should occur. In simple repos they are the same. In more advanced agent setups, the execution root can be a contained work area while the project root remains the canonical capability registry.

{ "env": { "AXF_PROJECT_ROOT": "/path/to/project", "AXF_EXECUTION_ROOT": "/path/to/execution-root" } }