Bearing Lite 1.0.3 documentation
Bearing Lite is a released, skills-first Agent Plugins package for teams using AI agents on real repositories. It gives work one visible path from an owner-approved plan to an independently assessed result.
- 1 · Input Scoped task Intent, scope, and constraints settled
- 2 · Gate Owner authorizes One human-only decision
- 3 · Process Build + assess Assurance and review assess the candidate
- 4 · Output Result + evidence Verified change, traceable record
Invalidating discovery returns to the owner.
At a glance: a scoped task enters, the human owner authorizes the exact plan, bounded work runs under independent checks, and the result closes out with evidence. Invalidating discoveries return to the owner; the diagram explains the path but grants no authority.
Start with a bounded task
Section titled “Start with a bounded task”Use Bearing Lite when a change needs an explicit scope, deterministic checks, and evidence that someone other than the candidate author can assess. It is for people directing work; Owner Authority remains human-only.
Install the published skills package, then give an agent a real bounded task:
pi install npm:@alphazede/bearing-liteUse Bearing Lite to add rate limiting to this API without changing its public responses. Require independent assurance at the configured cadence.
What happens next
Section titled “What happens next”Bearing Lite fills only missing planning stages, produces a five-artifact package, and pauses for one integrated owner review before bounded implementation dispatches. The package is portable: it supplies skills, references, templates, and optional client adapters without selecting a model, provider, credential, or launcher.
Use the installation guide to choose a host, then configuration to make explicit profile choices. Read the lifecycle before treating a visible diagram or a Definition of Done Manifest as authority.
When to use another path
Section titled “When to use another path”Bearing Lite never grants release or deployment authority, chooses an agent route, or certifies its own work. A scope, interface, security, acceptance, or authority change returns to the owner. An already-authorized deployment follows completion under separate operational checks and rollback readiness.