Risk before ritual
Depth follows blast radius, reversibility, ambiguity, and validation difficulty—not line count.
OPEN AGENT SKILLS · MIT
Engineering discipline for coding agents.
Baseline gives your agent a proportional path from a software request to a verified change—without adding a runtime, CLI, or project dependency.
Evidence before confidence
Expected behavior, fail-first checks, complete diffs, and fresh results stay visible.
Safety before speed
Trust, isolation, sensitive data, destructive actions, and blast radius remain explicit.
Authority stays human
Commits, releases, deployments, and irreversible actions require explicit permission.
WHY BASELINE
Fast code is useful. Unexamined confidence is not.
Coding agents can produce plausible changes quickly. Baseline keeps the engineering judgment around those changes proportional, inspectable, and under human authority.
THE PROBLEM
Speed can hide uncertainty.
An agent can write convincing code before it has identified the governing behavior, risk, or validation boundary.
THE RESPONSE
Route the work by risk.
Measure first, refine only real ambiguity, prove behavior at the smallest useful seam, then review at the selected depth.
THE RESULT
A workflow you can inspect.
Each step has a reason, a stopping condition, and an evidence boundary. Passing checks still do not become a claim of perfect correctness.
THE MAIN FLOW
Just enough process for the risk in front of you.
Start from governing input, measure risk, clarify only material ambiguity, check behavior fail-first, implement, verify proportionally, and keep Git authority explicit.
Open a branch only when the work calls for it.
New behavior or bug?
tdd starts from approved behavior. bugfix starts from a reproduced defect. Both preserve the fail-first boundary.
A real decision remains?
Clarify incompatible interpretations first, then compare viable options only when the choice is material.
Knowledge must persist?
Add durable documentation only for shipped behavior, stable boundaries, accepted decisions, operations, or incidents.
Security boundary involved?
Review authority, sensitive data, destructive actions, and trust boundaries with evidence proportionate to the risk.
COMPOSABLE WORKFLOWS
Use the path the change needs.
Baseline is a set of independent Agent Skills. Your agent can select a matching workflow from your request, or you can invoke one explicitly when you need a deterministic path.
Explore the skills$measurer Choose the depth
Classify risk, boundaries, reversibility, ambiguity, validation, and rollout exposure.
$tdd Prove the change
Start approved behavior with the smallest executable check that fails for the right reason.
$verify Review the whole story
Reconcile the request, expected behavior, tests, complete diff, fresh results, and residual risk.