File-based CLI basics

Create your first Lens

Goal: create .lenserfight/lenses/customer-summary/LENS.MD.

When to use it: you need a reusable prompt transformation.

Structure:

.lenserfight/lenses/customer-summary/LENS.MD

Item relationship: workflows can reference it with lens: customer-summary.

Validation: run lf validate .lenserfight/lenses/customer-summary/LENS.MD.

Common mistakes: missing frontmatter, vague output contract, placeholders without declared parameters.

Expected outcome: a reusable lens visible in project discovery.

Create your first Agent

Goal: create .lenserfight/lensers/researcher/LENSER.MD.

When to use it: you need a reusable AI role with operating rules.

Structure:

.lenserfight/lensers/researcher/LENSER.MD

Item relationship: workflows can reference it with agent: researcher.

Validation: run lf validate.

Common mistakes: mixing role instructions with one-off task input, omitting safety boundaries.

Expected outcome: a role that can be combined with lenses.

Create your first Workflow

Goal: compose a lens and agent into .lenserfight/colenses/research-brief/COLENS.MD.

When to use it: the same multi-step process happens repeatedly.

Structure:

.lenserfight/colenses/research-brief/COLENS.MD

Item relationship: nodes should reference discovered lens and agent slugs.

Validation: lf validate catches broken references.

Common mistakes: too many steps, unclear failure behavior, unowned output.

Expected outcome: a repeatable process with ordered steps.

Create your first Battle

Goal: compare two models, prompts, lenses, or workflows.

When to use it: the team needs evidence for a choice.

Structure:

.lenserfight/battles/review-comparison/BATTLE.MD

Item relationship: participants reference lenses, agents, or workflows.

Validation: lf validate catches broken participants.

Common mistakes: no rubric, comparing outputs that solve different tasks.

Expected outcome: a meaningful comparison with explicit criteria.

Create a Ray

Goal: create .lenserfight/rays/support/RAY.MD.

When to use it: templates need a shared category.

Validation: lf validate parses RAY.MD as a first-class item.

Common mistakes: using rays as vague hashtags instead of routeable categories.

Expected outcome: related templates become easier to browse.