Agent Skills - Lightsage

Lightsage provides three Agent Skills for turning workspace evidence into actions and operating Lightsage from a terminal.

Available skills

Skill	What it does
`from-prompt-runs-to-actions`	Turns stored prompt visibility evidence, mentions, sentiment, and citations into prioritized actions.
`from-eval-runs-to-actions`	Diagnoses eval outcomes and execution evidence, then proposes targeted fixes and validation runs.
`lightsage-cli`	Operates and troubleshoots the Lightsage CLI, including authentication, command discovery, resource IDs, writes, and recovery.

Install the skills

Install the repository with an Agent Skills-compatible client:

npx skills add lightsagehq/lightsage-skills

Choose the skills and supported agents you want when prompted. Clients that support automatic skill selection can choose the relevant skill from your request. You also need the Lightsage CLI and access to an authenticated Lightsage workspace. See Lightsage CLI for setup.

Example requests

“Review this week’s prompt runs and recommend the three highest-impact actions.”
“Diagnose the latest failed eval and propose the smallest useful fix.”
“Use the Lightsage CLI to show my recent eval results.”

Agent Skills are files installed in your coding agent. The resources returned by lightsage skills list are reusable skills attached to Lightsage evals and configurations. See Eval skills for those.

The skills are maintained in the lightsagehq/lightsage-skills GitHub repository.