clawperator-skill-author-by-recording

To turn a phone recording into an AI agent skill, you record a real device flow, export the artifacts, and scaffold a runtime skill. The workflow helps you derive a skill_id and run a self-test to produce a SkillResult.

The workflow for authoring a skill from a device recording involves recording a device flow, scaffolding the skill artifacts, and running a self-test. It automatically chooses replay or orchestrated execution based on recording evidence.

You do not need to manually choose replay or orchestrated execution. The workflow automatically decides the skill shape based on the recording evidence and your input, ensuring the generated skill matches the captured device flow.

If a skill self-test does not produce a passing SkillResult, the workflow enters a repair loop. This allows you to patch the skill artifacts and re-run the self-test until a passing result is achieved.

Skill authoring from a recording preserves workflow evidence by keeping a sanitized baseline and per-run artifacts. This ensures auditing accuracy while enabling the repair loop to reference original device flow data.

You can use a recorded workflow to test AI agent skills automatically through a self-test mechanism. The recording is scaffolded into a runtime skill that emits a SkillResult, proving the workflow functions as captured.