Skill Trace Sampler
Use this skill when an agent builder needs safe, representative examples from a
completed agent run for SKILL.md examples, regression fixtures, demos, or PR
evidence.
Inputs
- Plain text transcript, JSONL transcript, or markdown run notes.
- Optional
--max-per-categoryto keep the pack small.
Side-Effect Boundaries
- Reads only the files passed on the command line.
- Writes only when
--out <path>is provided. - Does not call external services, execute transcript commands, or inspect live accounts.
Approval Requirements
No approval is required for local parsing. Approval is required before sharing sample output outside the workspace, because transcripts may include sensitive context even after redaction.
Validation
Run npm test, npm run smoke, and inspect the generated redaction notes before
using samples in public artifacts.