Import an existing skill¶

Goal: Bring a skill defined elsewhere (a prompt, a small script, another agent framework's spec) into reyn's DSL.

When to use¶

You have a working prompt or workflow in another tool and want to keep its behavior while gaining reyn's structure (validation, events, replay).
You're rebuilding a one-shot prompt as a multi-phase reyn skill but want a draft to start from.

Use the `skill_importer` stdlib skill¶

reyn run skill_importer "<paste your existing prompt or workflow description>"

skill_importer reads the input, infers a phase graph, and writes a draft skill into reyn/local/<name>/. It uses lint Control IR ops to verify its output before declaring success.

Workflow¶

Paste or describe. Give skill_importer either the raw prompt text or a description of what the skill should do.
Review the draft. The importer writes skill.md, phases/*.md, and artifacts/*.yaml under reyn/local/<name>/.
Lint. reyn lint <name> should be clean. If not, the importer reports the issues at the end of the run.
Run. reyn run <name> "<sample input>". Iterate the phase instructions if the output isn't right.
Promote. When happy, move from reyn/local/ to reyn/project/ to check it in.

What the importer maps¶

Source concept	reyn equivalent
Single prompt	One phase + skill graph `entry → end`
"Step 1, then step 2"	Multiple phases connected linearly
"If X, do Y; otherwise Z"	A branching phase (`triage → [branchA, branchB]`)
Tool calls (read file, search)	Control IR ops
Repeated structured output	Artifact schema

The importer doesn't always pick the cleanest decomposition. Use skill_improver to refine after import:

reyn run skill_improver "improve <name>" --allow-shell

Promotion checklist¶

Before moving to reyn/project/:

[ ] reyn lint <name> is clean.
[ ] At least one happy-path eval case passes.
[ ] final_output artifact matches what callers actually need.
[ ] Phase instructions follow P8 (no schema enumeration, no Control IR syntax).