Create a skill

Under the Open .skill Protocol, agents author portable skills; humans review and approve release. Point your AI at the work — do not treat the CLI as a human click-ops checklist.

Want the full interview-driven walkthrough instead of a single prompt — every field explained, in order, with the reasoning behind it? Point your agent at examples/skillerr-authoring/SKILL.md in the skillerr repo. It takes an agent from a plain conversation to a minted .skill without hand-writing the contract JSON.

Prompt your agent

Run these exact commands in your terminal, in order:

1. npm i -g skillerr          (skip if `skill --version` already works)
2. export SKILL_HOST=cursor   (replace "cursor" with your actual tool name)

Then, from this conversation, create a portable .skill with a redacted journey
and exact sections I approved (secrets as {{refs}}). Checkpoint for handoff, or
compile --approve --mint when release-complete. Do not invent filler. Show
status and the output path.

What your agent will do

Step	Agent runs
Init workspace	skill init --title "…"
Redacted journey	skill journey --summary "…"
Propose sections	skill propose --json '[…]' (exact approved bodies)
Status	skill status
Mid-work handoff	skill checkpoint -m "WIP"
Release	skill compile -m "…" --approve --mint

Required env when creating:

export SKILL_HOST=cursor
export SKILL_AGENT_INVOCATION=1   # preferred

Workspace model

Local working tree (git-like):

.skill/
  config.json
  sections/*.json
  index.json          # staged ids
  HEAD.json
  objects/*.skill

git	skill (agent surface)
init	skill init
edit	skill propose
add	skill add
status	skill status
stash / WIP	skill checkpoint
commit	skill compile
tag / sign	skill mint

No marketplace publish in the happy path — share the .skill file.

Prefer verbatim human-approved bodies. Secrets only as . If incomplete → compile_refused with missing fields. Fix and compile again. Do not invent filler.

Contract helpers the agent may use: skill contract-template, skill contract-check contract.json --profile release.

Multi-skill from a journey

Prompt:

Using skillerr, run agent-guide then extract from ./journey.json into ./extraction.
For each candidate I select, open its own workspace, fill missing contract fields,
and only compile a release when complete — otherwise checkpoint.

What the agent runs: skill agent-guide → skill extract ./journey.json -o ./extraction → one workspace per selected candidate → contract-check → checkpoint or compile.

Local / offline create

Ask the agent to set Ollama (or similar) provenance, then the same propose / checkpoint / compile path. Provenance remains self-reported unless a deployment adds a verified issuer.

What success looks like

• Continuity: handoff file loads in another agent with gaps visible
• Release: complete package, mint attestation present, inspectable TrustView

Next: Ingest a skill · Agents