Skip to main content

Documentation Index

Fetch the complete documentation index at: https://labs.prompthon.io/llms.txt

Use this file to discover all available pages before exploring further.

Need help?
Ask in DiscordOpen a GitHub issueSupport guide
Skill packages are local-agent workflow contributions. They package a repeatable way of working so a learner or operator can run it with an AI assistant, inspect the steps, and adapt the workflow for their own environment. This contribution type is not limited to one assistant product. A good package defines the workflow, the human-facing explanation, the agent-facing instructions, and any helper files needed to make the workflow reviewable and repeatable.

​
What belongs here

  • Local-agent workflows that need more structure than a single prompt
  • Repeatable research, review, capture, cleanup, or monitoring routines
  • Helper scripts, references, or rules that make the workflow safer to run
  • Clear instructions for when a human should review, approve, or take over

​
Where skill packages belong

Skill packages live under skills/<skill-slug>/.
SurfaceWhat it coversExample packages
skills/<skill-slug>/README.mdHuman-facing explanation of the workflow and audience.Daily News Watcher overview, Garbage Collector workflow
skills/<skill-slug>/SKILL.mdAgent-facing invocation rules and operating contract.when to run the skill, what steps to follow
skills/<skill-slug>/scripts/Deterministic helpers that make the workflow repeatable.fetchers, analyzers, report writers
skills/<skill-slug>/references/Rules, source notes, checklists, or supporting context.cleanup rules, source lists, escalation checklists

​
Issue and review flow

Create a GitHub issue before building the package. Use the Practitioner Skill Package form so the workflow, package shape, local integrations, and safety boundaries are visible before work begins. The core team reviews the issue first. They may approve the scope, ask for changes, or decline the proposal. After the issue is approved or acknowledged, the contributor can fork the repository and make the change. The core team then reviews the pull request separately. They may approve it, request revisions, or decline the PR. If the issue or PR is declined, the contributor can still keep their fork and reuse the work outside this repository.

​
Package shape

Each package should stay small and legible: 
skills/<skill-name>/
  README.md
  SKILL.md
  agents/openai.yaml
  scripts/
  references/
Use README.md for the human-facing story: why the package exists, who it is for, and what workflow it teaches. Use SKILL.md for the agent-facing invocation contract. Use scripts/ only for deterministic helpers, and use references/ for rules, source notes, or supporting guidance.

​
Working rules

  • Keep the package focused on one repeatable workflow.
  • Make the human role clear before any automated step.
  • Prefer small helper scripts over long manual command blocks.
  • Include enough documentation for review, but do not turn the package into a general article.
  • Link from the package back to the relevant handbook page, project, or contributor discussion when useful.

​
Completion standard

A skill package is ready for review when it:
  • explains the workflow and audience
  • includes both human-facing and agent-facing instructions
  • keeps helper files minimal and deterministic
  • documents safety boundaries or approval points
  • can be reviewed as one coherent package