Skip to content

Installation

Jump to bottom
Jeremy Longworth edited this page Sep 12, 2026 · 5 revisions

Installation

AgentCannabis publishes portable Agent Skills packages. You can install a professional role package or an atomic skill with GitHub CLI, or load the repository folders directly in a local agent host.

Use a release tag for a reproducible installation. The current documented release is v1.0.5.

Prerequisites

  • Git and a destination Git project for project-scoped installs.
  • GitHub CLI with gh skill support.
  • PowerShell 5.1 or newer and Python 3.11 or newer for repository helpers and validation.

Check the installed command surface before choosing a host:

gh --version
gh skill install --help

Install a professional skillset with GitHub CLI

Run these commands from the destination project. The path is an exact folder inside the public repository, which avoids a full repository skill scan:

gh skill preview jeremylongworth-source/AgentCannabis skills/cannabis-compliance-specialist@v1.0.5
gh skill install jeremylongworth-source/AgentCannabis skills/cannabis-compliance-specialist --agent github-copilot --scope project --pin v1.0.5

The example targets GitHub Copilot. gh skill install supports other hosts; choose an --agent value supported by your installed GitHub CLI and follow the host's discovery instructions. Project scope places the package in the host's project skill location. Use --scope user when you deliberately want user-wide availability.

Do not combine an inline @VERSION with --pin. Put the ref in one place:

gh skill install jeremylongworth-source/AgentCannabis skills/cannabis-compliance-specialist@v1.0.5 --agent github-copilot --scope project

For development content, replace the tag with @main; a branch is mutable and should not be used when reproducibility matters.

Install an atomic skill

Choose a folder name from the Taxonomy page or skills/ directory:

gh skill preview jeremylongworth-source/AgentCannabis skills/build-crop-monitoring-plan@v1.0.5
gh skill install jeremylongworth-source/AgentCannabis skills/build-crop-monitoring-plan --agent github-copilot --scope project --pin v1.0.5

Atomic skills are useful for narrow tasks. Start with a professional wrapper when a request spans several families or needs role-level routing.

Install from a local clone

From an AgentCannabis checkout, the manifest helper validates the selected professional manifest and installs the self-contained wrapper into another Git project:

.\scripts\install_skillset.ps1 -Skillset cannabis-compliance-specialist -ProjectPath D:\MyProject -Ref v1.0.5 -Pin -WhatIf
.\scripts\install_skillset.ps1 -Skillset cannabis-compliance-specialist -ProjectPath D:\MyProject -Ref v1.0.5 -Pin

Add -IncludeMembers only when the destination host should discover every atomic member separately. The default wrapper includes its member index and references inside one installation folder.

Use the portable routing templates

For a local agent host that reads project instructions:

  1. Start with agents/AGENTS.base.md, agents/AGENTS.full.md, or the focused agents/AGENTS.<skillset-name>.md file.
  2. Load the selected skills/<name>/SKILL.md.
  3. Load referenced files only when needed for the task.
  4. Keep evidence gaps, source-currentness limits, and human approval gates in the output.

See Agent Routing and Architecture for the package and routing model.

Confirm availability

Use fictional records after installation:

Use $cannabis-compliance-specialist to identify missing licence, role,
inventory, quality, and source-currentness evidence in a fictional site
package. Do not approve, sign, submit, release, or operate anything.

The skill should identify the selected workflows, preserve missing evidence, state source limits, and route regulated or hazardous decisions to the responsible human or qualified professional.

Troubleshooting

  • If gh skill is unavailable, update GitHub CLI or use the portable local Agent Skills path.
  • If the package is not visible, confirm the destination project, exact folder, supported skill location, and host reload behavior.
  • If a skill already exists, preview the new version and inspect local changes before replacing it.
  • If a local validation command fails, run python scripts/validate_repository.py --stage full to isolate the first structural error.
  • If publishing from a Windows checkout reports a false Git ownership warning, use the repository's scoped wrapper rather than changing global Git config.

Publishing validation

From an AgentCannabis clone, preview the package without publishing:

.\scripts\publish_skill_repository.ps1 -DryRun

The wrapper scopes Git's safe.directory setting to the gh skill publish process when required. It does not change persistent Git configuration.

Clone this wiki locally