Skill Author

Use this skill when the operator wants Allbert to create a new skill or refine a skill draft.

Intake Ritual

Collect enough detail before writing a draft:

1. Name: kebab-case, unique, and no more than 64 characters.
2. Description: a short user-facing summary.
3. Capability summary: what the skill does, required inputs, expected outputs, and any boundaries.
4. Script needs: recommend Python if the operator does not state a preference. Mention Lua only when embedded scripting is enabled and lua is allowed by policy.
5. Tool needs: choose the smallest useful allowed-tools fence.
6. Optional agent contribution: include only if the skill should expose a delegated agent.

Use request_input for missing details, list_skills to avoid name collisions, and read_reference only for references in already-active skills.

Drafting Contract

• Always create drafts with create_skill and skip_quarantine: false.
• Never call create_skill with skip_quarantine: true; that path is reserved for first-party kernel seeding.
• Drafts must land under ~/.allbert/skills/incoming/<name>/ and carry provenance: self-authored.
• Explain that the operator must review and install the draft through the standard skill install preview before it becomes active.
• Keep raw transcript excerpts, secrets, credentials, and private tokens out of the skill body.

Skill Body Guidance

Write the generated skill in clear operational prose:

• Start with when to use the skill.
• Name any required inputs.
• State safety boundaries and when to ask the operator before acting.
• If scripts are needed, explain what each script does and keep interpreter choices within the configured allowlist.
• Prefer simple markdown references over scripts when a skill only needs reusable instructions.