Skills
On-demand instruction playbooks for Venna Code — a SKILL.md plus bundled resources the model loads only when it needs them.
A skill is an on-demand instruction playbook the model loads only when it's relevant: a
SKILL.md plus any resources it bundles, under ~/.venna/skills/<slug>/.
SKILL.md is YAML frontmatter plus a markdown body. Frontmatter carries name, description, and
an optional allowed-tools allowlist; the body is the playbook.
---
name: release-notes
description: Draft release notes from a tagged commit range.
allowed-tools: [read_file, bash]
---
# Release notes
Read CHANGELOG.md, summarise the range, and open a PR draft…How a skill loads
Each turn lists only every skill's name and description in an <available_skills> block —
never the body. When the model calls the skill tool with a name, the full body loads (capped at
32 KB) and becomes the turn's active skill. If that skill declares allowed-tools, the turn's tools
narrow to the intersection with the session's base tools — a skill can tighten, never widen,
what the model may do.
Install
Fetch from the registry
The Venna API's skills proxy returns the SKILL.md and its sibling files.
Stage and verify
Files are staged and path-validated (no .. or absolute-path escapes), then renamed into
~/.venna/skills/<slug>/ in one atomic move.
Load
The engine reloads and registers the skill tool for the primary agent and sub-agents alike.
Disable a skill per session and the tool rejects it, naming the loadable set back so the model
self-corrects.
The full SKILL.md reference lives in Skill authoring.