·4 min·Release

Leopoldo v2.1.0 ships: plugin packaging craft

A craft pass on the packaging surface. Playbooks you can opt into, a narrative project memory the system writes to across sessions, automatic compliance disclaimers on 129 regulated capabilities, and a new delivery-team plugin. Zero breaking changes.

Leopoldo v2.1.0 ships today. It is a craft release. No breaking changes, no user-visible UI shuffle. What changed is the depth of what happens when you invoke a Domain: the state the system carries between sessions, the compliance language it appends to regulated outputs, the way house methodologies slot into the same capabilities everyone else uses.

Five surfaces moved.

Playbook convention

Every consulting firm has its own IC memo format. Every legal team has its own contract review checklist. Every marketing group has its own campaign cadence. Until now, these lived as PDFs on a shared drive or as tribal knowledge in senior heads.

v2.1.0 adds a YAML convention for them. Drop a file into `.leopoldo/playbooks/ic-memo.yaml`, mark it active, and the ic-memo capability loads it at session start. The output follows your house style. No playbook present means commercial standard with a visible label, so there is never ambiguity about which methodology produced the artifact.

Six templates ship out of the box: ic-memo, contract-review, consulting-engagement, marketing-campaign, clinical-trial, engineering-delivery. Twenty-two high-traffic capabilities opt in.

Narrative project state

Claude Code and Cowork already remember conversations. What they did not remember was the *project*. A finance engagement that runs across twelve sessions over three weeks has decisions, patterns, and open loops that deserve to persist explicitly, not just implicitly in chat history.

Leopoldo v2.1.0 now maintains `.leopoldo/system.md`, a narrative project state file. The system loads it at session start (so every new conversation has continuity) and updates it at session end (so decisions survive). Plain markdown with three sections: Decisions, Patterns, Open Loops. You can read it, edit it, commit it to your repo. It is a living memory, not a black box.

Automatic compliance disclaimers

Regulated capabilities carry regulatory risk. A legal drafting output without a disclaimer, a medical analysis without the scope caveat, a tax recommendation without the jurisdictional warning: these are not just missing polish, they are failure modes.

v2.1.0 formalizes this. 129 capabilities across 5 categories (50 financial-advisory, 40 legal, 23 medical, 13 hr-regulatory, 3 tax) now carry a disclaimer frontmatter field. The builder reads it at package time and injects the canonical disclaimer in two places: the capability context (so the system knows to produce compliant language) and as an explicit output requirement (so the disclaimer text actually lands in the deliverable).

Opt-in only. Capabilities without the flag are untouched.

Studio plugin

Building a Leopoldo package for a new client used to be a shell script and some copy-paste. v2.1.0 turns it into a proper workflow with three commands (/onboard, /profile, /deliver) living inside a dedicated delivery-team plugin. Backed by a leads pipeline (backend table, admin UI, audit trail, GDPR doc) so new engagements are tracked from first contact to shipped package.

Not user-facing. This is for the delivery team. A separate leopoldo-setup plugin for end-user onboarding is planned for v2.2.

/examples page

A new public page at leopoldo.ai/examples with 5 deep-dive domain scenarios: finance fund PE, consulting engagement, legal contract review, medical clinical trial, engineering delivery. Each one shows the system working end-to-end on a realistic task, from intake to deliverable. Useful for anyone evaluating whether a Domain fits their workflow.

The numbers

SurfaceChange
Playbook templates6 new
Capabilities with playbook opt-in22
Capabilities with automatic disclaimer129
Trilateral-tagged capabilities41
Commands migrated to prompt card standard12
New tests (hooks, state, migration)54
Total test suite158 pass
Breaking changes0

Upgrade path

Run `/leopoldo update`. The update checker reads your manifest, pulls only managed capabilities with unchanged hashes, preserves every customization you made, and takes a snapshot before applying. If anything feels wrong, `/leopoldo rollback` restores the previous version.

The full migration guide is at `docs/guides/migration-v2.0-to-v2.1.md` in the public monorepo. A 5-minute read.

Why this release

v2.0.0 made the skill corpus leaner. v2.1.0 makes the packaging craft-level. State, compliance, methodology, delivery: four seams where the system used to ask you to trust it, now tightened to where it shows its work.

Choose a domain. Get the full system.

Frequently Asked Questions

What is new in Leopoldo v2.1.0?
Five surfaces changed. Playbook convention (opt-in YAML workflows in .leopoldo/playbooks/ with 6 templates). Narrative project state (a .leopoldo/system.md file the system loads at session start and updates across sessions). Automatic compliance disclaimers on 129 regulated capabilities across 5 domains (legal, medical, financial-advisory, tax, hr-regulatory). The studio delivery plugin for the build team. And a /examples public page with 5 deep-dive domain scenarios.
Are there breaking changes?
None. Full backward compatibility with v2.0.0. All user customizations, manifests, and domain selections are preserved on update. Run /leopoldo update in Claude Code and the new capabilities arrive cleanly.
What is a playbook?
A YAML file you drop into .leopoldo/playbooks/ that captures your team's house methodology for a recurring workflow. When present and marked active, the system loads it at session start and the relevant capability opts in. When absent, the capability falls back to the commercial standard with a labeled note. Six templates ship: ic-memo, contract-review, consulting-engagement, marketing-campaign, clinical-trial, engineering-delivery.
How does disclaimer injection work?
Regulated capabilities (legal drafting, medical analysis, financial advisory, tax, hr-regulatory) now carry a disclaimer frontmatter field that the builder reads at package time. The system injects the canonical disclaimer both in the capability context and as an explicit output requirement. 129 capabilities are covered today. Opt-in only, so capabilities without the flag stay clean.
What does the narrative project state do?
The system now maintains a .leopoldo/system.md file that captures decisions, patterns, and open loops for your project. It is loaded at session start so every new conversation has continuity, and updated at session end so decisions survive across sessions. Think of it as a living project memory the system writes to, in plain markdown you can read and edit.
How do I upgrade?
Run /leopoldo update in Claude Code or Cowork. The update checker reads your manifest, pulls only managed capabilities with unchanged hashes, preserves every modification you made, and takes a snapshot before applying. If anything feels off, /leopoldo rollback restores the previous version. The full migration guide is at docs/guides/migration-v2.0-to-v2.1.md in the public monorepo.
releasev2.1.0packagingplaybookscompliancestudio