Legal triages an inbound NDA against your playbook. Finance closes the month the same way it closed the last one. Support answers a queue the way your best agent would. These are open-source procedures a model follows step by step instead of improvising — and in Qua they run inside a Session, under the same tier clamps, the same masking before routing, and the same receipt as everything else you do here.
One playbook set per domain. Each bundles procedures — a call prep, a month-end reconciliation, an NDA triage — that a model follows step by step instead of improvising, so the same question gets the same treatment on a Tuesday as it does at quarter end. Every card carries its real status: a pack whose procedures reach for a CRM is not something we can honestly call live.
Your day planned from the tools you already work in.
Manage tasks, plan your day, and build up memory of important context about your work.
ExamplePlan my day — builds a task dashboard and remembers standing context between sessions.
Walk into every call already briefed.
Account research, call prep, outreach, and pipeline review — supercharged when your CRM is connected.
Example/sales:call-prep — prep for a call with account context, attendee research, and a suggested agenda.
A queue triaged the same way whoever is on shift.
Ticket triage, response drafting, escalation handling, and KB article generation.
ExampleTriage a ticket queue, draft responses, and turn a resolved issue into a KB article.
A spec that traces back to the research it came from.
Specs, sprint planning, roadmap updates, and research synthesis grounded in your tools.
Example/product-management:write-spec — draft a feature spec grounded in synthesized user research.
Content that sounds like your brand because the rules are written down.
Campaign planning, on-brand content, email sequences, SEO audits, and performance reporting.
ExamplePlan a campaign, draft on-brand content, and pull a cross-channel performance report.
An inbound contract triaged against your playbook, not someone's memory.
Contract review, NDA triage, risk assessment, and signature routing for in-house teams.
ExampleTriage an inbound NDA against your playbook, flag risky clauses, route for e-signature.
Month-end close that follows the same steps every month.
Close management, reconciliation, SOX testing, and variance analysis against your warehouse.
Example/finance:reconciliation — month-end account reconciliation and close support.
A question answered in SQL, checked before anyone acts on it.
Explore data, write and validate SQL, run statistical analysis, and build dashboards.
Example/data:write-query — write and validate SQL against the warehouse, then build a dashboard.
One question, answered from every system at once.
Search across all of your company's tools in one place.
ExampleOne query that finds a decision buried across email, Slack threads, and wiki pages.
Preclinical analysis wired to the databases it depends on.
Preclinical research tooling: literature search, genomics analysis, target prioritization.
ExampleQC a single-cell RNA-seq dataset and develop a Nextflow pipeline with literature grounding.
Productivity, Enterprise Search, Bio Research are built around live integrations rather than written knowledge — the procedure is mostly instructions for reaching a system Qua cannot reach yet. Running them today would produce something that looks like an answer and isn’t one, so we don’t. They stay catalogued, and their cards say why.
Skills tell a model what to do. They say nothing about what it may spend, what it may see, or what it must show you afterwards. That part is Qua’s, and a skill run inherits every bit of it.
Every run in a Qua workspace inherits the workspace’s policy, whoever or whatever started it.
A skill is Markdown, not a vendor binding. Nothing in a playbook welds it to one model or one provider.
The artefact at the end of a Qua run is not a vibe. It is a line-itemised record you can hand to finance.
The shape is real — the attribution badge, the receipt lines and the skill reference are what the product renders. The numbers are not: they are placeholders for layout, not a measured run, because what a skill costs depends entirely on the tier it resolves at and the length of your documents.
Illustrative figures for layout only — not a measured run, and not a claim about any particular workload. The tier ladder shown is Qua’s own, most expensive first: 01 Pro Models · 02 Fast Models · 03 Open Models · 04 Enterprise Search · 05 Personal Knowledge.
Built on the open-source knowledge-work plugins published at github.com/anthropics/knowledge-work-plugins under Apache-2.0, imported at a pinned commit and never a moving branch. Each pack’s full licence text is available in-app on its library page.
Anthropic does not endorse or sponsor Qua, and the two are not affiliated. The name is used descriptively, to identify the open-source work accurately; no trademark rights or licence are claimed or implied.
We would rather you hear it here than find out later. A skill in Qua is a written procedure the model follows against your prompt and your own knowledge base. What it cannot do is reach out — and the packs built entirely around reaching out say so.
Invoke one from the composer or by typing /legal:review-contract. It runs as an ordinary Session turn — same router, same clamps, same receipt — with the skill named on the answer and on the receipt.
Productivity, Enterprise Search, Bio Research depend on a live CRM, chat workspace, research database or bundled script that Qua does not run. They are catalogued and readable, never presented as ready, and we are giving no date.
Every procedure is stored in the open Agent Skills format — the same six frontmatter fields — so what you run here is readable, auditable, and not trapped here.