Packs
On this page
A pack gives Strata the skills, agent templates and connectors for one line of work, and you choose who gets it. Strata writes some packs, and you can write your organization's own. Packs live at Organization then Packs, which needs the Manage org settings permission.
#What a pack holds
- Skills: instructions the AI follows for that work. It reads a skill only when a question needs it, so a pack adds almost nothing to other chats.
- Agent templates: scheduled agents a member can start from Agents then Templates, under From your packs.
- Connectors it works with, listed so people know what to connect.
- A playbook (Strata's own packs): a few questions about how your organization works, answered once by an administrator. The answers go with the pack's skills every time the AI reads them.
#Finance & Accounting
The first of Strata's packs. It works from your own ledgers, spreadsheets and files, through SQL Server, OneDrive, SharePoint and Outlook.
| Skill | What it does |
|---|---|
| Month-End Close | The close as a checklist with business days, the accruals that are always needed, the reconciliations, a flux review and the sign-off |
| Account Reconciliation | A bank account, a subledger or an intercompany balance matched in passes, every reconciling item explained and aged, tied to zero |
| Variance Analysis and Commentary | Actual against budget, forecast or a prior period: the material lines, their drivers, favorable or unfavorable |
| Flash Report and Board Pack | The one-page monthly flash and the board pack, tied to the trial balance |
| 13-Week Cash Forecast | A direct-method, week-by-week forecast with the low point and the headroom |
Its agent templates: Month-end close kickoff (on the 1st of each month), Weekly cash forecast (Mondays) and Monthly flash draft (on the 5th). Each proposes, never posts or sends: entries are for your accountants to approve, and the flash is a draft for the person who runs the agent.
Its playbook asks when your fiscal year ends, your reporting currency, the scale you report in, how many business days your close takes, which variances are material enough to explain, and anything else about how you close and report. Leave a question blank and the skill uses its own default.
#Legal
For your legal team and the people who manage contracts. It works from the documents themselves, attached to the chat or kept in SharePoint, OneDrive or Outlook, and quotes the clause it relies on every time.
| Skill | What it does |
|---|---|
| Contract Review | A commercial agreement read clause by clause against your positions: each issue quoted, rated High, Medium or Low, with proposed wording |
| NDA Review | A non-disclosure agreement triaged as sign as it is, sign with listed edits, or send to a lawyer, with the reason for each |
| Redline Summary | Every change between two versions, whom it favors, its risk and a suggested response |
| Obligations and Key Dates | Parties, term, renewal and notice dates, payment terms and obligations pulled into a workbook, with each deadline worked out by formula |
| Data Processing Agreement Review | A DPA checked against what GDPR Article 28 requires, the transfer terms and breach notice: what is met, weak or missing |
Its agent templates: Contract renewal watch (Mondays), Contract inbox triage (weekday mornings, with Outlook connected) and Monthly obligations digest (on the 1st). Each lists what is due or what arrived for the person who runs it; none sends a notice, a reply or anything else.
Its playbook asks which governing law and courts you prefer and accept, the limitation of liability you accept and what stays uncapped, the indemnities you give and require, how you treat automatic renewal, who approves a deviation and when a contract goes to a lawyer, and anything else about how you review contracts.
The Legal pack helps your lawyers and contract managers; it does not give legal advice, and it never accepts, signs or sends anything. It has no case-law research: Strata has no legal database, so it would have to write citations it cannot check.
#IT Service Management
For IT operations and the service desk. It works from Jira Service Management (queues, requests, their SLAs and history), Confluence, Teams and Outlook, and from exports you attach, and names the source of every time and figure it uses.
| Skill | What it does |
|---|---|
| Incident Postmortem | A blameless postmortem from the incident's ticket, its history and the incident chat: impact, a timeline in UTC, time to detect and to mitigate, root cause and contributing factors, and action items with owners |
| Change Review | A change request checked before approval: risk, blast radius, test evidence, rollback, timing against other changes, and communication, with a recommendation for the approvers |
| Access Review | An export of accounts and permissions checked for leavers still active, orphaned and dormant accounts, privileged access, shared accounts and segregation-of-duties conflicts, each listed for a decision in a workbook |
| Service Desk Health | Volume in and out, the backlog by age, SLA breaches and requests about to breach, reopened requests and the largest request types, each with the query behind it |
| Runbook | A runbook or knowledge base article from resolved tickets or a procedure: when to use it, what is needed first, the steps, how to check it worked, how to undo it, and when to escalate |
Its agent templates: SLA breach watch (weekday mornings), Weekly service desk report (Mondays) and Postmortem action follow-up (Thursdays). They read Jira (and Confluence, for the follow-up) and list what needs attention for the person who runs them; none comments on, assigns or moves a request.
Its playbook asks which Jira Service Management projects are your service desk, your priority levels with their response and resolution targets, how you grade incident severity and which severities need a postmortem, who approves changes and what makes a change standard, normal or emergency, which systems are in scope for access reviews and who reviews each, and anything else about how your IT team works.
The pack never disables an account, approves a change, or comments on, assigns or moves a request on its own. A comment or a new issue happens only when your organization allows Jira writes and the person confirms it, as anywhere else in Strata.
#Sales & Revenue Operations
For sales leaders, revenue operations and account teams. It works from your CRM data copied into SQL Server or exported to Excel or CSV (attached, or in OneDrive or SharePoint), and from the deal's email threads in Outlook, and names the query or file behind every number.
| Skill | What it does |
|---|---|
| Pipeline Review | Coverage against the remaining target, the pipeline by stage, what was created, advanced, slipped, won or lost since the last review, deals past their close date or stalled, and the ones that need a decision |
| Forecast Call | Closed, commit, best case and pipeline against target, what changed since last week and why, and the deals that move the number; the owners' calls kept apart from the AI's observations |
| Deal Review | One opportunity against your qualification framework (MEDDICC unless you name another): what is known with its evidence, what is unknown, the risks, whether the stage is right, and the next steps |
| Account Plan | An account plan, a brief before a meeting or a QBR: what the customer has and uses, stakeholders, renewal risk with its signals, grounded expansion opportunities, and the plan for the quarter |
| Win-Loss Analysis | Win rate by segment, source, size and competitor with the number of deals behind each rate, cycle length, loss reasons grouped into themes, and what separates wins from losses |
Its agent templates: Weekly pipeline review (Mondays), Forecast call prep (Thursdays) and Renewal watch (on the 1st), each with your SQL Server connection. They read the sales data and list what needs attention for the person who runs them; none changes a record or emails anyone.
Its playbook asks where your sales data lives, your opportunity stages with their probability or forecast category, when your fiscal year starts and how quota is set, which qualification framework your teams use, the pipeline coverage you aim for and when a deal counts as stalled, and anything else about how your sales team works.
The pack never changes a CRM record, a stage, a close date or a forecast category, and never contacts a customer. A draft email happens only when the person asks, and it is never sent without their yes. It reports patterns by team, region and stage rather than ranking individual reps, unless the person asks for that.
#Who gets a pack
Open a pack with Manage. Under Who gets it, set a state for Everyone, and for any group that should have more:
| State | For a member |
|---|---|
| Off | Not offered: they do not see it |
| Available | Off until they turn it on in Customize then Packs |
| On by default | On until they turn it off |
| Required | On, and they cannot turn it off |
A member in several groups gets the most permissive state among Everyone's and their groups', the way Claude sets plugins per group. To give a pack to one group only, set Everyone to Off and that group to the state you want. With nothing set, a pack of Strata's is Available to everyone and a pack of your own is Off, so you can write its skills before anyone sees it.
Members manage their own packs in Customize, under Packs. A change reaches every chat within 20 seconds.
The pack's window also lists what it holds. Choose a skill of Strata's to read it as the AI reads it, with your playbook's answers; Back returns to the pack with your choices kept. Nothing is saved until Save.
#Writing your own pack
New pack asks for a name (up to 80 characters) and a description (up to 400). Then Add skill for each piece of work:
- Name, up to 80 characters.
- When to use it, up to 300 characters. The AI reads this to decide when the skill applies, so say what questions it is for.
- Instructions, up to 20,000 characters: what to do, step by step, in your organization's words.
A pack holds up to 20 skills, and an organization up to 25 packs of its own. In your own pack's window, each skill has Edit and Delete beside it, Edit name and description is under the pack's description, and Delete pack, at the bottom, deletes its skills with it.
Your skills shape how the work is done. They cannot change Strata's rules: approvals, permissions, privacy and safety still apply, and the AI is told so each time it reads one.
#How the AI uses packs
Only the packs that are on for a member are named to the AI, in one short line: each skill's name, and for a skill of your own the start of its When to use it. When the member asks for that work, the AI reads the skill first, ahead of a built-in skill on the same subject. A member with no pack on gets exactly what they had before packs existed. A scheduled agent uses the packs that are on for the person who owns it.
When a request asks for a pack's work, GPT is pointed at that skill. A request written in English that asks for a month-end close, a reconciliation, variance commentary, a flash report, a cash forecast, a contract or NDA review, a redline, renewal dates or a data processing agreement review (in its own words, or in the name or opening of a file attached to it) also tells GPT-5.4 which of the pack's skills to read. In our evaluation GPT-5.4 then read the right skill in 90 such requests of 90, against 25 of 90 with the line alone, and read none for requests that were not pack work. Claude Sonnet 5 reads them from the line alone (23 of 24). A request in another language has the line.
GPT-5.4 mini is not told about packs. In our evaluation it read a pack's skill in only 3 finance requests of 12, however the line was worded, and the line made it slightly less reliable on long multi-step requests. For pack work, choose GPT-5.4 or Claude Sonnet 5.
#In the audit trail
Every change is recorded with who made it: org.pack.offered (with the states before and after), org.pack.playbook_saved, org.pack.created, org.pack.updated, org.pack.deleted, and org.pack.skill_created, org.pack.skill_updated and org.pack.skill_deleted. See Audit log.