Skills & Workspace Knowledge

Last updated: November 17, 2026

Skill and workspace knowledge help you integrate your organization's standards and preferences that Rally cannot enforce as settings, so AI-assisted research follows how your organization actually works.

What skills and knowledge are

Most of this guide covers standards you configure. A locked consent form holds. A governance rule is enforced. A template preset applies every time.

Some standards have no setting. Which incentive fits an audience, how your team words an invitation, what makes a good screener question for your product, when a study needs a second reviewer. Rally gives you 2 places to put those, and both reach AI-assisted work in the Rally app and through connected tools.

Workspace knowledge holds facts and policy. A product glossary, your incentive rates, what a team calls its customer segments. It sits in front of every conversation as background. An edit reaches the next conversation.

Skills hold procedure. The steps your organization follows for a particular kind of work, read when that work comes up rather than carried in every conversation.

Rally ships starter skills, copied into your workspace as your own. Editing them to match how your organization works is the expected use. Changes Rally makes later do not reach back into your copies.

Neither one enforces anything

Both are guidance. They shape what an assistant proposes. They don't enforce any hard guardrails, but they are your next best approach to helping people make smart choices.

Workspace knowledge carries no authority to bypass an approval or a consent requirement, by design.

A skill is read when the AI assistant decides to read it. Strong descriptions for your skills help the agent recognize when it needs to consult that skill to complete the task.

Choosing between knowledge and a skill

If it answers what is true here, it is knowledge. If it answers how we do this, it is a skill.

Workspace knowledge

Skills

Holds

Facts, definitions, policy, rates

Procedures and sequences

Best for

Things nobody should have to ask

Work done the same way repeatedly

Reaches the AI assistant

In every conversation

When the work comes up

Scoped by team

Yes

Yes

Knowledge is cheaper to get right and usually the first worth writing. An assistant that knows your product names and your standard rates stops asking and stops guessing.

Skills earn their place when the sequence matters. If the value is in the order of steps, or in a check that has to happen before something else, that is a skill. If the value is in a fact, writing it as a procedure buries it.

What to write first

Your templates, properties, populations, and governance rules already carry some of your standards. Start with what they cannot hold.

  • Facts a newcomer would not know. Product and feature names, segment definitions, which population serves which kind of study, what your internal shorthand means. This is the highest-value knowledge in most workspaces and the fastest to write.

  • Standards with a number attached. Incentive rates by study type, typical sample sizes, session lengths. Documented where an assistant can read them, these get proposed rather than asked about. Left in a slide deck, they get guessed.

  • Procedures with an order. What happens before a study goes to legal, how a screener gets reviewed, what to do when feasibility comes back short.

  • The judgment calls the rest of this guide flagged as yours. Where a module says Rally cannot enforce something, that is a candidate.

Write at the level of the work

Instructions that name a specific tool, argument, or field can go stale if those referenced items change over time. Instructions written in terms of the work hold up.

“Check whether Rally already tracks this on the person’s profile” survives a scenario when the specific tool in Rally was renamed.

Descriptions decide whether a skill gets read

An assistant sees a skill’s name and description before it sees the instructions, and chooses from those. The same discipline that applies to template and property names applies here. Say what the skill is for, in the words your teams use for that work.

What to leave out

Anything a setting already carries. A template preset, a governance rule, a property description, a consent form attached to a template. Each applies on its own. A skill restating one adds a second copy, and the two can drift apart the first time someone changes only one.

Anything that would be wrong if followed literally. Guidance is applied cold by whoever reads it next. An instruction that needs context to be safe is an instruction that will be misapplied.

Volume for its own sake. The catalog is read before anything is loaded, so a long list of thin skills makes it harder to find the right one. The same problem as template sprawl, in a different place. Group conceptual topics or phases together if they are performed together, Example: Guidance around sourcing and outreach can be its own skill separate from building a study since sourcing can happen repeatedly after a study is created.

Anything that belongs to one study. These are standing instructions. A note about a particular project goes in the study. These facts and guidance must be relevant on a repeated basis to be useful in this format.

Scope and access

Scoped to teams. Much like templates, budgets, and populations, a knowledge entry or a skill applies to every team or just to selected ones. That makes it the useful if standards differ by team.

Managing knowledge and skills are separate permissions. You can grant anyone with a full paid seat the ability to manage knowledge or skills within the workspace. It's best to keep these to a small, centralized group so they do not conflict with each other.

Knowledge deserves a tighter hand than skills. A skill applies when it is read. Knowledge is present in every conversation, so a wrong entry is wrong everywhere at once.

Review and maintenance

These age the way templates, but with less visibility. Nobody opens a knowledge entry to notice it is out of date.

Considerations for a maintenance and review plan:

  • Who owns them. The same person or team who owns templates and governance. Guidance that contradicts your configuration is worse than no guidance, and only someone who sees both will catch it.

  • When they get reviewed. Tie it to something that already happens. A rate change, a product rename, or a new procedure are the moments an entry goes stale, and those are easier to notice than a calendar reminder.

  • What triggers a new one. The strongest signal is a question your team keeps answering. If someone has explained the same thing three times, it is a good candidate for a knowledge entry or a skill.

  • Keeping track of changes. Keep a copy of each skill and knowledge item offline. Include a version #, last modified date, and author in each object. Hold a changelog with your offline copies to summarize what each new version focused on addressing.