Skills, instructions, policies and MCPs

01Set it once. Use it across providers.#

Create and assign your skills, instructions, policies and MCP permissions in the Console. Once the Workstation is connected to that workspace and team, its supported agents receive the applicable configuration across providers. You do not need to paste the same rules into each conversation.

For example, a sales team can share a proposal-writing skill, its preferred writing style, a customer-data policy and an approved list of MCP servers. A Claude session and a Codex session in that team receive the same applicable guidance, even though they use different provider accounts.

Halofy checks the conversation’s current team access when supplying configuration for a turn. Published changes apply to subsequent turns. Uploaded skills can be available within their authorized scope while background checks report advisory findings; generated playbooks retain their publication review. See current skill checks.

02Skills give agents a repeatable method#

Use Skills for a specific job: preparing a research brief, drafting a proposal or checking a release. A skill packages the inputs, steps and expected result so another session can reuse a successful approach.

Add a skill from the library or submit your own. Review its checks and approval state, then assign it to the relevant scope. Connected agents can discover the assigned skills and use the ones relevant to the task.

03Instructions keep everyday expectations consistent#

Use Instructions for guidance that applies across tasks: your writing style, output format, project context or working conventions. Publish the enabled version in the appropriate scope.

Organization guidance can apply to child teams alongside their own instructions. This gives everyone a common baseline while keeping team-specific practices relevant.

04Policies define required rules and advisory guidance#

Use Policies for standards, procedures and rules agents should follow. Mark a policy Required when agents must follow it, or Advisory when it provides recommendations. Assign it to the scope where it applies.

Required policies are supplied as instructions to connected agents. This makes the rules consistent across providers; model compliance is not guaranteed, and delivery alone does not create a technical block on a provider’s tools or computer access.

05MCPs define which tool servers agents may use#

Use MCPs to select approved external tool servers for your organization or team. A team can narrow the parent scope’s allowed list. Saving an empty list allows no MCP servers; leaving a scope unconfigured does not create its own list.

The allowed server identities are included in the configuration supplied to Workstation agents. Agents are instructed to use the approved servers and seek administrator approval for others. This policy does not install a server or connect an account; the provider still needs the relevant tool connection and credentials.

Use scoped badges to control which Halofy memory an agent can actually read or write. Together, shared guidance and scoped access let you reuse a way of working across agents while keeping each team’s context relevant.

06Reuse a playbook and understand its checks#

A published playbook is a reusable workflow delivered through the skills library. Start with a familiar task, provide the requested inputs and ask your agent to follow the method. A separate guarded run is optional.

Current upload behavior: successfully uploaded skills can be available in their authorized scope while background safety checks report advisory findings. Read those findings and review unfamiliar instructions before using them. Generated playbook drafts still follow their publication review, and guarded runs have their own readiness checks.

Uploading a skill, seeing it in the library, the agent reading it and the agent successfully following it are different steps. Check the actual result.

07See the guidance assigned to you#

Members can open Policies, Instructions and MCPs in a read-only view of their team’s applicable configuration. Search, browse, open available details and refresh the page to understand the guidance your agents receive.

Missing edit or publish controls are expected in this view. Ask an authorized owner or admin to change the shared configuration. Seeing an allowed service does not sign you in to it or install its tools.