Agent Skills for Viva Insights

Start using

1. Install the Skill

   Run this in a terminal with a recent GitHub CLI. It copies the whole Skill folder, including its reference files and worked examples, into your personal Copilot Skills directory.

   gh skill install microsoft/viva-insights-sample-code frontier-analytics/skills/viva-insights-analysis --agent github-copilot --scope user

2. Load and check the Skill

   In GitHub Copilot CLI, run /skills reload and then /skills info viva-insights-analysis. For another agent, change --agent in the install command and restart the agent.

3. Try a starter prompt

   Use the Viva Insights analysis Skill to import my Person Query export, validate it and summarise any data-quality issues before we start the analysis.

   Add the location of your export and whether you prefer R or Python.

If gh skill is not available, or you use a different agent, follow the manual installation guide. Skills are not verified by GitHub, so review the installed files before use.

Overview

Questions this Skill helps you answer:

• Is my Person Query export complete enough to analyse, and which data-quality issues should I fix first?
• How is Copilot usage distributed across my organisation, and which usage segments stand out?
• Which collaboration metrics differ most between groups, without reporting groups below the privacy threshold?
• Can you turn this export into an adoption summary or an executive memo with reproducible code?

Results

What you can expect back:

• R or Python code that uses existing vivainsights functions before writing custom logic.
• A validation summary that flags common export pitfalls, such as text-coded flags, literal "#N/A" values, holiday weeks and groups below the privacy threshold.
• Standard visualisations and common deliverables, such as an adoption dashboard, an executive summary memo or an ROI estimate.

Limits to keep in mind

• Agent-guided local R/Python analysis. Export layouts and installed APIs require verification for each workflow.
• Verify installed APIs and generated outputs. Built-in and synthetic examples are not real-export validation.
• Prefer enabled-user/enabled-days fields. Documented native licensed-only M365 Copilot nulls mean unlicensed. Failed joins remain unresolved. These licence semantics do not apply to unlicensed Copilot Chat or Consumption credits.
• Use max(organisation requirement, workflow floor): generic 5, joined-query/dashboard/causal 10 distinct people. Protect complementary cells and never publish individual rankings.

Start using

1. Clone the repository and keep the checkout

   The Skill runs the dashboard runner from this checkout, so keep it after installation. Note where it is saved, because the agent will ask for it as repo_root.

   git clone https://github.com/microsoft/viva-insights-sample-code.git

2. Install both Skills from that checkout

   Run these from the folder where you ran git clone, which contains the viva-insights-sample-code folder. Installing both from the same checkout keeps the dashboard Skill and the shared analysis Skill at the same revision.

   gh skill install ./viva-insights-sample-code viva-insights-analysis --from-local --agent github-copilot --scope user
   gh skill install ./viva-insights-sample-code viva-insights-copilot-dashboards --from-local --agent github-copilot --scope user

3. Load and check the Skills

   In GitHub Copilot CLI, run /skills reload and then /skills info viva-insights-copilot-dashboards. For another agent, change --agent in the install commands and restart the agent. The dashboards are rendered with R Markdown, so install R, Pandoc and the packages the agent lists before building.

4. Try a starter prompt

   Reproduce the Consumption demo using the dashboard Skill. My checkout of viva-insights-sample-code is at:

   Paste the full path to your checkout after the prompt. For your own data, describe the query export and ask the agent to inspect it first.

If gh skill is not available, or you use a different agent, follow the manual installation guide. Skills are not verified by GitHub, so review the installed files before use.

Overview

Questions this Skill helps you answer:

• Can I reproduce the Consumption or Developer Experience reference dashboard from the synthetic demo data?
• Does my Consumption query export contain what the dashboard needs, and which sections would be excluded?
• Can you regroup the dashboard by a different organisational attribute or change its reporting periods?

Results

What you can expect back:

• A self-contained HTML dashboard with its local location, the reproducible command and configuration, and a validation summary.
• A list of the supported and excluded sections, with the evidence limitations for each.
• For inspection or a blocked build, a readiness assessment with the missing evidence and the next supported action.

These previews use the synthetic demo data that ships with the starter kit.

Limits to keep in mind

• Requires a retained sample-code checkout and R; reference demo rendering also requires Pandoc. Skill installation does not install dependencies.
• Consumption real builds support scoped credits-only panels. GitHub real data is inspect/readiness only; real Person/Agent dashboard adapters are not implemented.
• Reference reproductions are synthetic. Do not invent missing panels, tokens, currency, productivity, ROI or causal effects from activity.
• Use max(organisation requirement, 10 distinct people) across published and embedded output, protect complementary cells and never publish individual rankings.

Start using

1. Install both Skills

   Each command installs the current version from the main branch, so run them one after the other. To guarantee a matching revision, or to let the agent follow links to repository prompts, clone the repository and install both Skills from that checkout with --from-local, as in the dashboard Skill steps.

   gh skill install microsoft/viva-insights-sample-code frontier-analytics/skills/viva-insights-analysis --agent github-copilot --scope user
   gh skill install microsoft/viva-insights-sample-code frontier-analytics/skills/viva-insights-causal-analysis --agent github-copilot --scope user

2. Load and check the Skills

   In GitHub Copilot CLI, run /skills reload and then /skills info viva-insights-causal-analysis. For another agent, change --agent in the install commands and restart the agent.

3. Try a starter prompt

   Assess whether my Person Query and quarterly outcome can support within-person analysis before running models.

   Add the exposure and outcome you have in mind, how often the outcome is measured, and whether you prefer R or Python.

If gh skill is not available, or you use a different agent, follow the manual installation guide. Skills are not verified by GitHub, so review the installed files before use.

Overview

Questions this Skill helps you answer:

• Is my outcome measured often enough to support a within-person comparison?
• Does higher Copilot usage go with a change in a collaboration metric for the same people over time?
• Does the relationship survive prior-outcome adjustment, timing placebos and sensitivity checks?

Results

What you can expect back:

• A readiness assessment that states whether the data can support the question and what is missing, without fitting models.
• After a supported analysis, reproducible scripts, audit and results records, and a self-contained HTML report.
• Model-labelled results with visible caveats and plain-language definitions.

Limits to keep in mind

• Guided design/readiness workflow, not a general real-data causal engine. Bundled R/Python runners use fabricated balanced panels and limited linear comparisons.
• Choose a supported identifying design and compatible diagnostics first; readiness-only and unavailable checks are valid. No general PPML, DML, bootstrap or quasi-experimental runner is implemented here.
• The prescribed Copilot Causal Toolkit remains a separate notebook/DML workflow; do not silently substitute treatments or methods.
• Install alongside the core analysis skill. Use max(organisation requirement, 10 distinct people), protect complementary cells and never publish individual rankings or person-specific effects.

Start using

1. Install both Skills

   Install the process and shared analysis Skills from the same revision. For guaranteed revision parity, use a local checkout with –from-local, as in the dashboard Skill steps. Review customised copies before replacing them.

   gh skill install microsoft/viva-insights-sample-code frontier-analytics/skills/viva-insights-analysis --agent github-copilot --scope user
   gh skill install microsoft/viva-insights-sample-code frontier-analytics/skills/viva-insights-process-analysis --agent github-copilot --scope user

2. Load and check the Skills

   In GitHub Copilot CLI, run /skills reload and then /skills info viva-insights-process-analysis. For another agent, change –agent in the install commands and restart the agent. The Python rehearsal uses the standard library. The matched R classifier checks require jsonlite.

3. Try a starter prompt

   Help me define and compare the collaboration footprint of our annual audit using my Meeting Query exports. Start with scope and coverage before reporting totals.

   Add local file locations, preferred R/Python runtime and the process/calendar context. This is business-process analysis, not Purview audit-log parsing.

If gh skill is not available, or you use a different agent, follow the manual installation guide. Skills are not verified by GitHub, so review the installed files before use.

Overview

Questions this Skill helps you answer:

• Which meetings count towards the process, and what needs approval?
• How do matched observed process phases compare across years?
• Is a change associated with event volume, time per meeting or organisational composition?
• Can we refresh the report while retaining its scope, evidence and manual edits?

Results

What you can expect back:

• An analysis contract with reviewable title/date rules, worked examples and open scope questions.
• Agent-guided aggregate evidence for volume, format, attribution and matched periods, with reconciliations and caveats.
• A fabricated audit rehearsal and evidence-led reporting guidance. Real-export adapters, statistical engines and report renderers are not bundled.

Limits to keep in mind

• Agent-guided collaboration analysis around business processes using Meeting Query and optional purpose-specific Person Query exports.
• Requires approved title/date scope and verified coverage before reporting process totals. Purview audit logs and process mining are separate.
• Bundled Python rehearsal and R classifier checks use fabricated audit events; no general export adapter, inference engine or report renderer is implemented.
• Install alongside the core skill. Use max(organisation requirement, 10 distinct people), verify the applicable person base and keep event stability thresholds separate.