From e0004d4ec3239a06d5f216eef9ae97d7a9e3a268 Mon Sep 17 00:00:00 2001 From: SandraBcna Date: Tue, 4 Aug 2026 17:28:56 +0100 Subject: [PATCH 1/4] Add Copilot Agent Advisor skill Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- submissions/copilot-agent-advisor/README.md | 40 ++++++ submissions/copilot-agent-advisor/SKILL.md | 125 ++++++++++++++++++ .../copilot-agent-advisor/metadata.json | 12 ++ 3 files changed, 177 insertions(+) create mode 100644 submissions/copilot-agent-advisor/README.md create mode 100644 submissions/copilot-agent-advisor/SKILL.md create mode 100644 submissions/copilot-agent-advisor/metadata.json diff --git a/submissions/copilot-agent-advisor/README.md b/submissions/copilot-agent-advisor/README.md new file mode 100644 index 00000000..8efce370 --- /dev/null +++ b/submissions/copilot-agent-advisor/README.md @@ -0,0 +1,40 @@ +# Copilot Agent Advisor + +Not sure whether a scenario needs a full custom agent, a lightweight standard agent, or nothing more than Microsoft 365 Copilot as it ships? This skill makes that call for you. + +Describe the scenario and it recommends **one of three options**: + +| Option | What it is | +|--------|-----------| +| **Use Microsoft 365 Copilot as-is** | The built-in Copilot experience over your Microsoft 365 data — no build. | +| **Standard agent (declarative)** | Copilot tailored with your instructions, knowledge, and actions, running on Copilot's own orchestrator and models, inside Microsoft 365 apps. | +| **Custom agent (custom engine)** | A fully custom agent — your own orchestrator and models, custom workflows, its own hosting, and reach beyond Microsoft 365. | + +It also advises **which Copilot Studio authoring experience** to use — the new (modern) instruction-based experience vs the classic topic/flow experience — and always flags that there's no migration path between the two. + +## How it works + +It weighs the factors that actually drive the decision: + +- Where the **knowledge/data** lives (Microsoft 365 vs external systems) +- Whether you need **your own or fine-tuned models** +- Whether you need **custom orchestration** or deterministic business logic +- **Channels** (inside Microsoft 365 only, or external too) +- **Autonomy** (user-initiated vs proactive/triggered) +- **Audience** (individual vs group collaboration) +- **Compliance** (inherit Microsoft 365's, or own it) +- **Speed & skill** (fastest low-code, or full pro-code control) + +Then it returns a clear recommendation with the reasoning tied to your specifics, a runner-up and the single factor that would flip the decision, how to build it, and the trade-offs (cost, hosting, compliance ownership). + +## Example + +> **You:** "We want a Teams helpdesk assistant that answers from our SharePoint IT policies. Which agent should I build?" +> +> **It recommends:** a standard (declarative) knowledge agent in Microsoft 365 Copilot Agent Builder (new experience) — because the knowledge already lives in Microsoft 365, it runs inside Teams, it's user-initiated, and it inherits Microsoft 365 compliance. Runner-up (custom engine agent) only if you later add external-system write-back, your own model, or proactive behavior. + +## Good to know + +- **It advises — it doesn't build.** It recommends the choice; it won't provision, configure, or deploy the agent. +- **It stays current.** Agent capabilities and previews move fast, so it verifies fast-moving specifics against Microsoft Learn before asserting them. +- Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio "classic vs new agent experience" documentation. diff --git a/submissions/copilot-agent-advisor/SKILL.md b/submissions/copilot-agent-advisor/SKILL.md new file mode 100644 index 00000000..ed07952c --- /dev/null +++ b/submissions/copilot-agent-advisor/SKILL.md @@ -0,0 +1,125 @@ +--- +name: copilot-agent-advisor +description: | + Recommends which Microsoft 365 / Copilot Studio agent option fits a described + scenario — using Microsoft 365 Copilot as-is, building a standard (declarative) + agent, or building a custom (custom engine) agent — and which Copilot Studio + authoring experience to use (the new/modern experience vs classic). Use when + the user asks "which agent type should I build", "declarative vs custom engine + agent", "standard agent or custom agent", "should I use Copilot Studio or + Microsoft 365 Copilot", "new agent experience vs classic", "help me choose a + Copilot agent", or describes an agent scenario and wants a recommendation. + Do NOT use to actually build, deploy, configure, or write code for an agent, + and do NOT use for non-agent product comparisons — this skill only advises on + the choice. For general capability questions about Microsoft 365 Copilot, + answer directly instead of invoking this skill. +cowork: + category: analysis + icon: Lightbulb +--- + +## Overview + +This skill helps a user pick the right way to deliver an AI experience in the +Microsoft 365 + Copilot Studio ecosystem. It gathers the handful of factors that +actually drive the decision (data sources, models, orchestration, channels, +autonomy, collaboration, compliance, and skill/speed), maps them to one of three +options, and returns a clear, reasoned recommendation with a runner-up and how to +build it. It also advises which Copilot Studio **authoring experience** (new vs +classic) to use. + +The three options it chooses between: + +| Option | What it is | Use when | +|--------|-----------|----------| +| **Use Microsoft 365 Copilot as-is** | The built-in Copilot experience over Microsoft Graph data, no build. | Built-in capabilities + your Microsoft 365 data already cover the need; you don't need extra knowledge, custom instructions, or actions. | +| **Standard agent (declarative agent)** | Copilot configured with your instructions, knowledge, and actions, running on **Copilot's own orchestrator and foundation models**, inside Microsoft 365 apps. | You want to tailor Copilot for a focused scenario, stay inside Microsoft 365 (Teams, Outlook, SharePoint, Word), inherit Microsoft 365 compliance/RAI, and ship fast with low-code or a streamlined pro-code path. | +| **Custom agent (custom engine agent)** | A fully custom agent where you **bring your own orchestrator and models**, with custom/complex workflows, its own hosting, and multi-channel reach. | You need custom orchestration, your own/fine-tuned/domain models, deterministic complex business logic, proactive/autonomous behavior, group collaboration, delivery outside Microsoft 365, or you're integrating an existing external bot. | + +## When to Use + +- The user describes an agent/automation scenario and asks what to build. +- The user asks to compare declarative vs custom engine agents, standard vs custom agents, or Copilot Studio vs plain Microsoft 365 Copilot. +- The user asks whether to author in the Copilot Studio new (modern) experience or the classic experience. +- The user is unsure whether they even need to build an agent at all. + +## When NOT to Use + +- The user wants to **actually build, configure, deploy, or write code** for an agent — this skill only advises on the choice, not the build. +- The user asks a general "what can Microsoft 365 Copilot do" capability question — answer that directly. +- The user asks for a non-agent product comparison (e.g. licensing, Power Automate vs Logic Apps) — out of scope. + +## Quick Start + +``` +User: "We want a Teams helpdesk assistant that answers from our SharePoint IT + policies and can open a ticket in ServiceNow. Which agent should I build?" +1. Gather the decision factors (Phase 1) — most are already stated; fill gaps + with ONE clarifying question if a load-bearing factor is missing. +2. Apply the decision logic (Phase 2). +3. Recommend (Phase 3): primary option + why, runner-up, how to build, caveats. +4. (Optional) If the platform can browse the web, verify any current product + detail against Microsoft Learn before asserting it. +``` + +## Core Instructions + +### Phase 1: Gather the decision factors + +Collect the answers below. Take whatever the user already gave you; if a +**load-bearing** factor is missing (one that would flip the recommendation), ask +for just those with a single clarifying question (offer multiple choice where the +platform supports it). Never ask about factors the user already answered. + +1. **Knowledge / data** — Is the needed knowledge in Microsoft 365 (SharePoint, OneDrive, Teams, Graph, Copilot connectors), or in external systems/APIs? +2. **Models** — Do you need your own, fine-tuned, small, or domain-specific/multimodal models, or is Copilot's foundation model fine? +3. **Orchestration / logic** — Do you need custom orchestration, complex multi-step workflows, precise business rules, or deterministic step-by-step control? +4. **Channels / reach** — Will it live only inside Microsoft 365 apps (Teams, Outlook, Word, Copilot Chat), or also outside (own website/app, other platforms)? +5. **Autonomy** — Is it purely user-initiated, or must it act proactively / on triggers without direct user input? +6. **Audience** — Individual use, or shared group collaboration (a Teams channel/meeting, many users on the same agent)? +7. **Compliance** — Is inheriting Microsoft 365 security/compliance/RAI enough, or must you manage your own compliance posture? +8. **Speed & skill** — Do you want the fastest low-code route, or do you have pro-code capacity (.NET/Python/JS, Semantic Kernel/LangChain) and need full control? +9. **Existing assets** — Is there already a conversational bot built outside Copilot that you want to bring into Microsoft 365? + +### Phase 2: Apply the decision logic + +Evaluate in order; the first block that clearly matches is the recommendation. + +- **Recommend "Use Microsoft 365 Copilot as-is"** when the built-in Copilot plus the user's Microsoft 365 data already covers the need and there's no requirement for custom instructions, extra knowledge sources, or actions. Don't recommend building an agent that adds no capability. +- **Recommend a Standard (declarative) agent** when the scenario is a **focused** one that can run on Copilot's orchestrator and foundation models, the knowledge lives in Microsoft 365 (or reachable via Copilot connectors/actions), the workflow stays **inside Microsoft 365 apps**, it's mostly individual/user-initiated, the user wants faster/low-code delivery, and inheriting Microsoft 365 compliance & RAI is acceptable. Example: an IT helpdesk agent that answers @mentions in Teams, or a SharePoint document-summarization agent. +- **Recommend a Custom (custom engine) agent** when **any** of these is true: custom orchestration or complex/deterministic business logic; your own/fine-tuned/domain/multimodal models; delivery **outside** Microsoft 365; proactive/autonomous (trigger-driven) behavior; group collaboration where many users share one agent in a channel/meeting; or integrating an existing external bot. Note this route typically needs **its own hosting** (e.g. Azure) at additional cost and you must ensure your own compliance/RAI/security. Example: a loan-approval agent with strict rules and multiple credit-check integrations. + +If the scenario straddles standard and custom, lead with the **lighter** option that still meets every hard requirement, and name the specific factor(s) that would push it to custom. + +### Phase 3: Which Copilot Studio authoring experience (new vs classic) + +When the recommendation involves building in Copilot Studio, also advise the experience: + +- **New (modern) experience — the default for new agents.** Authoring starts from a natural-language description of the agent's purpose; a single consolidated surface holds identity, knowledge, tools, skills, and settings; behavior is driven by **instructions and reasoning** rather than explicit topic flows; all agents use the **enhanced orchestration runtime** (deeper reasoning, better answers, especially over Microsoft 365 data); and Evaluate + Monitor tabs are built into the authoring loop. Recommend it when the user is creating a new agent, wants a simpler instruction-based model, or is primarily reasoning over organizational/Microsoft 365 data. +- **Classic experience.** Topic/flow/branching-based design with explicit nodes and a choice of classic vs generative orchestration modes. Recommend it only when the user needs **precise, deterministic control** over each conversation step, relies on mature features not yet in the new experience, or is maintaining/extending an agent already built in classic. +- **Always warn:** there is **no migration path** between the two experiences (different architectures/runtimes), so the choice should be deliberate. Existing classic agents keep working, and the user can switch experiences at any time to build new agents. + +### Phase 4: Recommend + +Deliver the output in the format below. Keep it decision-focused, not a product essay. + +## Output + +Respond inline (no file unless asked) with: + +1. **Recommendation** — one primary option, stated plainly in the first line. +2. **Why** — 2-4 bullets tying the recommendation to the user's specific factors (name them). +3. **Authoring experience** — new vs classic, with the no-migration caveat when relevant. +4. **Runner-up / when to reconsider** — the next-best option and the single factor that would flip the decision. +5. **How to build it** — the tooling path (low-code Copilot Studio / Agent Builder, or pro-code Microsoft 365 Agents Toolkit / Visual Studio Code, plus hosting note for custom engine agents). +6. **Caveats** — cost/hosting and compliance ownership for custom engine agents; that capabilities evolve and specifics should be confirmed on Microsoft Learn. + +Offer a quick comparison table (or a card, if the platform renders them) only when 3+ options/attributes make a table clearer than prose. + +## Guardrails + +- **Advise, don't build.** This skill recommends a choice; it does not create, configure, deploy, or write agent code. If the user then wants to build, hand off rather than pretending to provision anything. +- **Ground current specifics.** Product names, capabilities, and preview status change quickly. When the user needs authoritative or current detail (feature availability, pricing, preview vs GA), verify against Microsoft Learn (`learn.microsoft.com`) using whatever web-lookup capability the platform provides before asserting, and say when something is subject to change. If the platform can't browse the web, flag the detail as time-sensitive and point the user to Microsoft Learn. +- **Never fabricate** feature availability, limits, or pricing. If you're unsure, say so and point to Microsoft Learn rather than guessing. +- **Ask only load-bearing questions.** Only ask a clarifying question for a factor that would actually change the recommendation; otherwise state a reasonable assumption and proceed. +- **Explain trade-offs, don't oversell.** Flag the cost, hosting, and compliance-ownership burden of custom engine agents, and the scope limits of declarative agents, so the user chooses with eyes open. diff --git a/submissions/copilot-agent-advisor/metadata.json b/submissions/copilot-agent-advisor/metadata.json new file mode 100644 index 00000000..032d0eb7 --- /dev/null +++ b/submissions/copilot-agent-advisor/metadata.json @@ -0,0 +1,12 @@ +{ + "name": "Copilot Agent Advisor", + "description": "Tells you which Microsoft 365 / Copilot Studio agent option fits your scenario — use M365 Copilot as-is, build a standard (declarative) agent, or build a custom (custom engine) agent — and whether to author in Copilot Studio's new or classic experience.", + "platforms": ["Cowork", "Copilot Studio", "Scout"], + "tags": ["copilot-studio", "agents", "decision-support", "architecture", "advisor"], + "author": "Sandra Boucenna", + "authorUrl": "https://github.com/SandraBcna", + "authorGithub": "SandraBcna", + "version": "1.0.0", + "createdAt": "2026-08-04", + "updatedAt": "2026-08-04" +} From 2ec50e3441ffdfaf86dbda877d93f5901f402da7 Mon Sep 17 00:00:00 2001 From: SandraBcna Date: Mon, 17 Aug 2026 17:14:41 +0100 Subject: [PATCH 2/4] Update Copilot Agent Advisor to harness terminology Replace the new-vs-classic authoring-experience framing with Copilot Studio's current three-harness model (GitHub Copilot, standard, Copilot chat) per maintainer review and https://learn.microsoft.com/en-us/microsoft-copilot-studio/harnesses-overview. Express the caveat as agents not being transferable between the GitHub Copilot and standard harnesses. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- submissions/copilot-agent-advisor/README.md | 6 ++--- submissions/copilot-agent-advisor/SKILL.md | 23 ++++++++++--------- .../copilot-agent-advisor/metadata.json | 4 ++-- 3 files changed, 17 insertions(+), 16 deletions(-) diff --git a/submissions/copilot-agent-advisor/README.md b/submissions/copilot-agent-advisor/README.md index 8efce370..5ed84d52 100644 --- a/submissions/copilot-agent-advisor/README.md +++ b/submissions/copilot-agent-advisor/README.md @@ -10,7 +10,7 @@ Describe the scenario and it recommends **one of three options**: | **Standard agent (declarative)** | Copilot tailored with your instructions, knowledge, and actions, running on Copilot's own orchestrator and models, inside Microsoft 365 apps. | | **Custom agent (custom engine)** | A fully custom agent — your own orchestrator and models, custom workflows, its own hosting, and reach beyond Microsoft 365. | -It also advises **which Copilot Studio authoring experience** to use — the new (modern) instruction-based experience vs the classic topic/flow experience — and always flags that there's no migration path between the two. +It also advises **which Copilot Studio harness** to build on — the GitHub Copilot harness (reasoning-heavy, multi-step work), the standard harness (rule-based and predictable), or the Copilot chat harness (extending Microsoft 365 Copilot Chat) — and flags that agents aren't transferable between the GitHub Copilot and standard harnesses. ## How it works @@ -31,10 +31,10 @@ Then it returns a clear recommendation with the reasoning tied to your specifics > **You:** "We want a Teams helpdesk assistant that answers from our SharePoint IT policies. Which agent should I build?" > -> **It recommends:** a standard (declarative) knowledge agent in Microsoft 365 Copilot Agent Builder (new experience) — because the knowledge already lives in Microsoft 365, it runs inside Teams, it's user-initiated, and it inherits Microsoft 365 compliance. Runner-up (custom engine agent) only if you later add external-system write-back, your own model, or proactive behavior. +> **It recommends:** a standard (declarative) knowledge agent in Microsoft 365 Copilot Agent Builder — because the knowledge already lives in Microsoft 365, it runs inside Teams, it's user-initiated, and it inherits Microsoft 365 compliance. Runner-up (custom engine agent) only if you later add external-system write-back, your own model, or proactive behavior. ## Good to know - **It advises — it doesn't build.** It recommends the choice; it won't provision, configure, or deploy the agent. - **It stays current.** Agent capabilities and previews move fast, so it verifies fast-moving specifics against Microsoft Learn before asserting them. -- Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio "classic vs new agent experience" documentation. +- Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio harness overview (GitHub Copilot, standard, and Copilot chat harnesses). diff --git a/submissions/copilot-agent-advisor/SKILL.md b/submissions/copilot-agent-advisor/SKILL.md index ed07952c..7d522bbc 100644 --- a/submissions/copilot-agent-advisor/SKILL.md +++ b/submissions/copilot-agent-advisor/SKILL.md @@ -4,10 +4,10 @@ description: | Recommends which Microsoft 365 / Copilot Studio agent option fits a described scenario — using Microsoft 365 Copilot as-is, building a standard (declarative) agent, or building a custom (custom engine) agent — and which Copilot Studio - authoring experience to use (the new/modern experience vs classic). Use when + harness to build on (GitHub Copilot, standard, or Copilot chat). Use when the user asks "which agent type should I build", "declarative vs custom engine agent", "standard agent or custom agent", "should I use Copilot Studio or - Microsoft 365 Copilot", "new agent experience vs classic", "help me choose a + Microsoft 365 Copilot", "which Copilot Studio harness", "help me choose a Copilot agent", or describes an agent scenario and wants a recommendation. Do NOT use to actually build, deploy, configure, or write code for an agent, and do NOT use for non-agent product comparisons — this skill only advises on @@ -25,8 +25,8 @@ Microsoft 365 + Copilot Studio ecosystem. It gathers the handful of factors that actually drive the decision (data sources, models, orchestration, channels, autonomy, collaboration, compliance, and skill/speed), maps them to one of three options, and returns a clear, reasoned recommendation with a runner-up and how to -build it. It also advises which Copilot Studio **authoring experience** (new vs -classic) to use. +build it. It also advises which Copilot Studio **harness** (GitHub Copilot, +standard, or Copilot chat) to build on. The three options it chooses between: @@ -40,7 +40,7 @@ The three options it chooses between: - The user describes an agent/automation scenario and asks what to build. - The user asks to compare declarative vs custom engine agents, standard vs custom agents, or Copilot Studio vs plain Microsoft 365 Copilot. -- The user asks whether to author in the Copilot Studio new (modern) experience or the classic experience. +- The user asks which Copilot Studio harness to build on (GitHub Copilot, standard, or Copilot chat). - The user is unsure whether they even need to build an agent at all. ## When NOT to Use @@ -91,13 +91,14 @@ Evaluate in order; the first block that clearly matches is the recommendation. If the scenario straddles standard and custom, lead with the **lighter** option that still meets every hard requirement, and name the specific factor(s) that would push it to custom. -### Phase 3: Which Copilot Studio authoring experience (new vs classic) +### Phase 3: Which Copilot Studio harness to build on -When the recommendation involves building in Copilot Studio, also advise the experience: +When the recommendation involves building in Copilot Studio, also advise the **harness** — the runtime that sits between your agent and the model, deciding when to call the model, what components to send it, and which tools to invoke. Copilot Studio documents three: -- **New (modern) experience — the default for new agents.** Authoring starts from a natural-language description of the agent's purpose; a single consolidated surface holds identity, knowledge, tools, skills, and settings; behavior is driven by **instructions and reasoning** rather than explicit topic flows; all agents use the **enhanced orchestration runtime** (deeper reasoning, better answers, especially over Microsoft 365 data); and Evaluate + Monitor tabs are built into the authoring loop. Recommend it when the user is creating a new agent, wants a simpler instruction-based model, or is primarily reasoning over organizational/Microsoft 365 data. -- **Classic experience.** Topic/flow/branching-based design with explicit nodes and a choice of classic vs generative orchestration modes. Recommend it only when the user needs **precise, deterministic control** over each conversation step, relies on mature features not yet in the new experience, or is maintaining/extending an agent already built in classic. -- **Always warn:** there is **no migration path** between the two experiences (different architectures/runtimes), so the choice should be deliberate. Existing classic agents keep working, and the user can switch experiences at any time to build new agents. +- **GitHub Copilot harness — the most capable, for reasoning-heavy work.** Takes a goal, breaks it into steps, and calls tools across connectors, knowledge, MCP, and connected agents, adjusting when a step fails. Natively creates and edits Word, Excel, PowerPoint, and PDF files, supports **skills** and **memory**, and runs each task in a governed sandbox. Billed with **Copilot Credits**. Recommend it for complex, multi-step business processes that work across tools and files (e.g. an accounts-payable agent that reads invoices, matches purchase orders, and routes exceptions). +- **Standard harness — dependable and rule-based.** You define the topics, prompts, and paths so a structured, repeatable conversation or workflow responds **predictably**, drawing on your prompt library and enterprise knowledge. Billed via **standard-harness licensing**. Recommend it when the scenario is well-defined and you want consistent answers (e.g. an internal help-desk agent that answers common questions and routes simple requests). +- **Copilot chat harness — for extending Microsoft 365 Copilot Chat.** Connects your enterprise knowledge to M365 Copilot Chat so employees get grounded answers without leaving their everyday experience; publishes to internal teams. Billing is consumption-based or included in the Microsoft 365 Copilot user subscription. Recommend it when the priority is connecting people to information (e.g. an onboarding agent grounded on SharePoint knowledge). +- **Always warn:** agents aren't transferable between the **GitHub Copilot** and **standard** harnesses (different runtimes), and the harnesses bill differently, so the choice should be deliberate. You switch which harness you build on via the **New experience** toggle on the Copilot Studio home page. ### Phase 4: Recommend @@ -109,7 +110,7 @@ Respond inline (no file unless asked) with: 1. **Recommendation** — one primary option, stated plainly in the first line. 2. **Why** — 2-4 bullets tying the recommendation to the user's specific factors (name them). -3. **Authoring experience** — new vs classic, with the no-migration caveat when relevant. +3. **Harness** — GitHub Copilot, standard, or Copilot chat, with the not-transferable-between-harnesses caveat when relevant. 4. **Runner-up / when to reconsider** — the next-best option and the single factor that would flip the decision. 5. **How to build it** — the tooling path (low-code Copilot Studio / Agent Builder, or pro-code Microsoft 365 Agents Toolkit / Visual Studio Code, plus hosting note for custom engine agents). 6. **Caveats** — cost/hosting and compliance ownership for custom engine agents; that capabilities evolve and specifics should be confirmed on Microsoft Learn. diff --git a/submissions/copilot-agent-advisor/metadata.json b/submissions/copilot-agent-advisor/metadata.json index 032d0eb7..bc1dd7fb 100644 --- a/submissions/copilot-agent-advisor/metadata.json +++ b/submissions/copilot-agent-advisor/metadata.json @@ -1,6 +1,6 @@ { "name": "Copilot Agent Advisor", - "description": "Tells you which Microsoft 365 / Copilot Studio agent option fits your scenario — use M365 Copilot as-is, build a standard (declarative) agent, or build a custom (custom engine) agent — and whether to author in Copilot Studio's new or classic experience.", + "description": "Tells you which Microsoft 365 / Copilot Studio agent option fits your scenario — use M365 Copilot as-is, build a standard (declarative) agent, or build a custom (custom engine) agent — and which Copilot Studio harness (GitHub Copilot, standard, or Copilot chat) to build on.", "platforms": ["Cowork", "Copilot Studio", "Scout"], "tags": ["copilot-studio", "agents", "decision-support", "architecture", "advisor"], "author": "Sandra Boucenna", @@ -8,5 +8,5 @@ "authorGithub": "SandraBcna", "version": "1.0.0", "createdAt": "2026-08-04", - "updatedAt": "2026-08-04" + "updatedAt": "2026-08-17" } From 06638c38889e112ef4313d079c51c5c17498d690 Mon Sep 17 00:00:00 2001 From: SandraBcna Date: Mon, 17 Aug 2026 17:26:20 +0100 Subject: [PATCH 3/4] Add MS Learn harness reference, platform-compatibility note, and differentiation Cite the Copilot Studio harnesses-overview doc (per @adilei) plus switch-experiences and billing-licensing as References; confirm platform-neutral operation across Cowork, Copilot Studio, and Scout; add a depth boundary and 'how this differs' section so the skill complements the Harness Picker and AI Platform Advisor rather than duplicating them. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- submissions/copilot-agent-advisor/README.md | 11 ++++++++++- submissions/copilot-agent-advisor/SKILL.md | 15 ++++++++++++++- 2 files changed, 24 insertions(+), 2 deletions(-) diff --git a/submissions/copilot-agent-advisor/README.md b/submissions/copilot-agent-advisor/README.md index 5ed84d52..0787715b 100644 --- a/submissions/copilot-agent-advisor/README.md +++ b/submissions/copilot-agent-advisor/README.md @@ -37,4 +37,13 @@ Then it returns a clear recommendation with the reasoning tied to your specifics - **It advises — it doesn't build.** It recommends the choice; it won't provision, configure, or deploy the agent. - **It stays current.** Agent capabilities and previews move fast, so it verifies fast-moving specifics against Microsoft Learn before asserting them. -- Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio harness overview (GitHub Copilot, standard, and Copilot chat harnesses). +- **Works everywhere.** Platform-neutral — the same advisory logic runs in **Cowork**, **Copilot Studio**, and **Scout**, with no dependency on platform-specific tools. +- Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio [harnesses overview](https://learn.microsoft.com/en-us/microsoft-copilot-studio/harnesses-overview) (GitHub Copilot, standard, and Copilot chat harnesses). + +## How this differs from similar skills + +This skill owns the **earliest, broadest** decision — *should you build at all, and which of the three agent types?* — and treats the harness as one downstream step, so it complements rather than duplicates the more specialized skills: + +- **Copilot Studio Harness Picker** goes deep on *only* the harness choice (Copilot Credit / licensing estimates, channel-and-identity implementation checks) once you've already decided to build in Copilot Studio. This skill hands off to that depth instead of reproducing it. +- **Microsoft AI Platform Advisor** picks across the wider Microsoft AI platform (Foundry, Agent 365, etc.). This skill stays focused on the M365 Copilot / Copilot Studio agent-type decision. +- It's also the only one of these that runs on **Scout** in addition to Cowork and Copilot Studio. diff --git a/submissions/copilot-agent-advisor/SKILL.md b/submissions/copilot-agent-advisor/SKILL.md index 7d522bbc..39114062 100644 --- a/submissions/copilot-agent-advisor/SKILL.md +++ b/submissions/copilot-agent-advisor/SKILL.md @@ -93,12 +93,13 @@ If the scenario straddles standard and custom, lead with the **lighter** option ### Phase 3: Which Copilot Studio harness to build on -When the recommendation involves building in Copilot Studio, also advise the **harness** — the runtime that sits between your agent and the model, deciding when to call the model, what components to send it, and which tools to invoke. Copilot Studio documents three: +When the recommendation involves building in Copilot Studio, also advise the **harness** — the runtime that sits between your agent and the model, deciding when to call the model, what components to send it, and which tools to invoke. Copilot Studio documents three (see [References](#references) for the authoritative, current comparison): - **GitHub Copilot harness — the most capable, for reasoning-heavy work.** Takes a goal, breaks it into steps, and calls tools across connectors, knowledge, MCP, and connected agents, adjusting when a step fails. Natively creates and edits Word, Excel, PowerPoint, and PDF files, supports **skills** and **memory**, and runs each task in a governed sandbox. Billed with **Copilot Credits**. Recommend it for complex, multi-step business processes that work across tools and files (e.g. an accounts-payable agent that reads invoices, matches purchase orders, and routes exceptions). - **Standard harness — dependable and rule-based.** You define the topics, prompts, and paths so a structured, repeatable conversation or workflow responds **predictably**, drawing on your prompt library and enterprise knowledge. Billed via **standard-harness licensing**. Recommend it when the scenario is well-defined and you want consistent answers (e.g. an internal help-desk agent that answers common questions and routes simple requests). - **Copilot chat harness — for extending Microsoft 365 Copilot Chat.** Connects your enterprise knowledge to M365 Copilot Chat so employees get grounded answers without leaving their everyday experience; publishes to internal teams. Billing is consumption-based or included in the Microsoft 365 Copilot user subscription. Recommend it when the priority is connecting people to information (e.g. an onboarding agent grounded on SharePoint knowledge). - **Always warn:** agents aren't transferable between the **GitHub Copilot** and **standard** harnesses (different runtimes), and the harnesses bill differently, so the choice should be deliberate. You switch which harness you build on via the **New experience** toggle on the Copilot Studio home page. +- **Depth boundary.** Advise the harness at a *decision* level — which one and why. For detailed Copilot Credit / licensing estimates, capacity math, or channel-and-identity implementation checks, hand off to a dedicated harness-selection skill or to Microsoft Learn rather than computing them here. This skill's job is the earlier, broader call — *whether to build at all* and *which agent type* — with the harness as one downstream step. ### Phase 4: Recommend @@ -124,3 +125,15 @@ Offer a quick comparison table (or a card, if the platform renders them) only wh - **Never fabricate** feature availability, limits, or pricing. If you're unsure, say so and point to Microsoft Learn rather than guessing. - **Ask only load-bearing questions.** Only ask a clarifying question for a factor that would actually change the recommendation; otherwise state a reasonable assumption and proceed. - **Explain trade-offs, don't oversell.** Flag the cost, hosting, and compliance-ownership burden of custom engine agents, and the scope limits of declarative agents, so the user chooses with eyes open. + +## References + +Ground harness and agent-type claims on current Microsoft Learn documentation. Verify volatile details (previews, pricing, Copilot Credits) against these before asserting, and note the date checked when the platform can browse the web. + +- **Choose a harness (GitHub Copilot, standard, Copilot chat):** https://learn.microsoft.com/en-us/microsoft-copilot-studio/harnesses-overview +- **Access standard vs GitHub Copilot harness (switching, what changes):** https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/switch-experiences +- **Copilot Studio licensing & Copilot Credits:** https://learn.microsoft.com/en-us/microsoft-copilot-studio/billing-licensing + +## Platform compatibility + +This skill is platform-neutral — its logic is pure advisory reasoning with no dependency on any platform-specific tool. Clarifying questions, comparison tables, and optional web lookups are described as *capabilities* ("if the platform supports it / can browse the web"), so it runs the same in **Cowork**, **Copilot Studio**, and **Scout**. The `cowork:` frontmatter only supplies a card category/icon in Cowork and is safely ignored elsewhere. From aeee1c44ea7a3fafa27eacb6b8ac18ffbcfa6c93 Mon Sep 17 00:00:00 2001 From: SandraBcna Date: Mon, 17 Aug 2026 17:44:50 +0100 Subject: [PATCH 4/4] Tighten scope wording in README, SKILL.md, and PR Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- submissions/copilot-agent-advisor/README.md | 8 -------- submissions/copilot-agent-advisor/SKILL.md | 2 +- 2 files changed, 1 insertion(+), 9 deletions(-) diff --git a/submissions/copilot-agent-advisor/README.md b/submissions/copilot-agent-advisor/README.md index 0787715b..72dad115 100644 --- a/submissions/copilot-agent-advisor/README.md +++ b/submissions/copilot-agent-advisor/README.md @@ -39,11 +39,3 @@ Then it returns a clear recommendation with the reasoning tied to your specifics - **It stays current.** Agent capabilities and previews move fast, so it verifies fast-moving specifics against Microsoft Learn before asserting them. - **Works everywhere.** Platform-neutral — the same advisory logic runs in **Cowork**, **Copilot Studio**, and **Scout**, with no dependency on platform-specific tools. - Grounded in Microsoft Learn's declarative-vs-custom-engine decision guidance and the Copilot Studio [harnesses overview](https://learn.microsoft.com/en-us/microsoft-copilot-studio/harnesses-overview) (GitHub Copilot, standard, and Copilot chat harnesses). - -## How this differs from similar skills - -This skill owns the **earliest, broadest** decision — *should you build at all, and which of the three agent types?* — and treats the harness as one downstream step, so it complements rather than duplicates the more specialized skills: - -- **Copilot Studio Harness Picker** goes deep on *only* the harness choice (Copilot Credit / licensing estimates, channel-and-identity implementation checks) once you've already decided to build in Copilot Studio. This skill hands off to that depth instead of reproducing it. -- **Microsoft AI Platform Advisor** picks across the wider Microsoft AI platform (Foundry, Agent 365, etc.). This skill stays focused on the M365 Copilot / Copilot Studio agent-type decision. -- It's also the only one of these that runs on **Scout** in addition to Cowork and Copilot Studio. diff --git a/submissions/copilot-agent-advisor/SKILL.md b/submissions/copilot-agent-advisor/SKILL.md index 39114062..06f1182a 100644 --- a/submissions/copilot-agent-advisor/SKILL.md +++ b/submissions/copilot-agent-advisor/SKILL.md @@ -99,7 +99,7 @@ When the recommendation involves building in Copilot Studio, also advise the **h - **Standard harness — dependable and rule-based.** You define the topics, prompts, and paths so a structured, repeatable conversation or workflow responds **predictably**, drawing on your prompt library and enterprise knowledge. Billed via **standard-harness licensing**. Recommend it when the scenario is well-defined and you want consistent answers (e.g. an internal help-desk agent that answers common questions and routes simple requests). - **Copilot chat harness — for extending Microsoft 365 Copilot Chat.** Connects your enterprise knowledge to M365 Copilot Chat so employees get grounded answers without leaving their everyday experience; publishes to internal teams. Billing is consumption-based or included in the Microsoft 365 Copilot user subscription. Recommend it when the priority is connecting people to information (e.g. an onboarding agent grounded on SharePoint knowledge). - **Always warn:** agents aren't transferable between the **GitHub Copilot** and **standard** harnesses (different runtimes), and the harnesses bill differently, so the choice should be deliberate. You switch which harness you build on via the **New experience** toggle on the Copilot Studio home page. -- **Depth boundary.** Advise the harness at a *decision* level — which one and why. For detailed Copilot Credit / licensing estimates, capacity math, or channel-and-identity implementation checks, hand off to a dedicated harness-selection skill or to Microsoft Learn rather than computing them here. This skill's job is the earlier, broader call — *whether to build at all* and *which agent type* — with the harness as one downstream step. +- **Depth boundary.** Advise the harness at a *decision* level — which one and why. For detailed Copilot Credit / licensing estimates, capacity math, or channel-and-identity implementation checks, hand off to a dedicated harness-selection skill or to Microsoft Learn rather than computing them here. This skill focuses on the choice of whether to build and which agent type, with the harness as one downstream step. ### Phase 4: Recommend