[New Mission] Operative NextGen: Author agent instructions
Summary
Author the mission where learners turn the baseline Hiring Agent into a controlled, testable orchestrator. Learners write its operating instructions, create the first focused skill, add Dataverse through MCP, and run the first evaluation.
This is the modern replacement for classic topic and flow-based behavior: instructions and skills define behavior, MCP exposes governed Dataverse capabilities, and Evaluate provides repeatable evidence. Duplicate checks found no issue proposing this consolidated mission.
Proposed mission
| Field |
Value |
| Section |
Operative NextGen |
| Folder |
docs/operative-v2/02-agent-instructions/ |
| Title |
Mission 02: Authoring Agent Instructions |
| Operation codename |
OPERATION SECRET DIRECTIVE |
| Difficulty |
2 |
| Time |
45 minutes |
Scenario
The Hiring Agent exists but lacks a reliable operating brief. The learner defines its role, process, constraints, delegation rules, and response expectations, then packages the first specialized behavior as a skill and proves the baseline with an evaluation.
Use an operations-manual analogy: instructions define standing orders, skills provide reusable procedures, and evaluations verify that the agent follows them.
Mission objectives
- Write clear, prioritized Hiring Agent instructions.
- Distinguish agent instructions, skill instructions, and tool descriptions.
- Create the first Hiring Agent skill.
- Add and authorize the Dataverse MCP server.
- Build and run a baseline evaluation.
Prerequisites to state in the mission
- Completion of Mission 01.
- Access to the Operative solution and seeded Dataverse records.
- Permission to add MCP tools and create connections.
- Access to the modern Evaluate experience.
Proposed lab outline
Author the operating brief
Write instructions covering role, supported hiring tasks, process order, data handling, delegation, response style, fairness, human review, and escalation. Test ambiguous and out-of-scope requests.
Create the first skill
Create a small reusable skill for a bounded hiring task, with a kebab-case name, selection-oriented description, and deterministic procedural instructions. Explain that skills replace appropriate classic topic behavior but do not own tools.
Add Dataverse MCP
Add the supported Dataverse MCP server, create the connection, review exposed tools and permissions, and test a read-only query against seeded Job Roles. Do not substitute a classic Power Automate flow.
Run the first evaluation
Create a small test set covering a supported request, an out-of-scope request, a safety boundary, and a Dataverse query. Run the evaluation, inspect failures, refine instructions, and rerun to establish a baseline.
Acceptance criteria
- Provide complete copyable Hiring Agent instructions with explicit priorities and no contradictions.
- Explain when to use agent instructions versus a skill or tool description.
- Create and behavior-test one valid skill.
- Use the currently supported Dataverse MCP onboarding and connection flow.
- Document exact exposed tool names only after live validation.
- Use least-privilege access and avoid destructive Dataverse tests.
- Create a versioned baseline test set with expected outcomes.
- Run, inspect, refine, and rerun the evaluation; record the baseline result.
- Do not use classic Topics or Power Automate flows to implement this mission.
- Preserve
<mission-meta /> and the Operative Mission 02 analytics tag.
- Follow
WRITING_STYLE.md; run validation, Markdown linting, and npm run docs:build.
Out of scope
- Full resume extraction or record creation.
- Connected agents and multi-agent orchestration.
- Model selection and comprehensive safety configuration.
- Production-scale evaluation suites.
Risks
- Dataverse MCP availability and tool names can vary by tenant rollout.
- Broad MCP permissions can expose more data or actions than the mission requires.
- Weak skill descriptions can cause selection overlap.
- Evaluation output can vary across model versions.
References
[New Mission] Operative NextGen: Author agent instructions
Summary
Author the mission where learners turn the baseline Hiring Agent into a controlled, testable orchestrator. Learners write its operating instructions, create the first focused skill, add Dataverse through MCP, and run the first evaluation.
This is the modern replacement for classic topic and flow-based behavior: instructions and skills define behavior, MCP exposes governed Dataverse capabilities, and Evaluate provides repeatable evidence. Duplicate checks found no issue proposing this consolidated mission.
Proposed mission
docs/operative-v2/02-agent-instructions/OPERATION SECRET DIRECTIVEScenario
The Hiring Agent exists but lacks a reliable operating brief. The learner defines its role, process, constraints, delegation rules, and response expectations, then packages the first specialized behavior as a skill and proves the baseline with an evaluation.
Use an operations-manual analogy: instructions define standing orders, skills provide reusable procedures, and evaluations verify that the agent follows them.
Mission objectives
Prerequisites to state in the mission
Proposed lab outline
Author the operating brief
Write instructions covering role, supported hiring tasks, process order, data handling, delegation, response style, fairness, human review, and escalation. Test ambiguous and out-of-scope requests.
Create the first skill
Create a small reusable skill for a bounded hiring task, with a kebab-case name, selection-oriented description, and deterministic procedural instructions. Explain that skills replace appropriate classic topic behavior but do not own tools.
Add Dataverse MCP
Add the supported Dataverse MCP server, create the connection, review exposed tools and permissions, and test a read-only query against seeded Job Roles. Do not substitute a classic Power Automate flow.
Run the first evaluation
Create a small test set covering a supported request, an out-of-scope request, a safety boundary, and a Dataverse query. Run the evaluation, inspect failures, refine instructions, and rerun to establish a baseline.
Acceptance criteria
<mission-meta />and the Operative Mission 02 analytics tag.WRITING_STYLE.md; run validation, Markdown linting, andnpm run docs:build.Out of scope
Risks
References