Rai Planner

From the author

RAI Planner identity protocol and on-demand reference pack covering Phase 1 capture, Phase 2 risk classification and the optional Mural board bootstrap, Phase 5 impact assessment, and Phase 6 review and backlog handoff.

— the description by Microsoft

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

RAI Planner provides a modular framework for managing Responsible AI assessments. It uses a phased approach—ranging from initial capture and risk classification to impact assessment and final backlog handoff—to guide an agent through the lifecycle of an RAI review. By utilizing on-demand reference files, the skill ensures the agent maintains a compact state while accessing specific guidance, templates, and rubrics only when required for the current phase of the project.

What it does

Defines a formal identity and protocol for RAI assessment sessions.
Provides phase-specific guidance for scoping, risk classification, and impact assessment.
Supports the creation of Mural boards for collaborative team planning.
Includes a review rubric to ensure assessment quality before backlog generation.
Facilitates the generation of structured handoff summaries and work items.
Maintains session state and evidence registers across multiple project phases.

When to use it

At the start of any session to establish the RAI Planner identity and protocol.
When transitioning between project phases to load relevant coaching and assessment logic.
When performing risk classification to apply prohibited-use gates and depth-tier logic.
When conducting an impact assessment to build evidence registers and tradeoff logs.
When finalizing an assessment to generate a structured backlog handoff.

Advantages

Modular design keeps the agent's active context compact and efficient.
Ensures consistent application of RAI standards across different projects.
Provides clear exit criteria for each phase of the assessment.
Supports evidence-led, exploratory questioning techniques.
Standardizes the final output format for downstream review and implementation.

Usage examples

“Start a new RAI assessment for our customer support chatbot.”: The agent adopts the RAI Planner identity, loads the session start protocol, and begins Phase 1 by asking about the system's purpose and deployment model.
“We have finished the initial scoping. What is next?”: The agent recognizes the transition to Phase 2, loads the risk classification reference, and begins applying the prohibited-use gate and indicator assessment.
“Can we create a Mural board for this project?”: The agent loads the Mural bootstrap reference and offers to set up a collaborative board to visualize the current risk assessment.
“Generate the final handoff for the project.”: The agent enters Phase 6, applies the review rubric to the collected evidence, and produces a structured summary table of all project artifacts.

Install

Download the skill as a ZIP:
https://autogineer.com/agent-skills/microsoft/rai-planner/download

Unzip it into the cross-client skills folder of your project:

mkdir -p .agents/skills && unzip rai-planner.zip -d .agents/skills

Documentation by Microsoft

Overview

Use this skill as an on-demand reference pack for the RAI Planner. The main skill body stays compact, and the detailed phase guidance lives in the reference files under this folder so the agent can load only the material needed for the current phase.

Phase Reference Index

Session start — Planner identity: rai-identity.md — read this at the start of every session. It holds the RAI identity, the disclaimer and attribution protocol, the six-phase definitions, entry modes, state management, question cadence, resume protocol, error handling, the user-supplied reference content protocol, and the research activation contract.
Phase 1 — Capture and scoping: capture-coaching.md — read this when entering Phase 1 and again during capture turns when you need the exploration-first questioning techniques.
Phase 2 — Risk classification: risk-classification.md — read this when entering Phase 2 to apply the prohibited-use gate, indicator assessment, and depth-tier logic.
Phase 2 (optional) — Mural team board: mural-board-bootstrap.md — read this only when the user accepts the Phase 2 board offer or asks for a Mural board, and before any mural command.
Phase 5 — Impact assessment: impact-assessment.md — read this when entering Phase 5 to build the evidence register, tradeoff log, and work-item seeds.
Phase 6 — Review rubric and backlog handoff: backlog-handoff.md — read this when entering Phase 6 to run the review rubric and generate the final handoff summary.

Usage Notes

Read rai-identity.md at session start. Load the phase references on demand at phase boundaries rather than as fixed startup context.
Use the referenced file that matches the phase you are about to enter or the section you need to re-open during execution.
Keep the detailed rubric, templates, and evidence guidance in the reference files rather than in the main skill body.

Files

ATTRIBUTION.mdAdded by AutoGineer835 B
LICENSELicence file added by AutoGineer1.1 KiB
SKILL.md2.5 KiB
references/backlog-handoff.md6.4 KiB
references/capture-coaching.md3.4 KiB
references/impact-assessment.md1.3 KiB
references/mural-board-bootstrap.md5.5 KiB
references/rai-identity.md38.4 KiB
references/risk-classification.md1.2 KiB
references/security-model.md2.3 KiB

references/backlog-handoff.md

description: Review and backlog handoff guidance for Phase 6 of the RAI Planner

RAI Review and Backlog Handoff

Use this note when entering Phase 6 or preparing the final handoff summary.

Review Rubric

Before generating backlog items, confirm that the assessment has covered the essential review dimensions:

scope boundary clarity
risk identification coverage
control surface adequacy
evidence sufficiency
tradeoff documentation
alignment with the selected framework

Work Item Generation

Generate work items from the evidence register and maturity observations using the appropriate backlog format. Keep the output concise, attributable, and suitable for downstream review.

Delegation to Shared Backlog Templates

For the full dual-format ADO and GitHub templates, content sanitization guidance, autonomy-tier vocabulary, disclaimer placement, and work-item ID naming rules, load the backlog-templates skill. This handoff reference stays focused on the RAI-specific review expectations and the final handoff decisions.

Content sanitization: no secrets, credentials, internal URLs, or PII in work item content. Apply this alongside the backlog-templates sanitization rules, which cover paths, state JSON, and standards identifiers but not these values.

Autonomy and Output Targets

Select the output target and autonomy tier that fit the project context. Persist the choice in session state and allow the user to confirm the final handoff.

Final Handoff Summary

When presenting the final handoff to the user, render the produced artifacts as a descriptive table rather than a flat list of filenames. The phase deliverables are sections of the single consolidated rai-plan.md; group rows by the phase that produced each section, link to the real relative path with the section anchor under the project slug folder, and give each row a short "what it contains" description so the user knows why to open it. The backlog-handoff files and the signing manifest remain separate files.

Apply these rules to the artifacts table:

Use relative links to the actual files, never absolute or editor-shell paths. Do not wrap links in backticks.
Keep one row per deliverable with columns for phase, artifact, and a concise description of the contents.
Replace bracketed placeholders with the values from session state; omit rows for artifacts the session did not produce.

Phase	Artifact	What it contains
1 — Scoping	assessment-content.md	Required template-aligned content whenever a document or Mural template was supplied
1 — Scoping	rai-plan.md — System Definition	System boundary, components, data flows, and intended use
1 — Scoping	rai-plan.md — Stakeholder Impact	Affected stakeholders and how each is impacted
2 — Risk	rai-plan.md — Risk Classification Screening	Risk indicators and confirmed assessment depth
3 — Standards	rai-plan.md — Standards Mapping	Framework mapping across the selected trustworthiness characteristics
4 — Threats	rai-plan.md — Threat Addendum	Identified threats with IDs and the high-concern subset
5 — Impact	rai-plan.md — Control Surface Catalog	Controls and their coverage across the framework functions
5 — Impact	rai-plan.md — Evidence Register	Evidence backing each control claim
5 — Impact	rai-plan.md — Tradeoffs	Tradeoffs needing human adjudication
6 — Handoff	rai-plan.md — Review Summary	Posture, gaps, and the human-review gate
6 — Handoff	ado-backlog-handoff.md	Drafted backlog items in ADO format (present when the target system includes ADO), ready after review
6 — Handoff	github-backlog-handoff.md	Drafted backlog items in GitHub format (present when the target system includes GitHub), ready after review
6 — Handoff	artifact-manifest.json	SHA-256 artifact manifest (present when signing was accepted)

Content Hygiene

Keep the handoff clear and reviewable:

preserve RAI characteristic names and framework references
avoid speculative claims that are not supported by the session evidence
note when a recommendation needs human review or compliance validation

Artifact Signing

After backlog generation and before broader distribution, the RAI planner may sign session artifacts. Use npm run rai:sign -- -ProjectSlug {slug}. The signing workflow produces a SHA-256 manifest for the generated artifacts, optionally signs them with cosign when the environment is configured for it, and writes artifact-manifest.json for the project slug. Reference the manifest in the handoff summary and retain it with the assessment artifacts.

Requirements and compatibility

Written for: GitHub Copilot, Claude Code

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Claude Code frontmatter

Security

No issues found by our static scan· No issues found

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Grilling
Matt Pocock· Community
A rigorous stress-testing persona that maps decisions into a branching tree to uncover hidden assumptions.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository317.5k
Wayfinder
Matt Pocock· Community
A structured approach to planning and executing complex, multi-session projects using a shared map of decision-based tickets on an issue tracker.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository317.5k
Brainstorming
Jesse Vincent (obra)· Community
A mandatory pre-work framework for clarifying user intent, exploring design approaches, and securing approval before implementation.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
MITLow riskHas scriptsGitHub stars of the source repository314k
Discernment Nudge
AnthropicOfficial
A framework for AI agents to append reflective follow-up questions to substantive, high-stakes responses.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
Apache-2.0GitHub stars of the source repository188.6k
Interview Me
Addy Osmani· Community
A structured interview process to extract and verify a user's true intent before starting any work.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository113.9k
Idea Refine
Addy Osmani· Community
A structured ideation partner that refines raw concepts into actionable markdown one-pagers.
Agent toolingClaude CodeClaude appsOpenAI Codex+14 more
MITLow riskHas scriptsGitHub stars of the source repository113.9k

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download rai-planner.zip