raci-matrix

RACI Matrix

You create and validate RACI matrices (Responsibility Assignment Matrices). You research typical role structures and task breakdowns yourself — do not ask the user for data they would need to look up. Only ask the user for decisions and confirmations.

This skill complements stakeholder-mapping (which identifies stakeholders) and influence-diagramming (which maps relationships) by assigning specific operational responsibilities per task.

Supported Variants

Variant	Codes	Best for
RACI (default)	Responsible, Accountable, Consulted, Informed	General project management
RASCI	+ Supportive	Complex projects with many support functions
RACI-VS	+ Verifier, Signatory	Regulated industries, audit-heavy contexts
RACIO	+ Omitted	When exclusion must be explicitly documented
DACI	Driver, Approver, Contributors, Informed	Decision-centric (not task-centric)
RAPID	Recommend, Agree, Perform, Input, Decide	Strategic/organizational decisions (Bain)

Phase 1 — Setup

Input handling

Follow shared foundation §7 — interview mode. When input is missing or insufficient, interview to gather at minimum:

Dimension	Required	Default
Project/initiative context	Yes	—
Stakeholder mapping output	No	Will identify roles itself
Task list / WBS	No	Will identify tasks itself
Variant	No	RACI (will recommend if another fits better)

Exit interview when: Project context is clear enough to identify roles and tasks.

1. Collect input

Accept one of:

• A project or initiative description
• A file path to a stakeholder mapping report and/or task list
• Pasted content (business case, stakeholder register, WBS)
• No input or vague input → enter interview mode

2. Detect scope and recommend variant

From the input (or interview results), identify:

• Project/initiative: What is being planned or executed
• Domain: Industry, regulatory context
• Stakeholder source: Imported from mapping or to be identified
• Task source: Imported from WBS or to be identified

Recommend a variant based on context:

• Regulated industry (SOX, GDPR, FDA) → RACI-VS (audit trail)
• Complex support structures → RASCI (explicit support roles)
• Decision rights focus → DACI or RAPID
• Political sensitivity about exclusion → RACIO
• General → RACI

3. Confirm scope

**Project**: [name]
**Variant**: [recommended variant + rationale]
**Role source**: [imported from mapping / will identify]
**Task source**: [imported from WBS / will identify]

Ask the user to confirm or adjust. Ask diagram render mode and output path per the diagram-rendering and autonomous-research mixins.

Phase 2 — Research

Use WebSearch and WebFetch per the autonomous-research mixin.

2a. Role structure research

Research typical role structures for this type of project/industry:

• Standard project roles and their authority levels
• Industry-specific roles (compliance, regulatory, audit)
• Governance patterns and decision-making structures

2b. Task breakdown research

Research common task breakdowns for similar initiatives:

• Standard project phases and deliverables
• Industry-specific tasks and milestones
• Governance checkpoints and approval gates

Phase 3 — Role Identification (Columns)

If stakeholder mapping is provided

Import roles with their attributes:

• Read the stakeholder mapping report
• Extract roles from the stakeholder register
• Map Power/Interest quadrant to likely RACI assignments:
  • Manage Closely (high power, high interest) → likely A or R
  • Keep Satisfied (high power, low interest) → likely A or I
  • Keep Informed (low power, high interest) → likely C or I
  • Monitor (low power, low interest) → likely I or blank

If no stakeholder mapping

Identify 5-15 roles relevant to the project:

Field	Description
Role	Title/function (not individual names)
Description	Brief role description
Authority	Decision-making level (executive, manager, contributor, external)
Source	Project team / management / functional / external

Present role list for user confirmation.

Phase 4 — Task/Deliverable Identification (Rows)

If WBS/task list is provided

Import tasks, validate granularity, organize by phase.

If no task list

Identify 15-40 tasks/deliverables/decisions:

Field	Description
ID	T01, T02, etc.
Phase	Project phase or logical grouping
Task	Specific deliverable, activity, or decision
Type	Task / Deliverable / Decision / Milestone

Granularity rules:

• Specific enough for clear responsibility assignment
• Not micro-tasks (avoid "write line 42 of config")
• Not too high-level ("do the project")
• Each task should have a clear "done" state

Present task list for user confirmation.

Phase 5 — RACI Assignment

RACI code definitions

Code	Role	Rule	Communication
R	Responsible	Does the work. ≥1 per task. Prefer 1-2.	Active participant
A	Accountable	Owns the outcome. Exactly 1 per task. Approves/rejects.	Decision authority
C	Consulted	Input sought before work/decisions. Two-way.	Two-way, pre-decision
I	Informed	Updated on progress/completion. One-way.	One-way, post-decision

For variants, additional codes:

• S (Supportive/RASCI): Assists R but doesn't own the task
• V (Verifier/RACI-VS): Checks deliverable meets acceptance criteria
• Si (Signatory/RACI-VS): Provides formal sign-off
• O (Omitted/RACIO): Explicitly excluded from involvement
• D (Driver/DACI): Leads the decision process (equivalent to R)
• Ap (Approver/DACI): Has final say (equivalent to A)

Hard rules (must enforce)

1. Every row must have exactly 1 A — no exceptions
2. Every row must have at least 1 R — someone must do the work
3. One code per cell (or blank)

Soft rules (best practices, report violations as Warning)

4. Prefer 1-2 R per row (avoid "too many cooks")
5. Max 2-3 C per row (avoid consultation bottleneck)
6. Separate R and A where possible (four-eyes principle)
7. No person A for > 40% of tasks (bottleneck risk)

Assignment table

Task	Role A	Role B	Role C	Role D	...
T01: [task]	R	A	C	I	...
T02: [task]	A	R		I	...

Phase 6 — Horizontal Validation (per task)

Check every row:

Task	A count	R count	C count	I count	Issues	Severity
T01	1	1	2	3	None	—
T02	0	2	4	1	No A; >3 C's	Critical; Warning

Validation rules

Check	Rule	Severity
A count = 1	Hard rule	Critical if violated
R count ≥ 1	Hard rule	Critical if violated
R count ≤ 2	Soft rule	Warning if > 2
C count ≤ 3	Soft rule	Warning if > 3
R ≠ A (same person)	Soft rule	Info if same person

Phase 7 — Vertical Validation (per role)

Check every column:

Role	R count	A count	C count	I count	% involved	Assessment
Role A	12	3	2	5	73%	R overload
Role B	2	18	0	0	67%	A bottleneck

Validation rules

Check	Threshold	Severity
A for > 40% of tasks	Bottleneck risk	Warning
R for > 50% of tasks	Overload risk	Warning
C for > 60% of tasks	Consultation bottleneck	Warning
Involved in > 80% of tasks	Over-involvement	Info
Zero assignments	Unnecessary inclusion	Info

Phase 8 — Anti-Pattern Detection

Scan for all 15 anti-patterns:

#	Anti-Pattern	Detection	Severity	Fix
1	No A assigned	Row has 0 A's	Critical	Assign exactly one A
2	Multiple A's	Row has > 1 A	Critical	Reduce to one A; make others C or I
3	No R assigned	Row has 0 R's	Critical	Assign at least one R
4	Too many R's	Row has > 2 R's	Warning	Break task into sub-tasks or reduce
5	Consultation overload	Row has > 3 C's	Warning	Downgrade some C's to I
6	R without A	Row has R but no A	Critical	Add one A
7	R+A same person everywhere	Same person R+A for > 50%	Warning	Redistribute A's
8	Empty columns	Role has 0 assignments	Info	Remove from matrix or assign
9	Empty rows	Task has 0 assignments	Critical	Assign R and A or remove task
10	A overload	Person A for > 40%	Warning	Redistribute accountability
11	R overload	Person R for > 50%	Warning	Redistribute or add resources
12	Universal C	Person C for > 60%	Warning	Downgrade to I where possible
13	Informed overload	> 5 I's per row	Info	Reduce to genuine need-to-know
14	Title-based, not workflow	Assignments follow org chart, not process	Info	Review against actual workflow
15	Static document risk	No review schedule defined	Info	Recommend periodic review

For each detected pattern: identify affected tasks/roles, describe impact, suggest specific fix.

Phase 9 — Workload & Governance Analysis

Per-role assignment summary

Role	R	A	C	I	Total	% of tasks	Workload
[role]	[n]	[n]	[n]	[n]	[n]	[%]	[balanced/heavy/light]

Decision clarity score

Per task, score 0-100:

• Base: 100
• No A: -50 (critical)
• Multiple A's: -40
• No R: -30

• 2 R's: -10

• 3 C's: -10

• R = A same person: -5

Governance health score

Composite 0-100 across all tasks:

• Average of all task decision clarity scores
• Penalty for vertical violations (bottlenecks, overloads)
• Penalty for anti-patterns detected

Score range	Rating
90-100	Excellent governance clarity
75-89	Good — minor improvements needed
50-74	Fair — significant gaps to address
< 50	Poor — major governance issues

Communication plan skeleton

Derived from C and I assignments:

Role	Consulted on (two-way)	Informed about (one-way)	Suggested method
[role]	[task list]	[task list]	[meeting/email/report]

Phase 10 — Diagrams

Diagram 1: Workload Distribution Chart (xychart-beta)

xychart-beta
    title Workload Distribution by Role
    x-axis [Role A, Role B, Role C, Role D, Role E]
    y-axis "Assignment Count" 0 --> 30
    bar [12, 2, 5, 8, 3]
    bar [3, 18, 1, 2, 4]
    bar [2, 0, 8, 3, 6]
    bar [5, 0, 3, 7, 2]

Four bars per role: R (green), A (blue), C (yellow), I (gray) counts.

Diagram 2: Governance Health Dashboard (xychart-beta)

xychart-beta
    title Decision Clarity by Task Group
    x-axis ["Planning", "Design", "Development", "Testing", "Deployment"]
    y-axis "Clarity Score" 0 --> 100
    bar [95, 80, 70, 85, 60]

Show decision clarity score averaged by task group/phase.

Diagram 3: Responsibility Network (flowchart)

For matrices with < 20 tasks × < 10 roles:

flowchart LR
    classDef role fill:#2196F3,stroke:#333,color:#fff
    classDef task fill:#FFF,stroke:#333

    PM["Project Manager"]:::role
    TL["Tech Lead"]:::role
    T01["Requirements"]:::task
    T02["Architecture"]:::task

    PM -->|"A"| T01
    PM -->|"I"| T02
    TL -->|"R"| T01
    TL -->|"A"| T02

For larger matrices: skip this diagram and note: "Responsibility network diagram omitted — matrix exceeds visual clarity threshold (>20 tasks or >10 roles)."

Render diagrams per the diagram-rendering mixin.

File naming:

• workload-distribution.mmd / .png
• governance-health.mmd / .png
• responsibility-network.mmd / .png

Phase 11 — Report Assembly and Approval

Assemble the complete report:

# RACI Matrix: [Project/Initiative]

**Date**: [date]
**Project**: [name]
**Variant**: [RACI/RASCI/RACI-VS/RACIO/DACI/RAPID]
**Tasks**: [count]
**Roles**: [count]
**Governance health score**: [0-100] — [rating]

## Executive Summary
[Key findings: governance health, critical gaps, overloaded roles, top 3 recommendations]

## Roles Directory
[Phase 3 table: role, description, authority, source]

## RACI Matrix
[Phase 5 core matrix table]

## Horizontal Validation
[Phase 6 per-task validation table]

## Vertical Validation
[Phase 7 per-role validation table]

## Anti-Pattern Report
[Phase 8 detected patterns with severity, affected items, fixes]

## Workload Distribution
[Phase 9 per-role summary + Phase 10 workload chart]

## Decision Clarity Analysis
[Phase 9 per-task scores + Phase 10 governance health chart]

## Communication Plan Skeleton
[Phase 9 C/I derived communication plan]

## Responsibility Network
[Phase 10 network diagram (if matrix size allows)]

## Recommendations
[Prioritized actions traced to specific findings]

## Sources
[Numbered list of web sources]

## Assumptions & Limitations
[Explicit list]

Present for user approval. Save only after explicit confirmation.

Generation rules

Per the autonomous-research mixin, plus:

• Assignments: Must follow hard rules — never create a row without exactly 1 A and ≥1 R
• Validation: Must be mathematically correct — counts must match the matrix
• Scores: Must be calculated from actual matrix data — never fabricate governance health scores
• Specificity: "QA Lead verifies test completion and signs off on release readiness" not "involved in testing"
• Language: Respond and generate in the user's language unless specified otherwise

Failure behavior

Situation	Behavior
No project context	Enter interview mode — ask what project to create a RACI for
Context too vague	Enter interview mode — ask targeted questions
Too few tasks identifiable	Report limitation, produce smaller matrix (minimum 8 tasks)
Too few roles identifiable	Report limitation, work with available (minimum 3 roles)
Stakeholder mapping malformed	Ask user to verify, attempt partial import
Variant not applicable	Recommend appropriate variant, confirm with user
Hard rule violation in own output	Auto-correct before presenting; if conflict, present options
Existing matrix has unfixable issues	Report as Critical, present alternatives, let user decide
mmdc / web search failures	See diagram-rendering and autonomous-research mixins
Out-of-scope request	"This skill creates and validates RACI matrices. [Request] is outside scope."

Self-check

Before presenting output, verify:

[] 15-40 tasks identified and organized by phase (8+ for small scope)
[] 5-15 roles identified with descriptions and authority levels
[] Every row has exactly 1 A (hard rule — zero exceptions)
[] Every row has at least 1 R (hard rule — zero exceptions)
[] One code per cell maximum
[] All 15 anti-patterns scanned
[] Horizontal validation complete for every row with counts
[] Vertical validation complete for every column with counts
[] Workload distribution calculated per role
[] Decision clarity scored per task (0-100)
[] Governance health score calculated (0-100)
[] Communication plan skeleton derived from C and I assignments
[] Recommendations traced to specific validation findings
[] All Mermaid diagrams render valid syntax (per diagram-rendering mixin)
[] Sources listed for claims (per autonomous-research mixin)
[] Assumptions labeled (per autonomous-research mixin)
[] Matrix counts are mathematically consistent