Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when clarifying who does what across roles for a project, process, or initiative: lay responsibilities out as a RACI matrix — activities as rows, roles as columns, cells holding only R, A, C, or I, with exactly one A per activity — which cheaper models do not do by default. Do NOT use for org charts, reporting lines, project timelines, or staffing and capacity plans.
.claude/skills/raci-matrix/SKILL.md| Model | Lift | Δ tokens | Δ turns | Cases | Verified |
|---|---|---|---|---|---|
| gemini-3.6-flashbest | +36% | +55% | 0% | 22 | 54d ago |
| gemini-3.5-flash | pending re-run | — | |||
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | — | — |
| case-10 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-22 | ✗→✓ | ▲ Improved | — | — |
| case-07 | ✗→✓ | ▲ Improved | — | — |
Enforces the RACI responsibility-assignment convention whenever responsibilities for a set of activities must be allocated across a set of roles: a grid with activities as rows and roles as columns, every cell holding one of the letters R, A, C, I (or staying blank), and every row carrying exactly one A. Not for drawing org charts, building schedules, or headcount planning.
an activity with two A's has no owner at all (the classic violation is a cell pair sold as "shared accountability"). When two roles both claim ownership, exactly one keeps the A; the other becomes R or C.
A row may carry several R's; it must carry at least one.
transpose it, and never replace the grid with prose paragraphs about collaboration.
personal names, no percentages, no "leads" / "supports" / "helps". Nuance goes in a footnote beneath the grid, never inside a cell.
out letters "to be safe" produces a wall of noise that says nothing.
row's single A.
is done. I is one-way — the role hears about it afterwards. Marking every stakeholder C manufactures review bottlenecks; when input is not truly needed, downgrade to I.
Prose ownership → grid. The default answer describes collaboration; the convention demands letters.
BEFORE:
> The office relocation will be coordinated by the operations manager, working closely with > facilities. IT will take care of the network cutover, and department heads should be kept in > the loop throughout.
AFTER:
| Activity | Ops manager | Facilities | IT lead | Dept heads | | --------------------- | ----------- | ---------- | ------- | ---------- | | Floor-plan selection | A | R | C | C | | Packing and labelling | R | R | | I | | Network cutover | I | C | A/R | I | | Move-day logistics | A/R | R | | I |
Two A's → one A. A monthly customer newsletter where the editor and the marketing lead "co-own" publication:
BEFORE (violates rule 2):
| Activity | Editor | Staff writer | Marketing lead | Support desk | | ----------------- | ------ | ------------ | -------------- | ------------ | | Issue publication | A | R | A | I |
AFTER:
| Activity | Editor | Staff writer | Marketing lead | Support desk | | ----------------- | ------ | ------------ | -------------- | ------------ | | Issue publication | A | R | C | I |
The editor answers for the issue going out; the marketing lead is consulted on timing; the support desk merely hears that it shipped.
Prose in cells → letters. A cell reading "reviews the draft and suggests edits" becomes C. A cell reading "keeps everyone posted" dissolves: the receiving roles get I, and the sender is already covered by its R on that row.
yield, split the activity into two rows with genuinely separate outcomes; each new row still gets exactly one A.
single role. Put the chair (or one named office) in the A position rather than the whole body.
of C's is a bottleneck warning sign.
row keeps a single A.
admit the activity is unstaffed.
none.
would do.
grid.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +36 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.5-flash | verified | 7/10/2026 | +95% |
Other measured skills in the registry, with their headline benchmark lift.