Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Produce a toolbox talk or pre-task safety briefing from the day's planned construction work. Use when asked to write a toolbox talk, prepare a pre-task plan or JHA/JSA briefing, brief a crew on today's hazards, or plan safety for a specific task like a crane pick, excavation, or hot work. Produces a crew-ready briefing with task-specific hazards, controls ordered by the hierarchy of controls, required permits, and explicit stop-work triggers.
.claude/skills/mohitagw15856-site-safety-briefing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 117% | 0% |
A toolbox talk that could apply to any site on any day protects nobody. This skill builds the briefing from today's actual work: what tasks are running, what they conflict with, what the weather and site conditions add, and — critically — what conditions mean the crew stops and calls the supervisor. Controls are ordered by the hierarchy of controls, because "wear your PPE" is the last line of defence, not a plan.
Ask for what's missing; from a bare task name, build the briefing from typical hazards and label site-specific items [confirm on site]:
Identify hazards per task, not generically. Sweep these classes for each task: gravity (falls, dropped objects), mechanical/energy (struck-by, caught-between, stored energy, electrical), excavation/ground (collapse, utilities, water), atmosphere (confined space, dust, fumes, hot work), environment (heat/cold, wind — especially for crane picks and panel handling), and human factors (new crew members, fatigue, language barriers).
Order every control by the hierarchy — and say which level it is:
| Level | Control | Example | |---|---|---| | 1 | Eliminate | Prefabricate at ground level instead of working at height | | 2 | Substitute | Mechanical lifting instead of manual handling; low-VOC product | | 3 | Engineering | Guardrails, trench box, ventilation, physical barricades | | 4 | Administrative | Permits, exclusion zones, spotters, rotation, signage | | 5 | PPE | Harness, respirator, cut gloves — the last layer, never the plan |
If a briefing's controls are all level 4–5, flag it: the task plan itself may need rework.
Stop-work triggers must be observable. Not "if unsafe" — write conditions a crew member can see: wind above the crane chart limit; water entering the excavation; utility strike or unexpected line; barricade down in the exclusion zone; anyone in the fall zone during the pick. State plainly that anyone can stop work without repercussions.
1. Today's work — tasks, crews, locations, in one glance. 2. Task hazard & control table — | Task | Hazard | Control (hierarchy level) | Who verifies | 3. Permits & inspections required today — permit type, task, who holds it, valid when. 4. SIMOPS / crew interfaces — who's working near whom, and the separation rule. 5. Stop-work triggers — bulleted, observable, with "stop, make safe, call supervisor]". 6. Emergency info — muster point, first aider, nearest hospital, who calls emergency services. 7. Sign-on sheet — name/signature lines; space for crew-raised concerns (record them — that's the point of the talk).
[confirm on site] for the supervisor to fill in| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 22,327 | 36,667 | +64% | 1 | 1 | 0% | 3,623 | 5,366 | +48% | 0 | 0 | — |
case-05 | pass→pass | 43,058 | 46,218 | +7% | 1 | 1 | 0% | 6,824 | 7,188 | +5% | 0 | 0 | — |
case-06 | pass→pass | 24,392 | 41,934 | +72% | 1 | 1 | 0% | 3,069 | 6,722 | +119% | 0 | 0 | — |
case-07 | pass→pass | 31,586 | 34,524 | +9% | 1 | 1 | 0% | 3,498 | 5,236 | +50% | 0 | 0 | — |
case-01 | fail→pass | 43,823 | 39,608 | -10% | 1 | 1 | 0% | 5,149 | 5,991 | +16% | 0 | 0 | — |
case-02 | fail→pass | 28,750 | 31,008 | +8% | 1 | 1 | 0% | 3,641 | 5,212 | +43% | 0 | 0 | — |
case-03 | fail→pass | 30,685 | 33,447 | +9% | 1 | 1 | 0% | 4,194 | 5,066 | +21% | 0 | 0 | — |
case-08 | fail→pass | 29,786 | 34,278 | +15% | 1 | 1 | 0% | 3,474 | 5,145 | +48% | 0 | 0 | — |
case-09 | pass→pass | 27,991 | 37,559 | +34% | 1 | 1 | 0% | 2,888 | 5,527 | +91% | 0 | 0 | — |
case-10 | pass→pass | 29,248 | 35,627 | +22% | 1 | 1 | 0% | 3,178 | 5,409 | +70% | 0 | 0 | — |
case-11 | fail→fail | 22,072 | 29,348 | +33% | 1 | 1 | 0% | 2,604 | 4,927 | +89% | 0 | 0 | — |
case-12 | fail→pass | 26,298 | 38,066 | +45% | 1 | 1 | 0% | 2,676 | 5,803 | +117% | 0 | 0 | — |
case-13 | fail→pass | 25,245 | 30,039 | +19% | 1 | 1 | 0% | 3,027 | 5,200 | +72% | 0 | 0 | — |
case-14 | fail→pass | 19,452 | 35,948 | +85% | 1 | 1 | 0% | 2,445 | 5,561 | +127% | 0 | 0 | — |
case-15 | pass→pass | 27,885 | 36,326 | +30% | 1 | 1 | 0% | 3,217 | 5,214 | +62% | 0 | 0 | — |
case-16 | fail→pass | 23,371 | 31,352 | +34% | 1 | 1 | 0% | 2,524 | 5,130 | +103% | 0 | 0 | — |
case-17 | fail→pass | 21,465 | 37,450 | +74% | 1 | 1 | 0% | 2,772 | 5,426 | +96% | 0 | 0 | — |
case-18 | fail→pass | 22,534 | 36,006 | +60% | 1 | 1 | 0% | 2,579 | 5,129 | +99% | 0 | 0 | — |
case-19 | fail→pass | 24,834 | 31,567 | +27% | 1 | 1 | 0% | 2,567 | 5,116 | +99% | 0 | 0 | — |
case-20 | fail→pass | 26,019 | 29,460 | +13% | 1 | 1 | 0% | 3,234 | 4,178 | +29% | 0 | 0 | — |
case-21 | fail→pass | 24,655 | 26,440 | +7% | 1 | 1 | 0% | 2,769 | 4,599 | +66% | 0 | 0 | — |
case-22 | fail→pass | 19,089 | 20,521 | +8% | 1 | 1 | 0% | 1,920 | 3,696 | +93% | 0 | 0 | — |
case-23 | pass→pass | 21,093 | 28,904 | +37% | 1 | 1 | 0% | 2,379 | 4,310 | +81% | 0 | 0 | — |
case-24 | fail→pass | 28,702 | 27,264 | -5% | 1 | 1 | 0% | 3,053 | 4,861 | +59% | 0 | 0 | — |
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. 24 cases were attempted. The headline lift of +63 percentage points is the difference between those two pass rates over the 24 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.