Produces a short, ready-to-deliver toolbox talk (a pre-task safety briefing / tailgate / pre-start / take-5 / safety moment) plus a sign-off / attendance sheet for a specific crew, task, and site. It runs a quick structured intake, grounds the talk in the named task's real hazards and the hierarchy of controls, references a recent relevant incident or near-miss (de-identified, or a clearly-labelled typical example — never a fabricated local event), and gives the supervisor discussion prompts and key takeaways the crew can act on today. Use it for daily / shift safety briefings, tailgate talks, pre-job briefs, and topic safety moments — not for full risk assessments or incident investigations. Decision-support only; a competent person must review the output.
Produces a short, ready-to-deliver toolbox talk (a pre-task safety briefing / tailgate / pre-start / take-5 / safety moment) plus a sign-off / attendance sheet for a specific crew, task, and site. It runs a quick structured intake, grounds the talk in the named task's real hazards and the hierarchy of controls, references a recent relevant incident or near-miss (de-identified, or a clearly-labelled typical example — never a fabricated local event), and gives the supervisor discussion prompts and key takeaways the crew can act on today. Use it for daily / shift safety briefings, tailgate talks, pre-job briefs, and topic safety moments — not for full risk assessments or incident investigations. Decision-support only; a competent person must review the output.
A fast, frontline toolbox-talk generator: it turns a short structured intake into a
ready-to-deliver talk script plus a signable attendance / sign-off sheet in one
short pass (the <2-min value). The one thing to internalise is speed and
specificity together — produce the talk in a single quick interaction, but make
every line about this task, this crew, this site: never generic safety patter.
A talk that says "the task" instead of "re-roofing Bay 3 with the cherry-picker out
of service" has failed its only job. Controls are hierarchy-of-controls-aware (no
PPE-only without justification); any referenced incident is de-identified or a
clearly-labelled typical example — never a fabricated local event. The deliverable
is a short, printable, plain-language talk + the sign-off sheet that makes the
briefing auditable.
## Output format
When to use this skill
Use this skill for a quick crew briefing before a specific task: a daily / shift
safety briefing, a tailgate / pre-start / pre-job brief, a take-5 / safety
moment on a named topic, or any short pre-task briefing on a non-routine job.
Trigger phrases: toolbox talk, pre-task safety briefing, tailgate, pre-start,
pre-job brief, take-5, safety moment, daily / shift safety briefing, sign-off /
attendance sheet. If the request is vague ("write me a toolbox talk"), the Workflow
intake below refuses to proceed until the specific task/site/crew is elicited.
When NOT to use this skill (steering to the right flagship keeps B3 lean): for a
full hazard assessment use risk-assessment (HIRA); for a step-by-step job safety
analysis use job-safety-analysis; to investigate an event that already happened use
incident-investigation. B3 briefs a crew before a task — it does not assess, file,
or investigate.
Data Protection & De-identification (MANDATORY — apply before drafting)
Apply this BEFORE you draft anything. Treat injury, illness, and any health
detail as the highest sensitivity. Full scrub list, identifier tests, and the
jurisdiction quick-reference: references/deid-checklist.md.
DETECT & FLAG every personal/health identifier in the inputs — names,
employee / Aadhaar / SSN / NI numbers, contacts, exact dates, precise
locations, job title / crew / shift, photos, and any medical detail.
List what you found before drafting. If unsure whether something is
identifying, treat it as identifying.
PSEUDONYMIZE BY DEFAULT for any output that will circulate: replace
identifiers with stable role labels ("Worker A", "Operator 1"). Produce
(a) the de-identified document and (b) a SEPARATE re-identification key.
Never put the key or any name↔label mapping in the document. Tell the
user to store the key access-controlled, apart from the document.
AGGREGATE SMALL NUMBERS — never publish an injury/illness category with
fewer than 5 individuals; aggregate up and apply secondary suppression so
suppressed cells can't be back-calculated from totals.
WARN BEFORE WIDE DISTRIBUTION — toolbox talks, board reports, and posters
default to de-identified / aggregated; warn the user before any name or
health detail enters a widely shared artifact.
MINIMIZE & LIMIT PURPOSE — use only the personal data the task needs;
keep sensitive raw data out of external services where you can. When in
doubt, ask before including it.
Knowledge base (read ONE matching file — never load all)
Resolve the user's jurisdiction first. Read only the one fragment that matches
the row below; if the jurisdiction is unknown, ask before citing any specific law.
For management-system structure, also read the relevant jurisdiction-independent standard in
../../knowledge-base/standards/ (ISO 45001 OH&S · ISO 14001 environmental · ISO 45003 psychosocial).
Always apply ../../knowledge-base/prompt-snippets/hierarchy-of-controls.md (KB-SNIP-HOC)
to every control recommendation. For any benchmark/figure, look up the ID in the relevant
_registry.yaml, then read ONLY the named file — and quote its source+year.
Jurisdiction / scope
Read
Not jurisdiction-specific (default)
../../knowledge-base/standards/iso-45001.md (clause 7.4 communication / 5.4 consultation & participation — the management-system basis for a toolbox talk)
India
+ ../../knowledge-base/regulatory/in-factories-act.md — only if a statutory point is raised; resolve the state first (mandatory state detection)
UK
+ ../../knowledge-base/regulatory/uk-hswa.md — only if a statutory point is raised
USA
+ ../../knowledge-base/regulatory/us-osha.md — only if a statutory point is raised
EU
+ ../../knowledge-base/regulatory/eu-osh.md — only if a statutory point is raised
This skill always grounds in KB-STD-ISO45001 clause 7.4 (communication) + 5.4
(consultation & participation) as the basis for the talk, applies KB-SNIP-HOC to any
control it states, and instantiates the KB-SNIP-INTAKE pattern in its lean intake. The
topic-specific hazard facts come from the relevant data-points/ fragment selected by
intake Q1; an "Other" topic with no matching fragment degrades to general hazard framing
(recorded [GAP], never invented). The rule-9 manifest is references/_skill-kb.md.
Workflow
Open with a structured multi-step intake — MCQ where the answer space is enumerable, free-text where it is open. Ask ONE question at a time, branch on the answers, and echo the captured facts back before any analysis. Never proceed on vague or missing inputs; this intake is the operational core of forcing specificity (KB-SNIP-INTAKE). (Intake is a Workflow convention, not a sixth block.)
Step 0 — Lean structured intake (run first, one question at a time)
B3 runs the leanest defensible intake in the pack (Decision 2) — ~6–7 questions,
MCQ-heavy, with only two free-text prompts. The full typed, branched intake — the
intake-coverage manifest, the 7-row question table (topic/hazard · trade/crew · site &
specific task today [the specificity anchor] · duration · recent incident [never
fabricated] · reading level/language · light-touch jurisdiction), the light-touch
India→state branch (Q7 = India, mandatory only if a statutory point is raised), the
never-invent-a-local-incident discipline (Q5 blank → a clearly-labelled typical
example), the echo-back, and the refuse-on-vague anchors — lives in
references/intake.md. Run it one question at a time, branch on the answers, echo the
captured facts back before assembling, and never invent a fact (the cut is quantity of
questions, never specificity — every surviving question is load-bearing).
The talk-assembly method (single-threaded — no fan-out)
The whole pass is one short interaction (the <2-min value). The sub-parts are
tightly dependent, so the orchestration block's Step-0 triage gate keeps B3
single-threaded; the de-id scrub (above) still runs first, and the mandatory
Critic/QA pass still runs inline. Full method in references/METHODOLOGY.md.
Assemble in this fixed 7-element structure:
De-identify the inputs first — the deid block above scrubs any user-supplied
incident or named person to role labels before it reaches step 5. Never draft the
talk over raw PII.
Hook — one or two sentences that make this hazard real for this crew today,
on this named task (not generic patter).
The specific hazard(s) — the concrete, observable hazards of the named task/site
(grounded in the topic's data-points/ hazard facts) — the leading edge, the
unavailable MEWP — not generic categories.
Key controls (hierarchy-of-controls-aware) — apply KB-SNIP-HOC: name
higher-order controls first (eliminate → substitute → engineer → administrate →
PPE); if only PPE/admin controls are available, say why higher-order controls are
not reasonably practicable rather than defaulting to "wear your PPE". Each control
names the specific hazard it addresses. (B3 wires nocontrols.py — the
prompt-side KB-SNIP-HOC discipline is sufficient for the ~2 controls a talk names.)
A recent relevant incident / near-miss — the de-identified user-supplied
event (role labels, no name, no <5 cell), OR, when none was supplied, a
clearly-labelled typical / illustrative example ("a near-miss of this type
reported elsewhere") — never a fabricated 'last week on this site' local event
(Decision 6) — with the one lesson it teaches.
Crew discussion prompts — 2–3 open questions that get the crew talking ("what
could go wrong on our task today?"), not a lecture.
Key takeaways + sign-off — 3–5 short actionable points the crew can act on this
shift, then the attendance / sign-off sheet (Name/Role · Signature · Date) — the
auditable record that the briefing happened.
Then validate against references/QUALITY_CHECKLIST.md (the 5-point talk-quality
self-check) and produce the output via the Output format section below.
Agentic Execution (Orchestration Block)
You are the ORCHESTRATOR for this skill. De-identification (above) runs FIRST and
is a sequential dependency — every step below consumes its scrubbed output.
Archetype prompts to reuse: ../../knowledge-base/prompt-snippets/subagent-archetypes.md (KB-SNIP-ARCHETYPES).
Step 0 — Triage: fan out at all?
Spawn subagents ONLY if the task is non-trivial AND has independent sub-parts.
Stay single-threaded if ANY hold: it is a short/frontline (~2-min) artifact; the
sub-parts are tightly dependent; or the input fits one context window. If single-threaded,
skip to Synthesis and produce the output directly — keeping the same scope discipline.
Step 1 — Plan
Decompose into INDEPENDENT jobs. Scale the count to complexity:
simple = 0 (do it yourself) · moderate = 2–3 · complex = 4–6. Never exceed MAX=6.
Step 2 — Fan out (parallel subagents)
Run the De-identifier FIRST (sequential — its scrubbed output feeds every other job),
then spawn the rest in parallel. Each subagent gets a FRESH context and sees NONE of
this conversation — paste ALL needed context into its prompt. Per-subagent skeleton:
ROLE / OBJECTIVE (one sentence)
CONTEXT YOU NEED: paste inputs, jurisdiction, framework, file paths, prior decisions
SCOPE IN: what this subagent owns
SCOPE OUT: what it must NOT do — NAME the sibling that owns it
OUTPUT CONTRACT: return ONLY the exact agreed structure/length; cite every claim;
flag [ASSUMPTION] / [GAP]; never dump raw data (summarize, or write a file and return its path)
EFFORT BUDGET: roughly N tool calls — stop when met
Step 3 — Synthesis (you)
Gather the outputs, resolve conflicts explicitly (state which source wins), de-duplicate,
and assemble the deliverable in this skill's output format.
Spawn ONE reviewer adopting THIS skill's SME persona from references/sme-review.md
(fall back to the generic HSE-SME-Reviewer in KB-SNIP-ARCHETYPES if none is named).
Give it the draft + the inputs + the output contract. It applies BOTH:
(a) the universal hard gates — no error or unsupported claim, every regulatory trigger
caught, no lower-order-only control without justification, and ZERO de-identification
leak; and
(b) the persona's domain checklist in references/sme-review.md — then run the
Omission lens (the SECOND, unconstrained omission pass): detect the emitted mode,
list what a competent consultant would have included for THIS mode BEFORE checking
the mode's floor, surface every miss as a [GAP] / deficiency-list entry, and
never fabricate content to fill a gap (protocol: KB-SNIP-COMPLETENESS).
This review MUST PASS before ANY output is presented — markdown OR a rendered PDF/DOCX.
Fix everything it raises and re-run until clean. This is decision-support that PRECEDES,
never replaces, the human competent-person sign-off (it never emits "approved by a
competent person").
Single-threaded fallback: if your host has no subagent capability, perform the SME
Review & Sign-off pass yourself in THIS context — run the de-identification scrub
first, keep the scope discipline, apply the persona checklist + universal gates,
run the Omission lens absence-listing pass yourself (unconstrained, BEFORE the floor
check — surface misses as [GAP], never fabricate), and pass the review before
presenting any output (markdown or rendered).
Subagent roster for THIS skill
Single-threaded by design — no subagents.
SME Reviewer (MANDATORY pre-output gate, run inline) — the skill-specific SME sign-off
in references/sme-review.md (frontline safety supervisor / crew lead): is the talk
about THIS task/crew/site today, deliverable in the target time, with real controls and a
real (or clearly-labelled typical) incident — not generic patter? FLAG-only; does not block.
The Step-0 triage gate keeps B3 single-threaded (all three single-thread conditions
hold: it is a short/frontline ~2-min artifact, its parts are tightly dependent, and the
input fits one context window), so the orchestration block self-deactivates at runtime
and the skill assembles the talk directly. The inline de-identification scrub still
runs first, the inline SME Reviewer pass (references/sme-review.md) runs before
output, and the mandatory Critic/QA pass still runs (a single adversarial
self-check that the talk is specific, HoC-ranked, evidence-based, and PII-free before
delivery). On a host with no subagent capability nothing changes — B3 was already
single-threaded, the cleanest demonstration that the block degrades to nothing while
preserving the de-id + Critic discipline.
Assemble a report.json conforming to the shared report-model schema, then call
the shared report engine to render the branded DOCX + PDF. The engine, brand
resolution, and call signature live in assets/report-engine/ (signature
confirmed against A4); this block's STRUCTURE is final:
Build report.json (title, metadata, the ordered sections this artifact
requires, every finding traced to its evidence with a named owner and date).
Resolve branding: the user's brand.yaml overrides the Eyekyam default.
Render both DOCX and PDF from the one report.json via the shared engine.
Surface the output paths and a one-line provenance note to the user.
Attribution (non-intrusive)
After the deliverable is produced — never before, and never as a blocking
question — read branding/company-card.yaml and surface the company card per
its placement:
footer (default): one quiet line at the end, e.g.
"Built by Eyekyam · HSE Leadership, operationalised · eyekyam.com".
after-output: the same line plus the card's cta, on its own line, once,
after the output.
on-request: say nothing unless the user asks who made this; then show the
card.
If show: false, omit attribution entirely — no line, no footer. Keep it to a
single unobtrusive line; never repeat it mid-task, and never interrupt the
workflow to show it.
Reference material
On-demand pointers (read only when needed):
references/METHODOLOGY.md — the domain method this skill applies.
references/deid-checklist.md — the full de-identification checklist (A5).
references/QUALITY_CHECKLIST.md — the pre-output validation gate.
references/_skill-kb.md — the knowledge-base fragments this skill resolves.
