| name | incident-investigation |
| description | Investigates an HSE incident — injury, illness, near-miss, property or environmental loss — and produces a defensible, de-identified investigation report: timeline, evidence log, root-cause analysis (5-Whys / ICAM / SCAT / Fishbone / Swiss-Cheese), root and contributing causes traced to evidence, a hierarchy-of-controls corrective-action plan with named owners and due dates, and the jurisdiction's reporting/notification requirements. Use it to investigate an accident, conduct a root-cause analysis / RCA, write up an incident, build a CAPA from an event, or determine whether an incident is reportable (RIDDOR / OSHA 29 CFR 1904 / India state accident form). Decision-support only; a competent person must review the output. |
| license | Apache-2.0 |
| metadata | {"author":"eyekyam","version":"1.0","category":"incident-management","tier":1,"audience":["M","C"],"industry":["All"],"jurisdiction":["All"],"status":"stable","plugin":"hse-core","hse_reviewed_by":"","hse_reviewed_date":""} |
Incident Investigation
A consultant-grade HSE skill that turns an event — injury, illness, near-miss,
property or environmental loss — into a defensible, de-identified investigation
report: a timeline, a numbered evidence log, a root-cause analysis, root and
contributing causes each traced to a numbered evidence item, a
hierarchy-of-controls CAPA with named owners and ISO due dates, and the
jurisdiction's reporting/notification verdict. It is grounded in ISO 45001 clause
10.2 (incident, nonconformity & corrective action). Two non-negotiables hold
throughout: de-identify first (no analysis or drafting ever touches raw
PII/health data), and trace every conclusion to evidence (no un-evidenced cause,
no anonymous action, no invented citation). Root-cause structure and systemic reach,
control ranking, and action traceability are deterministic (the A7 rca,
controls, smart_actions engines), never prose judgement.
When to use this skill
Use this skill after an incident, accident, or near-miss — when a root-cause
analysis, a CAPA, or a reportability decision is needed for a concrete event at a
named site/asset. Trigger phrases: investigate an incident, incident investigation,
root cause analysis / RCA, 5-Whys / ICAM / SCAT / Fishbone (Ishikawa) / Swiss-Cheese,
accident investigation, near-miss investigation, write up an incident, CAPA from an
incident, corrective action plan, is this reportable / reportability, RIDDOR, OSHA
recordable / 29 CFR 1904, India accident notice / state accident form. Not for
live emergency response (a different skill), and not for routine incident-rate
dashboards (use incident-rate-calculator, B10 — this skill may surface only one
contextual rate). If the request is vague, the Workflow intake below forces the
specifics — and the de-identification of all personal/health detail — before any
analysis.
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 | Read (for the reporting/notification verdict) |
|---|
| India | ../../knowledge-base/regulatory/in-factories-act.md + in-state-forms.md — resolve the STATE first (mandatory), then read the accident-notice row for that state (e.g. Maharashtra → Form 24 within 24h via KB-REG-IN-STATEFORMS); never a national form number; append the OSH-Code transition note |
| UK | ../../knowledge-base/regulatory/uk-hswa.md (RIDDOR 2013 reportable events + timelines) |
| USA | ../../knowledge-base/regulatory/us-osha.md (29 CFR 1904 recordkeeping + 1904.39 reporting timelines: 8h fatality / 24h in-patient-hospitalization/amputation/eye-loss) |
| EU | ../../knowledge-base/regulatory/eu-osh.md |
| Unknown | Ask before citing any specific law |
This skill always grounds in KB-STD-ISO45001 clause 10.2 (incident,
nonconformity & corrective action) — the clause that structures the RCA + CAPA — and
applies KB-SNIP-HOC to every CAPA control (paired with the A7 controls.rank_controls
check). For an India site it resolves the state via KB-REG-IN-STATEFORMS (mandatory
state detection — confirm the state before citing any form; an un-seeded state →
[GAP] + "verify the state accident form with a competent person", never an invented
form). Every citation quotes the fragment's source+year; B5 never invents a
citation. 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 — Structured intake (run this first, one question at a time)
The full typed, branched B5 intake — the intake-coverage manifest, the question table
(narrative · when/where · incident type · severity · people-involved [IMMEDIATE de-id] ·
immediate causes · evidence · permits-in-force · RCA-method preference · jurisdiction →
mandatory India→state · investigation team · optional rate context), the four
RCA-method sub-prompt branches, the mandatory India→state branch (Q9 = India → Q9a +
KB-REG-IN-STATEFORMS), the de-identified echo-back, and the refuse-on-vague anchors —
lives in references/intake.md. Run it one question at a time, branch on the answers,
and never proceed on vague inputs (record [GAP] for missing evidence and
[ASSUMPTION] for anything inferred — never invent evidence or causes). Q5 (people
involved) and Q2 (when & where) are scrubbed in Workflow step 2 before any analysis, so
the echo-back in references/intake.md shows role labels only ("Worker A", witness
"W-1"), never raw names.
The investigation method (de-id-first → evidence → cause → control → report)
Full method in references/METHODOLOGY.md. The sequence is strict: evidence before
cause, cause before control. Steps:
- Structured intake (Step 0 above) — gather facts; never proceed on vague inputs.
- De-identify FIRST (the
deid block above — a sequential dependency). Before
anything else, run the De-identifier over all intake inputs: detect and list
every identifier; pseudonymize to stable role labels ("Worker A", witness "W-1");
aggregate any injury/illness cell <5; emit the re-identification key as a
separate artifact returned to the user — never in the report or any subagent
prompt. Every subsequent step consumes only the scrubbed, role-labelled text. (On
a single-thread host the orchestration fallback runs this inline, first.)
- Establish the timeline & gather evidence. Reconstruct the factual sequence from
the scrubbed inputs into a numbered, time-ordered event list; build the evidence
log with each item numbered
E-1, E-2, … and typed (statement / photo-ref / log /
reading / document). Flag [GAP] where a needed fact is absent. No causes yet —
facts only.
- Root-cause analysis (A7
rca — the user-chosen method, one of five). Take the
method chosen at Q8 (5-Whys / ICAM / SCAT / Fishbone / Swiss-Cheese — all
A7-validated). Run rca.scaffold(method, problem) to get the method's required
slots; fill them from the evidence log (each causal claim cites an E-n); then
run rca.validate(method, analysis) and resolve every issue it raises.
reaches_systemic is enforced for ALL FIVE methods — it is not optional: a
5-Whys terminating at individual error fails; an ICAM with no Organisational Factor
fails; a SCAT missing Basic-Causes / Lack-of-Control fails; a Fishbone with only
the "Man" branch populated fails (all-individual-blame); a Swiss-Cheese with no
organisational-influence layer fails (no latent/systemic reach). The skill must
reach a systemic/organisational factor whatever the method.
- Identify root + contributing causes — traced to evidence (defensibility). Emit
causes as
{id: "RC-n", statement, tier: root|contributing|immediate, evidence_ref: "E-n"}. Every cause must carry an evidence_ref into the step-3 evidence log; a
cause with no evidence is dropped or re-tagged [ASSUMPTION] and cannot be a root
cause. This is the defensibility lever made mechanical.
Contextual incident rate (optional, off by default). Only if Q10 captured period
hours and recordable counts, call incident_rates.compute_all(counts, hours_worked, period) to surface a single contextual TRIR/LTIFR metrics KPI ("this LTI moves the
site LTIFR to X"). It is never the investigation's conclusion, and never
fabricated — incident_rates raises on zero/negative hours, so it is called inside a
guard and omitted when the inputs are absent. Rate dashboards belong to B10.
The orchestration block (below) governs the agentic execution: for a non-trivial
incident the triage gate fans out steps 3–7 across the 4-agent roster after the
De-identifier; for a trivial single-witness near-miss it stays single-threaded but
still runs the de-id scrub first and the mandatory Critic/QA pass. The De-identifier
is never one of the parallel jobs — it is the sequential gate before fan-out.
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.
Step 4 — SME Review & Sign-off (MANDATORY — regulatory/safety output)
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
- De-identifier — runs FIRST (sequential dependency); scrub all PII/health detail —
injured-party, witnesses, diagnoses, exact dates/locations, small injury cells — into role
labels before any analysis. Returns the re-identification key SEPARATELY (to the orchestrator,
not to any sibling). Everything below consumes only its scrubbed output.
- A · Evidence & Timeline Reconstructor — assemble the numbered, time-ordered event
sequence and the numbered evidence log (E-1…) from the scrubbed inputs; flag [GAP].
SCOPE-OUT: does not assign causes (B owns it) or decide reportability (C owns it).
- B · Root-Cause Analyst — apply the chosen method (5-Whys / ICAM / SCAT / Fishbone /
Swiss-Cheese) via rca.py; every causal claim cites an evidence item (E-n); reach a
systemic/organisational factor (rca.validate
reaches_systemic true for whichever method).
SCOPE-OUT: reportability (C owns it) and control selection (D owns it).
- C · Regulatory Reportability Checker — for the resolved jurisdiction (India → resolved
STATE first), return verdict + clause/section + deadline + form (India state accident form
via KB-REG-IN-STATEFORMS / UK RIDDOR / US OSHA 29 CFR 1904). Conservative — flag [GAP] and
"ask a competent person" when unsure. SCOPE-OUT: does not draft the report or invent a form number.
- D · Corrective-Action Drafter — hierarchy-of-controls-tagged CAPAs, each tracing to a
named cause (RC-n) with a named owner + ISO due date + measure; prefer higher-order controls,
justify any PPE/admin-only. SCOPE-OUT: does not score causes (B) or check law (C).
- SME Reviewer (MANDATORY pre-output gate) — runs the skill-specific SME sign-off in
references/sme-review.md (ICAM-trained lead investigator + regulatory-reportability
specialist) before any output: does the RCA reach a SYSTEMIC/organisational factor (not stop
at "worker error"), is every cause traced to a numbered evidence item, and is the
reportability verdict correct, conservative, cited to the right rule/form/deadline — and
surfaced even when NOT reportable? FLAG-only; does not block on a flag.
- Critic/QA (MANDATORY) — adversarial read-only review: every cause evidence-backed, RCA
reaches a systemic factor, reportability cited conservatively to the matched KB row, every
CAPA traces to a cause with owner+date, no PPE/admin-only without justification, and ZERO
PII/health leak (no residual identifier, no <5 cell, no re-id key in the output).
## Output format
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.
