safety-audit

name	safety-audit
description	Conducts consultant-grade safety and compliance audits of a specific site, system, or process against a defined standard or checklist. Use this skill whenever a user asks to run a safety audit, an inspection, a compliance audit, a management-system audit, or a conformity assessment; to audit a site or process against ISO 45001, a regulation, or a custom checklist; to record clause-by-clause findings, nonconformities (major or minor), observations, or opportunities for improvement; to rate conformity; or to produce an audit report with corrective actions. Grounds the audit method in ISO 45001 clause 9.2 (internal audit), traces every finding to objective evidence, risk-rates each nonconformity, and drives corrective actions through the hierarchy of controls (no PPE-only treatments without justification) with named owners and due dates — emitting a branded audit report. Decision-support only; a competent person must review the output.
license	Apache-2.0
metadata	{"author":"eyekyam","version":"1.0","category":"compliance","tier":1,"audience":["M","F","C"],"industry":["All"],"jurisdiction":["All"],"status":"stable","plugin":"hse-core","hse_reviewed_by":"","hse_reviewed_date":""}

Safety Audit (conformity assessment against a standard or checklist)

A consultant-grade HSE skill that audits a specific site, system, or process against a defined standard or checklist and produces a clause-by-clause audit report. It grounds the audit method in ISO 45001 clause 9.2 (internal audit), traces every finding to objective evidence, risk-rates each nonconformity, and drives corrective actions through the full hierarchy of controls — the single lever that separates a defensible audit from copy-paste paperwork. An audit finding with no evidence trail is, by definition, indefensible; this skill refuses to produce one.

When to use this skill

Use this skill whenever a user asks to:

• Run a safety audit, inspection, compliance audit, management-system audit, or conformity assessment of a named site, system, or process.
• Audit something against ISO 45001, a regulation (OSHA / Factories Act / HSWA), or a custom checklist.
• Record clause-by-clause findings — conformity, nonconformity (major or minor), observation, or opportunity for improvement — each traced to evidence.
• Rate conformity for an audited scope, or produce an audit report with corrective actions (a CAPA register).

Do not use this skill to investigate an incident or run a root-cause analysis (that is incident-investigation / B5), to build a single-task JSA (job-safety-analysis), or to manage an existing CAPA register's lifecycle over time (capa-manager / B7 — this skill produces the register B7 then tracks). If the request is vague ("audit the site"), the Workflow intake below forces the specific scope and the audit criteria before any drafting.

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.

1. 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.
2. 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.
3. 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.
4. 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.
5. 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 / criteria	Read
India	../../knowledge-base/regulatory/in-factories-act.md (+ in-state-forms.md for the user's state — mandatory state detection before citing any form)
UK	../../knowledge-base/regulatory/uk-hswa.md
USA	../../knowledge-base/regulatory/us-osha.md
EU	../../knowledge-base/regulatory/eu-osh.md
Unknown	Ask before citing any specific law
Audit method (always)	../../knowledge-base/standards/iso-45001.md (KB-STD-ISO45001, clause 9.2 — internal audit: the method backbone, scope→criteria→evidence→findings→reporting)
Criteria = ISO 45001 (Q-Crit)	../../knowledge-base/standards/iso-45001.md (the audited clauses — e.g. 6.1.2, 7.2, 8.1, 9.1 — become the criteria set the audit walks; distinct from 9.2, the method)
Criteria = custom checklist (Q-Crit)	Use the user-supplied checklist items as the criteria set; cite no external clause beyond what the user names (a private checklist has no KB fragment)

Method vs criteria (the distinction this skill turns on): ISO 45001 9.2 is always the method (how the audit is planned and conducted). The criteria — what is audited against — are chosen at intake (Q-Crit): ISO 45001 itself, a regulatory regime, or a custom checklist. The same 9.2 loop audits a site against any of them.

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 (one at a time; branch; echo back; never proceed on vague)

The full typed, branched intake — the intake-coverage manifest, the question table (jurisdiction · the scope/boundary anchor Q-Scope · the criteria gate Q-Crit · audit type · evidence available · industry · nonconformity-rating matrix · the audit team / lead auditor Q-Team · the audit date + CAPA cycle Q-When · physical location), the mandatory India→state branch (Q-Juris = India → Q-Juris-a), the criteria-gate branches (Q-Crit = ISO 45001 → KB-STD-ISO45001; A regulatory regime → the jurisdiction fragment; A custom checklist → no external clause), 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 any analysis, and refuse to proceed on a vague scope ("general site audit") or an undeclared criteria set (Q-Scope and Q-Crit are load-bearing — record [GAP], never invent a clause). An audit with no defined boundary is unauditable; one with no criteria set cannot classify a finding.

The audit method (ISO 45001 9.2 internal-audit loop)

Full method — finding-classification rules, sampling guidance, evidence-sufficiency tests — in references/METHODOLOGY.md. The steps:

1. De-identify the inputs FIRST (the deid block + the De-identifier-runs-first rule). Audit evidence routinely carries worker names, signatures, training/injury records — scrub all of it to role labels before any assessment; everything downstream (every finding, every quoted evidence item) consumes the scrubbed text. A personal identifier surviving into a finding's evidence trail is a de-id auto-fail.
2. Establish the scope & criteria set. Fix the boundary (Q-Scope) and resolve Q-Crit into a concrete, walkable clause/checklist list: ISO 45001 → the relevant clauses from KB-STD-ISO45001; a regulatory regime → the duties from the matched jurisdiction fragment; a custom checklist → the user's items. State the audit type (Q-Type) and methodology (interview / document review / observation / sampling). Record [GAP] where the criteria set is incomplete — never invent a clause.
3. Gather & assess objective evidence per criterion. For each clause/checklist item, assess the available evidence against that specific criterion: what it shows, whether it is sufficient, what is missing. Each assessment names the specific evidence item (document id, observation, interview-role, record) it rests on. Flag [GAP] where evidence is absent — an unassessed criterion is "insufficient evidence," never silently passed.
4. Record one finding per criterion, each classified + traced to evidence (the defensibility core). Classify with the 4-class ISO scheme:
   • Conformity — evidence demonstrates the requirement is met.
   • Nonconformity — Major — a systemic failure or total absence of a required arrangement.
   • Nonconformity — Minor — an isolated lapse against an otherwise-conforming arrangement.
   • Observation — conforming now, but a risk of drift / an early warning.
   • Opportunity for improvement — conforming, but a better practice is available. Every finding names the clause/checklist item, the objective evidence, and the classification. Risk-rate EVERY nonconformity by default via risk_matrix.load_matrix(config) then risk_matrix.score(likelihood, severity, matrix) (config from Q-NCrate, default 5×5) so remediation can be prioritised — the band is the engine's, not prose. Conformities/observations need no score.
5. Rate overall conformity. Count by classification; derive a conformity rating for the scope (conformity % of assessed criteria, count of major/minor nonconformities, highest residual nonconformity risk band) — computed from the per-finding ratings, not asserted in prose.

The orchestration block below sits after this Workflow so the triage gate judges the assembled work before fanning out. The deterministic rating/ranking steps (4 risk-rating via risk_matrix; 6 ranking via controls, register validation via smart_actions) are A7 script calls in every case — never a fan-out job (there is no "Conformity-Scorer" subagent: deterministic work runs in the script).

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

The standard moderate roster (A6 "moderate = 2–3"). The De-identifier is the sequential first gate, not a fan-out peer; the 3 fan-out jobs are Evidence-Assessor + Regulatory-Checker + Drafter; Critic/QA is mandatory. There is no Conformity-Scorer — nonconformity risk-rating, control ranking, and CAPA-register validation are deterministic A7 script calls at steps 4/6 (the project's core value: deterministic work runs in the script).

• De-identifier — runs FIRST (sequential gate, not a fan-out peer); scrub all PII/health detail in the audit evidence (names, signatures, training/injury records) to role labels before any assessment. Every fan-out job below consumes only its scrubbed output.
• Evidence-Assessor — for each criterion in the resolved criteria set, assess the scrubbed objective evidence against that specific clause/checklist item: what it shows, sufficiency, what is missing; cite the specific evidence item per assessment; flag [GAP] where evidence is absent. SCOPE-OUT: law/criteria wording (Regulatory-Checker owns it), drafting the report (Drafter owns it). (Conformity rating is NOT a subagent — it is the A7 risk_matrix/controls scripts.)
• Regulatory-Checker — for the resolved jurisdiction + criteria, return the applicable duty + clause/section + (India → resolved STATE first) the state form via KB-REG-IN-STATEFORMS, and confirm each cited clause exists; conservative — flag [GAP] when unsure. SCOPE-OUT: assessing evidence (Evidence-Assessor), drafting (Drafter), inventing a form number.
• Drafter — write the clause-by-clause findings + nonconformity log + conformity summary + corrective-action plan + CAPA register to the output template using role placeholders; each finding tagged its classification + evidence trail, each control tagged its KB-SNIP-HOC tier (consumes the De-identifier's scrubbed text + the Evidence-Assessor's per-criterion assessments + the A7 risk_matrix/controls ratings + the Regulatory-Checker's verdict). SCOPE-OUT: assessing evidence (Evidence-Assessor), checking law (Regulatory-Checker).
• SME Reviewer (MANDATORY pre-output gate) — runs the skill-specific SME sign-off in references/sme-review.md (lead auditor, ISO 19011 / ISO 45001 9.2) before any output: every finding traced to OBJECTIVE evidence against a named criterion, classified correctly (major vs minor NC vs observation vs OFI), every nonconformity risk-rated, and the emitted CAPA register handoff-clean for the B6→B7 seam.
• Critic/QA (MANDATORY) — every criterion assessed or [GAP]-flagged, every finding classified + traced to objective evidence, every nonconformity risk-rated (via the A7 engine), every corrective action HoC-ranked + owned + dated + finding-linked, no PPE/admin-only treatment without justification, every cited clause traces to the KB, zero PII/health leak. PASS/FAIL.

Single-thread fallback (the block's canonical line): on a host with no subagent capability, the same jobs run sequentially in one context — De-identifier scrub first, the A7 rating/ranking/validation calls still made deterministically, scope discipline kept, the Critic/QA pass still performed. (This skill is not single-threaded-by-design; it ships the roster above.)

## 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:

1. Build report.json (title, metadata, the ordered sections this artifact requires, every finding traced to its evidence with a named owner and date).
2. Resolve branding: the user's brand.yaml overrides the Eyekyam default.
3. Render both DOCX and PDF from the one report.json via the shared engine.
4. 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.