hse-skill-forge

name	hse-skill-forge
description	Scaffold a new, contract-conformant HSE Leadership skill at build time: this build-time authoring tool runs an author-interview, then emits a born-conformant skills/<name>/ with the five inline blocks byte-matching template/blocks/*, the folder skeleton, vocab-validated metadata, the seeded runtime intake step, and an A8-shaped evals skeleton. It is for skill authors and maintainers extending the pack, not an HSE-advice skill — it produces no risk assessment, talk, or investigation.
license	Apache-2.0
metadata	{"author":"eyekyam","version":"1.0","category":"ms-admin","tier":1,"audience":["C"],"industry":["All"],"jurisdiction":["All"],"status":"stable","plugin":"hse-systems","hse_reviewed_by":"","hse_reviewed_date":""}

hse-skill-forge — the build-time scaffolder

A build-time authoring tool for the people who EXTEND this skill pack — skill authors and maintainers, not HSE end users. It runs a short author-interview about the skill you want to build, then emits a born-conformant skills/<name>/ that lint_skills.py passes with zero hand-edits: the five inline blocks copied byte-for-byte from template/blocks/*, the folder skeleton, vocab-validated metadata, the §2.7 runtime-intake STEP seeded into the new skill's Workflow, the A7 component shim and A9 branding link where declared, and an A8-shaped evals/ skeleton (including the de-id fixture pair). It produces no HSE artifact — no risk assessment, no toolbox talk, no investigation; it produces the next skill.

The forge dogfoods the very contract it scaffolds: this SKILL.md is itself a conformant six-block skill (it is how the pack proves the contract is satisfiable in practice — A10 AC#8).

When to use this skill

Use this skill at build time when you are adding a new skill to the HSE Leadership pack or regenerating an existing skill's contract surface — for example:

• "Scaffold a new permit-to-work skill" / "add a behaviour-based-safety skill".
• "Re-sync this skill's inline blocks against the canonical template/blocks/*" (the --sync anti-drift verb).
• "Give me a contract-clean skeleton I can drop a domain body into."

Do not reach for this skill to produce an HSE deliverable for a real task or site — that is what the hse-core advice skills (risk-assessment, toolbox-talk, incident-investigation, …) are for. This skill lives in the hse-systems bundle, deliberately outside the advice grouping, so it is never mistaken for one.

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	Read
India	../../knowledge-base/regulatory/in-factories-act.md (+ in-state-forms.md for the user's state)
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

Workflow

The forge's Workflow is an author-interview: a build-time Q&A that asks WHAT skill to build, not what HSE problem to solve.

The two intakes are categorically distinct — never conflate them (A10 D4, "the single most important error to avoid"). The author-interview below is the forge's own Workflow; it asks the author about the skill they want to create. It is NOT the §2.7 runtime intake. The runtime intake is a seeded STEP — a <!-- TODO --> scaffold (templates/intake-todo.md) that the forge plants into the NEW skill's Workflow for the author to fill with that skill's domain questions later. The forge never invents the new skill's domain question VALUES; it only seeds the intake STEP. See references/mandatory-blocks.md for the full contrast.

Step 1 — Author-interview (one question at a time)

Ask these ten questions, MCQ where the answer space is enumerable, free-text where it is open; branch on the answers; echo the captured facts back before scaffolding:

1. Skill name — lowercase [a-z0-9-], ≤64 chars, MCP-safe (free-text).
2. One-line domain — what artifact does it produce? (free-text → description).
3. Category — MCQ from the A2 §4.1 category vocab (risk-assessment, training, incident-management, compliance, ms-admin, …).
4. Audience — MCQ, one or more of M / E / F / C.
5. Industry — MCQ from the industry vocab (All, Con, Chem, Min, Avi, …).
6. Jurisdictions — which the skill serves (All / UK / US / IN / EU).
7. Single-thread vs roster — MCQ: single-thread · moderate fan-out · the flagship-b5 four-agent roster (references/orchestration-patterns.md).
8. A7 components — MCQ: which deterministic engines it calls (risk_matrix, controls, rca, smart_actions, incident_rates) — or none (no scripts/).
9. Trigger keywords — the discovery phrases (reinforce description).
10. Tier + status — MCQ from the tier (1–4) and status (stable/beta/assistive) vocab.

Step 2 — Serialize → scaffold → validate → handoff

• Serialize the answers to an answers.json (the metadata{} block + roster / a7_components / kb_ids flags — see references/mandatory-blocks.md).
• Run scripts/scaffold.py --name <name> --answers <answers.json> (add --from-skill-creator <draft.md> to wrap a skill-creator domain draft, or --standalone to use the templates/* fallback skeletons).
• The scaffolder self-validates: it refuses exit 0 unless its output lints clean.
• Run scripts/validate_repo_skill.py <skill> (the lint_skills.py re-export) to confirm; --sync re-pulls any drifted canonical block from template/blocks/*.
• Hand off assets/forge-handoff.template.json listing the seeded TODOs the author must fill (the domain body, the intake question VALUES, the eval expectations).

This Workflow performs no HSE analysis and circulates no personal data; the deterministic blocks above are inherited because the contract is invariant.

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

• Single-threaded by design — no subagents. The forge's author-interview + scaffold + self-lint runs in one context. (NOTE: this is the forge's own runtime roster — distinct from the BUILD-TIME §3.7 campaign fan-out, where one subagent authors each skill in parallel; the forge SEEDS rosters into other skills via references/orchestration-patterns.md, it does not run a fan-out itself.)

## 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/mandatory-blocks.md — the human-readable six-block contract; what each block is, its marker, and that its text is canonical in template/blocks/*.
• references/orchestration-patterns.md — the roster-authoring menu (single-thread · moderate · flagship-b5), pointing at KB-SNIP-ARCHETYPES.
• references/_skill-kb.md — the knowledge-base fragments this skill resolves.
• references/METHODOLOGY.md — the scaffold→validate→handoff method.
• references/deid-checklist.md — the full de-identification checklist (A5).
• references/QUALITY_CHECKLIST.md — the pre-output validation gate.