--- name: write-report description: Synthesise research or data into a polished single-file B&W HTML report — BLUF, evidence, recommendation, confidence, source-evidence index. Verified by default. visibility: public type: draft director: CIO base: draft url: navigateengine.com/skill/write-report --- # Write a Report **The CIO's report-writing verb.** Take research, data, or a brief and produce a polished single-file HTML deliverable that a decision-maker can read in under ten minutes and act on. The report must lead with the verdict and end with the evidence index — never the reverse. --- ## input A research brief, a dataset, a meeting transcript, or a prior draft. Examples: - `/write-report from the gather-info output above` - `"Write me a board-pack-style update from these three GA4 + Ads CSVs"` - `"Turn this 40-page consultant deck into a one-page report for the CEO"` If the input is sparse, ask the user **one** sentence: audience + decision the report enables. Then proceed. --- ## process — ASK → PREFLIGHT → DRAFT → VERIFY → AUDIT 1. **ASK** — identify the audience and the decision this report enables. One round only. 2. **PREFLIGHT** — emit the output skeleton (BLUF + N section headings + recommendation + confidence + sources) before writing a single paragraph. 3. **DRAFT** — fill the skeleton, applying the structure below. 4. **VERIFY** — if the report contains ≥ 5 numeric claims, invoke `/verify` against the draft and embed the verdict matrix above the Sources block. 5. **AUDIT** — run the contract. Revise until all items pass. Show the audit table inline. ### Steps inside DRAFT 1. **BLUF** — one sentence, the single most important finding, with the verdict in **bold**. 2. **Sections** — 3 to 5, ordered by importance. Each section: **finding** (one sentence) → **data** (cited) → **implication** (one sentence on what changes). 3. **Recommendation** — actionable, owner-assignable, deadline-bearing. No "consider exploring..." 4. **Confidence** — `HIGH` / `MEDIUM` / `LOW`, justified by source quality and source agreement. 5. **Next action** — one concrete step, with owner and timeframe. --- ## output A single self-contained HTML file using the NavigateEngine **B&W skin** from the canonical styleguide (`navigate-engine.s3.ap-southeast-2.amazonaws.com/ne-brand/styleguide.html`). ### inline skeleton ```html {{ report_title }}

CIO · REPORT

{{ report_title }}

{{ audience }} · {{ date }} · {{ n_sources }} sources

BLUF: {{ one-sentence finding with verdict in bold }}

1. {{ section_title }}

Finding: {{ one sentence }}.

Data: {{ cited evidence }} [1]

Implication: {{ what changes }}.

Recommendation

{{ actionable, owner-assignable, deadline-bearing }}

Confidence

{{ HIGH | MEDIUM | LOW }} — {{ justification }}.

Verification

{{ embedded /verify verdict matrix when ≥5 numeric claims }}

Sources

    {{ [n] cited source · publisher · date · URL }}
``` The canonical template at `navigateengine.com/skill/write-report/assets/template.html` ships with all B&W tokens inlined, all sections pre-styled, and slot markers (`{{ ... }}`) that the agent fills. --- ## contract - Audience explicitly identified. - BLUF is one sentence with the verdict in ``. - 3–5 sections, ordered by importance. - Each section has explicit `Finding`, `Data`, `Implication` blocks. - Every numeric claim carries `[n]` citation OR `[UNVERIFIED]`. - Recommendation names owner + timeframe. - Confidence is `HIGH` / `MEDIUM` / `LOW` with a justification clause. - If ≥ 5 numeric claims: `/verify` verdict matrix embedded. - Sources block contains 5–15 entries. - Output is a single `` document with no external runtime dependencies (CSS may be linked). --- ## inversion If your recommendation aligns with the user's pre-existing direction, name the condition under which it would be wrong. If you cannot, the report has not earned its weight.