Report writing

Skill mriffle/findings-ai-collab-workflow/skills/report-writing

How to assemble a report in a Findings Workflow project as a projection of the findings graph: structure, the shared-spec discipline for parallel writers, the per-section write/review loop, and the coherence/assembly pass. Use when running Stage 6 (the stage6-report command) or drafting/assembling report sections.From its SKILL.md

Install

npx -y skills add mriffle/findings-ai-collab-workflow --skill report-writing

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.9 KB, as published. Nobody here has run it

Report-writing procedure

Authoritative rules: conventions/reporting.md. A report is a projection of selected findings into prose — writers select, order, and connect; they never invent content. Methods, figures, and references come essentially for free from the findings' provenance.

Confirm the mode first

QC / data-quality — exhaustive, descriptive; completeness so collaborators trust the data; built largely from Stage 3 outputs + state files.
Research — selective, narrative; a focused disseminable artifact (not a manuscript). Findings deliberately omitted to tell the story.

The selection discipline is inverted between them — everyone writing must know which one.

Structure

Title · Abstract/Summary · Methods · Results · Discussion · References

Results = selected findings rendered into narrative, each sentence carrying its finding id.
Methods = union of those findings' methods + pinned script/environment provenance.
Figures = the exact artifacts the findings already point to (svg/png + legend).
References = the findings' fact-checked references + software/environment citations, deduped and styled at compile time.

Process (matches the stage6-report command)

Select + outline — the scientist chooses which findings the report is about and their order (human checkpoint). Record the chosen finding ids.
Build the shared spec — a glossary, defined abbreviations, the agreed narrative arc, the mode, and per-section finding assignments. This is what keeps parallel writers from drifting in terminology or repeating each other. Distribute it to every writer.
Dispatch per section — one writer + report-reviewer pass per logical section, each with the shared spec and its assigned findings.
Write / review / iterate — until each section passes the report-reviewer (claim-source, status/caveat propagation, no invented references).
Coherence / editor pass — the report-reviewer stitches: resolve repetition, terminology drift, and cross-references. This is real editing, not concatenation.
Assemble the final markdown under reports/. Write the abstract last, since it summarizes the assembled whole.
Optional PDF — pandoc + a CSL citation style for dissemination.

Integrity (non-negotiable)

Every Results/Discussion sentence maps to a finding id; exploratory findings keep exploratory confidence and their caveats; no reference appears that isn't in a source finding. These are enforced by the report-reviewer — write to them from the start rather than fixing them at review.