write-technical-report

a skill to take an experiment's logs, outputs and artifacts, and write them into a technical report.

new to this? open a folder holding the experiment's logs, outputs and artifacts in Claude Code or Codex, and say:

use the repository at https://github.com/torsor/claude-plugins and write a technical report from the material in this folder

a Claude Code skill. give it an experiment’s protocols, run logs, outputs, tests, evaluator notes, interventions, and repository state, and it writes them into a reproducible technical report: the question, the methods, the results, the validation, the failures, the limitations, and the next experiment.

it is written for a reader who does not believe you yet. provenance is frozen first, then every run and artifact is audited into an evidence ledger, and the ledger gates what may be claimed. only then does it fix a six-sentence spine and a section plan for you to approve, and only then does it write.

failures and limitations are first-class sections rather than a closing concession.

the shape

phase 0   scope the report, freeze provenance   ->  experiment brief + reader profile
phase A   audit runs, artifacts, validation     ->  report-notes/*.md   (evidence gate)
phase B   six-sentence spine and section plan   ->  you approve the claims
phase C   author, build, inspect                ->  the report, four formats

invoke

/write-technical-report the overnight run in ~/experiments/2026-09-sweep/

produces

a report in the house format — LaTeX source with PDF, HTML, EPUB and Markdown — with the run manifest, evidence ledger, and artifact index kept alongside.