Turn agent output into something a person will actually read.
Ask for a report instead of prose, and you make two decisions the agent otherwise makes for you: the lens (the question you want answered) and the projection (the shape the answer arrives in). The result is one self-contained, on-brand HTML file you can open from a bare path, share as a link, or email.
| Skill | What it does |
|---|---|
/html-report |
Picks a lens and a projection with you, gathers real context, renders one shareable HTML file |
/writing-style |
The house prose style the reports depend on. Also useful on its own for status updates, Slack posts, and client communication |
Both are installed together. html-report defers prose style to writing-style,
so installing only the first gives you reports that look right and read like
generic AI output.
- Reviewing a pull request, a plan, or a spec
- Explaining something technical to somebody who cannot hold the detail
- Post-mortems and incident write-ups
- Content inventories and migration analysis
- Synthesising workshop notes and transcripts
- Onboarding into a codebase you have just inherited
/plugin marketplace add darylldoyle/html-report
/plugin install html-report@html-report-marketplace
Copy both skill directories into your skills folder:
git clone https://github.com/darylldoyle/html-report.git
cp -r html-report/html-report html-report/writing-style ~/.claude/skills/Point it at something and say what you want. If you leave out the lens or the projection, it will ask.
/html-report review this PR for correctness and risk as an annotated diff
/html-report explain why the new products aren't showing on the site,
for the client, with a diagram
/html-report where did people actually disagree across these workshop notes,
as a decision map with owners and next actions
Without the skill installed, the prompt shape is:
Give me a self-contained HTML report on the reliability of this change (edge cases, tests, performance), as a sequence diagram of where it could break. Cite the file and line for each finding, and separate confirmed failure modes from things I should check myself.
Every report cites what it read and draws a visible line between confirmed findings and hypotheses. That gives you the material to check the claims that matter. It does not check them for you, and a well-designed report is more persuasive than it has earned the right to be. Read the sources line.
html-report/SKILL.md carries the brand tokens and logo inline, which is why
reports come out looking like they came from us. To rebrand, edit the brand
token block and the inlined logo SVG in that one file.