reading-pagelisted
Install: claude install-skill Traction-AI-Labs/skills
<!-- SOURCE-REVISION 9c099e97 -->
# Reading page
People read long agent output far better as a designed page than as raw markdown. Anything longer
than a chat message that the reader is meant to read goes out as one self-contained HTML page: a
contents rail, the bottom line first, callouts, sortable tables, folded evidence, light and dark.
The markdown can stay the agent's record beside it; the page is how the reader reads it.
**Use it for:** research reports, plans, briefs, handoffs, reviews, summaries, investigations,
anything longer than a chat message that they are meant to read.
**Not for:** chat replies (stay in chat); decisions they must rule on (`ruling-page`; a report that
also needs rulings links to its ruling page from a decision callout); call prep (`call-prep`).
## Build it
1. Write the report as markdown beside where it belongs (the project's repo: `research/`, the
plan or status folder). The markdown stays the agent record and goes in git.
2. Build the page beside it, same name, `.html`:
```bash
uv run <skill-dir>/scripts/build.py <report>.md \
--author "Claude" --status "Research complete"
```
`<skill-dir>` is the folder holding this SKILL.md. Output defaults to `<report>.html`. Dependencies are pinned in the script header (PEP 723),
so `uv run` needs nothing installed. Optional `--title`, `--date YYYY-MM-DD`, `--template`;
front matter (`title`, `date`, `author`, `status`) works too; `status` also sets the page state, which ever