AbhiramDwivedi/ai-tools

Skills and small tools I use with AI coding agents. Each skill is a folder with a SKILL.md,
which both Claude Code and Codex read, so one copy serves both.
## Skills
### technical-writer
Drafts, revises and reviews technical prose (blog posts, reference docs, newspaper pieces,
READMEs) so it reads like a sharp engineer wrote it.
I used to ask for three things on every piece: the humanizer skill, Google's developer
documentation style, and Zinsser's principles. The output still needed hand edits, so this
skill adds two sources to those three: research on how AI prose differs from human prose, and
a voice built from the edits AI drafts keep needing.
The most useful research finding is that people who use chatbots a lot spot machine text
almost perfectly, and they judge the whole piece (no point of view, nothing specific, uniform
smoothness) rather than single words. So the skill fixes the thinking first and the wording
second. The sources are in
references/ai-tells.md, along with the
phrases current models overuse from Graphite's October 2026 study, such as "this matters",
"why X matters" and "looking ahead". Those phrases have replaced the em dash as the commonest
sign of a model.
The default voice is in references/voice.md,
with before/after edits taken from revisions of real AI drafts. You can give the skill your
own voice instead (see below). For kinds of piece the voice doesn't cover, such as news
reports, READMEs and release notes, the skill copies writers who do them well. They are listed
in references/craft.md.
The skill isn't meant to fool anyone. It adds a one-line AI disclosure when your voice
profile, you or the publication ask for one, and offers one otherwise. Either way, the prose
has to pass the same checks.
It also ships a linter for the mechanical tells (contrast framing, narrated significance,
trailing "-ing" clauses, em dashes, AI vocabulary, uniform sentence length and more). It needs
only the Python standard library:
sh
python skills/technical-writer/scripts/tells.py draft.md
cat draft.md | python skills/technical-writer/scripts/tells.py -
python skills/technical-writer/scripts/tells.py --json draft.md
A clean run doesn't make a draft good, because the linter can't see a missing point of view,
structural sameness or invented facts. The skill's revision passes cover those.
## Install
macOS / Linux:
sh
git clone https://github.com/AbhiramDwivedi/ai-tools.git
cd ai-tools && ./install.sh # symlinks; add --copy to copy instead
Windows (PowerShell):
powershell
git clone https://github.com/AbhiramDwivedi/ai-tools.git
cd ai-tools; .\install.ps1 # directory junctions; add -Copy to copy instead
Both scripts install into ~/.claude/skills/ and ~/.codex/skills/ (or $CODEX_HOME/skills).
With links, a git pull updates the installed skill.