cli-faq-shortcuts reads your Claude Code and Codex history and writes the slash commands you keep asking for
Kishor Morol has released cli-faq-shortcuts, an Apache-2.0 agent skill that reads your Claude Code and Codex prompt history in a project, groups the questions you keep asking, and writes each one as a short slash command committed to the repo.
Image: GitHub
Why it mattersRepeated requests to an AI coding agent become short slash commands that already point at the project's own scripts and record the mistakes made on that path.
Every developer using an AI coding agent has a set of questions they type again and again, in slightly different words each time, and each retype loses the small lesson from the last one. Kishor Morol has released cli-faq-shortcuts, an Apache-2.0 agent skill that reads your Claude Code and Codex history in a project, groups your repeated asks by intent, and writes each one as a short slash command committed to the repository. The repository was created on 18 September, was last pushed on 22 September, and carries 103 stars, 2 forks and 4 open issues on GitHub as of 24 September, six days after it was published.
What it does, in the author's own numbers
The skill runs one Python extractor over your Claude Code and Codex prompt history for the current project, then hands the list to an agent that groups the prompts by intent no matter how they were worded. The README states that on one real project, the most repeated exact request appeared only four times, and the top ask grouped by intent had been typed 127 times, roughly 32 times as often. That is why the grouping step exists: string counting misses most of the repetition.
The output is a table: shortcut name, the ask it replaces, and how often the writer asked it. The writer picks. Each accepted shortcut becomes a .claude/skills/<name>/SKILL.md file under the project, linked into ~/.agents/skills/ so Codex reads the same skill, and Cursor reads both directories. The README's own example turns "did the newsletter go out, did it reach everyone?" into /sent, "how many people signed up this week, not counting test accounts?" into /signups, and "any update on the billing migration, what's blocked?" into /project billing.
What the SKILL.md file carries
Each shortcut file the skill writes names the script or query the project already has, records the traps already found on that path, and states what to report. The README gives one example: a sent shortcut points at scripts/email-status.js --batch <name>, notes that "a 'delivered' status to a shared inbox proves nobody read it (happened 3 May)", and asks for sent, delivered and bounced counts plus the bounced addresses. Anything that writes or sends runs a dry pass first and waits for a yes, so a typo stops at the preview screen.
The extractor is local. The README states that nothing leaves the machine at that step, because the extractor reads the history files and prints them on standard output for the agent in the same session.
Why the shape works for a team already using these agents
A shortcut committed to the repository is a piece of shared context: the next teammate to type "did the newsletter go out" runs /sent instead, and the trap about shared inboxes is already in the file. The skill mines the reader's own history, so the shortcuts already reflect the language the team uses and the scripts the project already has, and they work the first day they land in the repository. A team that has been using Claude Code or Codex for more than a few weeks and already types the same requests each day is the case where this skill saves the most typing.
Source
- Kishor Morol, kishormorol/cli-faq-shortcuts on GitHub, Apache-2.0 licence.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
