Human Content Prompts
Slash commands that generate writing prompts from session context.
Agents Prompting Humans
I want to write and share more of what I do, and I live in VS Code, Obsidian, and an agent TUI. While the dopamine drips from the AI pair programming session with infinite possibility, when is there time to write anything down?
My thesis is that reducing friction in the writing process using the data we have already collected in a coding session has the potential to turn anyone into an author.
So today I added a /devlog and /article command into my Claude user settings that reviews the work we have done via the agent session we are already in along with the recent git history. And if I don’t write more, at least I have a snapshot of what I thought was important enough to share.
What Changed
Created two Claude Code slash commands (~/.claude/commands/devlog.md and article.md) that generate content drafts from session context. Each command gathers conversation history and recent git activity, presents topic options, checks whether the topic fits devlog or article scope, and writes a properly frontmattered markdown file to the correct content directory of our developer website, and references the project that was being worked on.
Notes
- Commands live in
~/.claude/commands/(user-level, not project-level) /devlogtargetssrc/content/devlog/,/articletargetssrc/content/articles/- Both use
AskUserQuestiontool to confirm topic selection and content-type fit - Word targets: devlog 150-300, article 800-1500
- Content-type crosscheck: each skill flags if the topic belongs in the other format
- Always sets
draft: true— these are writing prompts, not published posts