The DotShare agent skill
The DotShare skill teaches your AI agent to write strong posts for every platform and to publish them the safe way: a draft, a preview you can read, and nothing public until you say so. It works with any agent that reads skills, on any model, and setup installs it for you.
What your agent does with it
| Without the skill | With the skill |
|---|---|
| Generic text, the same on every platform | A native shape per platform: a thread on X and Bluesky, one rich post on LinkedIn with the link in the first comment, CamelCase hashtags on Mastodon |
| Long explanations, or none | Talks at your level: just the result if you want it done, what each step does if you ask how it works |
| Choices buried in paragraphs | Choices side by side in a small table, with a recommendation first: thread or single post, post now or schedule, where the link goes |
| Previews as walls of text | One summary row per account (format, length, media, cost), then the exact text |
| Might promise what DotShare can't do | Says up front what isn't possible, such as replying to other people's posts |
| Guesses after a timeout | Checks what went out before doing anything, and never posts twice |
It also brings the details agents usually get wrong: each platform's character and media limits, Dev.to article metadata, the X budget when X posts go through DotSuite, and how to recover a post that timed out.
Install it
The skill comes with dotshare-cli. Setup installs it where your agent loads skills,
along with the MCP server:
npm install -g dotshare-cli
dotshare-agent setup # Codex
dotshare-agent setup --client claude # Claude Code
dotshare-agent setup --client cursor # Cursor
dotshare-agent setup --client antigravity # Antigravity| Agent | Skill location |
|---|---|
| Claude Code | ~/.claude/skills/dotshare |
| Codex | ~/.codex/skills/dotshare (or $CODEX_HOME/skills/dotshare) |
| Cursor | ~/.cursor/skills/dotshare |
| Antigravity | ~/.gemini/config/skills/dotshare |
Any other agent: the skill is a standard SKILL.md folder (the open Agent Skills
format), so point setup at the directory your agent reads skills from:
dotshare-agent setup --client generic --skills-dir ~/.agents/skills--client generic also prints the MCP server entry to add to your agent. Start a new
agent session afterwards so it loads the skill.
Keep it up to date
Run setup again after you upgrade dotshare-cli. It replaces the old copy of the
skill as a whole, in one step. To check it, run doctor with the same client:
dotshare-agent doctor --client claudeDoctor warns when the skill is missing, from another DotShare version, or replaced by a different skill with the same name.
Your own skill named dotshare
If you already have a different skill named dotshare, setup keeps it and tells you.
--replace replaces it and saves a backup in ~/.dotshare/skill-backups, outside
your agent's skills folder so it isn't loaded twice. A symlinked skill is never
written through, and --no-skill skips the skill entirely.
What's inside
SKILL.md stays short: how to talk to you, how to compare options, the workflow and
the safety rules. The agent opens the rest only when it needs it:
| File | Covers |
|---|---|
references/writing.md | Hooks, the best shape for each platform, style rules, a worked example of one source adapted everywhere, Dev.to articles, other languages |
references/platforms.md | Text and media limits per platform, and what X costs through DotSuite |
references/tools.md | The exact fields of every DotShare tool |
references/scheduling.md | Posting later through DotSuite, and what can't be scheduled |
references/recovery.md | Timeouts, partial failures and duplicates |
references/setup.md | Setup, accounts and images |
references/cli.md | The same workflow with terminal commands, when MCP isn't connected |
The limit tables are checked against DotShare's code by a test, so they change when the code does. Read the skill on GitHub.
For the full CLI and MCP reference, see Set up the CLI and MCP.