How to use the Hermes Tweet X/Twitter agent skill

What an agent skill is and what this one does

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a task matches the skill's description. This one, hermes-tweet, tells the agent how to work with the Hermes Tweet toolset.

Once set up, the agent can search a bundled catalog of endpoints, read public posts, accounts, trends and search results, and summarize what people are saying about a topic or account. Private or state-changing actions such as posting, creating monitors or running giveaway draws are meant to be blocked until a human explicitly approves them.

• tweet_explore: searches the bundled endpoint catalog without using the API key.
• tweet_read: calls catalog-listed public read-only endpoints when XQUIK_API_KEY is set.
• tweet_action: handles private reads, writes, monitors, webhooks, extractions, draws and media operations, only when action gating is enabled and you approve.

Install the skill (ZIP upload or folder copy)

Download the hermes-tweet ZIP from this page and unzip it. For a personal setup, place the skill folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

The skill's own setup section also covers the Hermes plugin itself. Install and enable the plugin through Hermes with the command below. Hermes scans plugins during installation and updates; review each warning, since a dangerous verdict blocks installation or disables an update.

Set your API key and action gate

On the Hermes runtime host, set your Xquik API account key in the XQUIK_API_KEY environment variable. Keep credentials in environment variables or the Hermes runtime env file, and never paste them into prompts or tool inputs.

Keep HERMES_TWEET_ENABLE_ACTIONS set to false for read-first sessions. Set it to true only for an approved private read, write, monitor, webhook, extraction, draw or media operation. Restart gateway, cron or long-running Hermes sessions after changing environment variables.

Trigger the skill with natural requests

You do not need special syntax. Describe the goal and the agent will pick up the skill when your request matches its purpose. The skill asks for the objective, target, time window and whether any private or state-changing operation is in scope before it chooses routes.

State clearly that you want read-only work. Phrases like "do not post" or "keep actions disabled" help the agent stay inside the read-first model.

• Research an account before drafting a reply to it.
• Track mentions of a launch keyword today and summarize the themes.
• Draft a launch post and list the exact tool call you would use, waiting for approval before any action.
• Audit the public replies and reposts on a giveaway tweet.

Run a read-first workflow step by step

A typical session starts with discovery, then a public read, then a summary. The agent searches the catalog with tweet_explore, reads the chosen public route with tweet_read, and returns a short report. It should only use catalog-listed paths and should not guess endpoints.

For broader jobs, the skill has reference workflow patterns for social listening, support triage and giveaway audits. Social listening groups results by theme, account, link and recency. Support triage summarizes the problem and drafts replies separately from any action calls. Giveaway audits keep an evidence trail of route paths and result summaries.

• Ask for a Summary, Read routes, Action plan and Next check in the output.
• Ask the agent to list the complete action it would take, then approve or reject it.
• Re-run tweet_explore if a route does not match.

Safety: scripts, network and approval

This skill runs scripts and uses the network. The plugin calls the external Xquik API, so read results depend on that service and on the agent's summaries. Treat outputs as useful context to verify, not as a guaranteed-accurate record of X data.

Before any tweet_action call, the approval checklist requires the endpoint and method, the target, the complete payload without credentials, the expected side effects, and your explicit approval. Use Desktop or gateway operator review for private or state-changing calls. Do not expose credentials in issues, PR comments or prompts.

• Keep HERMES_TWEET_ENABLE_ACTIONS=false unless you are approving a specific operation.
• Review scanner warnings during plugin installation and updates.
• Require approval before creating or changing monitors.