x-thought (independent legacy skill)

An autonomous X writing skill for Hermes Agent. It is not the current ReplyNodes MCP, does not publish or schedule content, and is retained separately from ReplyNodes' canonical agent context surface.

It does not generate content to fill a calendar. It reads sources, decides whether there is a real thought worth expressing, and returns SKIP when there is not.

What it does

x-thought combines the complete writing runtimes of:

• blader/humanizer: voice-aware rewriting and a two-pass audit for AI-writing patterns.
• sergebulaev/x-skills: X-native posts, threads, replies, quote posts, hook study, repurposing, planning, profile work, and audience insight.

On top, x-thought adds an editorial decision layer:

sources → signals → thought selection → X draft → humanizer → X audit → QUEUE | HOLD | SKIP

A post is earned. Silence is a valid output.

Modes

Mode	Use
thought-scan	Read source feeds, research, discussions, or news and find a thought worth sharing.
post	Write one native X post from a supplied claim or idea.
thread	Develop multiple independently useful claims into a thread.
reply / quote	Add a useful reply or quote post; skip empty agreement.
hook study	Study a post's mechanism and make original options.
repurpose	Rebuild an article, LinkedIn post, video, newsletter, or notes for X.
plan	Create an editorial X plan that leaves room for reactive posts and SKIP.
profile	Write an X bio, header guidance, and pinned post.
audience insight	Find questions, language, and gaps in available audience or competitor data.

Autonomous output

<MODE IN UPPERCASE>
<the X text>

WHY NOW: <one source-grounded sentence>
CONFIDENCE: high | medium
ACTION: QUEUE | HOLD
SOURCE: <URL or source identifier>

For thought-scan, the skill returns:

SKIP — <brief reason>

when it cannot form an evidence-backed, personally plausible, non-obvious thought.

QUEUE is reserved for high-confidence drafts that pass the editorial selection, Humanizer, and X-native checks. HOLD is for drafts that should not enter the queue.

Install in Hermes

Clone this repository into Hermes's skills directory:

git clone https://github.com/replynodes/x-thought.git ~/.hermes/skills/x-thought

Or add the repository path to Hermes's skills.external_dirs configuration.

Then use it in an agent request:

Use x-thought in thought-scan mode. Read these sources, find at most one thought worth queuing, otherwise return SKIP.

Design rules

• Never invent a personal experience, opinion, metric, customer conversation, or certainty.
• Do not summarize a source and call it a thought.
• Do not manufacture contrarianism, vulnerability, takeaways, or engagement bait.
• Match the user's voice profile rather than generic “human” prose.
• Run the bundled Humanizer rewrite and final audit on every draft.
• Use X-native structure, not compressed LinkedIn copy.
• Do not publish, schedule, or manage accounts. ReplyNodes owns execution.

Repository structure

SKILL.md                         orchestration and editorial gate
references/                      x-thought's selection and voice profile
components/humanizer/            complete blader/humanizer runtime
components/x-skills/             complete sergebulaev/x-skills runtime

The bundled upstream components retain their MIT licenses:

• components/humanizer/LICENSE
• components/x-skills/LICENSE

ReplyNodes integration

ReplyNodes should treat x-thought as the thinking and writing layer:

1. Collect sources and user context.
2. Invoke x-thought.
3. Persist the output, source, result, and outcome.
4. Automatically queue only ACTION: QUEUE according to the product's policy.
5. Feed publish, rejection, edits, and performance outcomes back into the user's voice and judgment profile.

Publishing, scheduling, social API access, and account management belong outside this repository.