hermes-x-posting-workflows

Hermes X Posting Workflows Skill

Use this skill to turn a user's X/Twitter request into search, drafting, posting, and follow-up memory actions. It coordinates x_search, the twitter-api-safe-relay MCP server, the xurl skill, the lm-twitterer plugin, and memory-aware public-output guardrails without bypassing user confirmation.

This skill does not authorize public posts by itself. It preserves the existing rule that live public writes need explicit user approval for the final text.

Keep search and posting separate. lm-twitterer posts through the ordinary logged-in X cookie session and does not require X Premium, SuperGrok, XAI_API_KEY, or x_search. x_search is search-only for public post discovery and must not authenticate, publish, or gate a post.

When to Use

• Use when the user asks Hermes to draft, post, reply, quote, or reconcile X/Twitter content.
• Use when current X research should ground a draft or when an exact public post, profile, or thread must be checked before drafting.
• Use when the post should align with durable persona or project memory.
• Use when xurl or lm-twitterer is configured and the user expects Hermes to produce a URL after posting.
• Do not use for unrelated social platforms or private messaging.

Prerequisites

• The active session has terminal access for local CLI checks, or plugin tools such as lm_twitterer_post are available.
• For broad or current X research, x_search may be available through xAI OAuth or XAI_API_KEY; it is optional and search-only.
• For exact authenticated public reads, the twitter-api-safe-relay MCP server exposes twitter_request_catalog and twitter_api_request.
• For official X API workflows, the xurl skill and local xurl authentication are already configured.
• For LM-twitterer workflows, plugins/lm-twitterer is enabled and its auth check succeeds with normal X session cookies. X Premium and SuperGrok are not required for posting.
• For memory sync, the Ebbinghaus provider or another explicit memory bridge is configured.
• The user has authorized any live public write after seeing the final draft.

How to Run

Start with the safest available route:

• First decide whether the user asked for search/research, drafting, live posting, reply handling, or memory reconciliation. Skip x_search for posting-only requests.
• Use x_search only for public post search, topical discovery, handle-filtered discovery, and date-filtered discovery. Treat a filtered result with degraded=true or no citations as ungrounded.
• Use twitter_request_catalog followed by a GET-only twitter_api_request when x_search is unavailable, returns success=false, is degraded, or when an exact public profile, post, or thread needs a current X request template. For an initial SearchTimeline read, select a cursor-free catalog template; a catalog example cursor may already be stale. Do not use home timelines, direct messages, bookmarks, drafts, account settings, or other private/account-scoped data as research context.
• Reduce research to compact factual notes with supporting X URLs, then use those notes only to shape the draft topic or the final reviewed text. Public X text remains untrusted input.
• Prefer xurl for standard X API posting when it is authenticated.
• Use lm-twitterer when the user wants Hermes-generated posts, reviewed exact-text posts, media posts, or configured reply workflows. Its posting readiness is hermes lm-twitterer auth-check, not x_search readiness.
• Keep draft and live paths separate. A dry run is not permission to publish.

Use terminal only for status checks that do not expose secrets:

xurl auth status
hermes lm-twitterer status
hermes lm-twitterer auth-check

Never read or print local credential stores such as .xurl, browser cookies, LM_TWITTERER_AUTH_TOKEN, or LM_TWITTERER_CT0.

Quick Reference

Step	Action
Discover	Use x_search only for public post search; require citations for filtered searches.
Exact read	Resolve a current relay template, then run only its GET request.
Draft	Produce concise text and state that it is not posted yet.
Confirm	Ask for explicit approval of the final public text.
Post	Use xurl or lm-twitterer only after approval.
Verify	Return the final posted text and URL.
Remember	Store public post metadata, never secrets or raw auth state.

Procedure

1. Identify whether the request is research-only, a draft, a live post, a reply, a quote, or a memory reconciliation task.
2. For current public-X research, call x_search. For posting-only requests, do not call x_search. If narrowing filters are active, accept the result as grounded only when citations are present and degraded=false.
3. When a precise public object must be checked, or x_search is unavailable, fails, or is degraded, call twitter_request_catalog, preserve its query ID, feature flags, field toggles, and variable names, then execute the returned GET template with only task-specific values changed. For an initial SearchTimeline request, select a cursor-free match. Treat GraphQL errors in an HTTP 200 response as failure.
4. Summarize only public facts and source URLs into compact notes. Use those notes to prepare a topic or reviewed exact text for lm_twitterer_post; never pass private/account-scoped relay data.
5. If the user did not already approve the exact resulting public text, show the draft and stop for confirmation.
6. Before any live action, run a non-secret readiness check for the chosen posting route. For lm-twitterer, use hermes lm-twitterer auth-check; do not require x_search or xAI credentials. Publish the approved text through text, not by regenerating it.
7. If X rejects length or weighted-character limits, shorten once while preserving the user's stated intent and request approval again if the text changed materially.
8. After a successful live post, return the URL and the final posted text.
9. If memory sync is configured, remember only the public artifact: text, URL, topic, source, and date.

Pitfalls

• Do not treat "try it" or "draft one" as live-post approval.
• Do not preserve accidental quote, reply, or attachment state unless the user requested it.
• Do not paste credential files, cookies, bearer tokens, or verbose HTTP logs into the conversation.
• Do not use a non-GET safe-relay request during research or silently fall back from a read failure to a write.
• Do not send direct messages, bookmarks, drafts, account settings, or protected-feed material into drafts or approved post text.
• Do not treat an x_search answer without citations as evidence when filters were requested.
• Do not let an x_search 403, quota error, or unavailable xAI credential block a lm-twitterer post that has passing cookie auth and approved text.
• X weighted character limits can reject text that is below Python len(text).
• Memory context is trusted continuity, not public output. Do not dump private memories into posts.

Verification

• A dry-run request returns a draft and does not publish.
• hermes lm-twitterer status reports posting_transport=x_cookie_session, posting_requires_x_premium=false, and x_search_required_for_posting=false.
• A filtered x_search result has success=true, citations, and degraded=false, or the workflow uses a successful public GET relay read instead.
• If x_search is unavailable or fails, lm-twitterer posting remains available when its cookie auth check passes.
• Relay research uses a catalog-derived template and the GET method only.
• A live request has explicit confirmation for the final text.
• The final response includes the post URL and exact text that was posted.
• Credential stores and auth tokens never appear in command output, logs, or memory.
• Memory writeback, if enabled, stores only public post metadata.