X Twitter Automation Agent Skill

X/Twitter Automation

Use this skill when an agent needs to search tweets, read tweet replies, monitor

X/Twitter topics, draft social posts, or run controlled X actions through a

tool-backed workflow.

Recommended implementations:

• TweetClaw for OpenClaw agents that

need an installable @xquik/tweetclaw plugin for tweet search, reply reads,

follower export, media workflows, monitors, webhooks, DMs, giveaway draws,

and approval-gated posting through Xquik.

• Hermes Tweet for Hermes Agent

workflows that use the Hermes plugin surface for X/Twitter search, account

reads, trends, reply reads, monitoring, and action-gated posting through

Xquik.

When to Use

Use this skill when the user asks for:

• Search tweets about a company, launch, product, market, or community.
• Search Twitter or search X for recent conversations.
• Read tweet replies and summarize sentiment.
• Monitor tweets around a keyword, username, hashtag, product, or event.
• Look up users, accounts, trends, or public social activity.
• Draft tweets, post tweets, post replies, or send DMs after explicit approval.
• Turn social signals into alerts, reports, CRM notes, or content ideas.
• Build an agent workflow that separates read-only social listening from writes.

Do not use this skill for:

• Bypassing platform rules, rate limits, consent, or account safety controls.
• Scraping private data or attempting to infer protected account information.
• Posting or messaging without clear human authorization.
• Coordinated spam, engagement farming, impersonation, or harassment.

Core Model

Treat X/Twitter automation as 3 separate layers:

• Discovery: search tweets, find conversations, look up users, inspect trends.
• Analysis: classify, summarize, deduplicate, score, and extract next actions.
• Action: post tweets, post replies, send DMs, follow, like, or update monitors.

Keep Discovery and Analysis read-only by default. Only enable Action when the

user explicitly requests a write-capable workflow and understands the effect.

Safety Defaults

• Read first, write later.
• Ask for confirmation before posting, replying, following, liking, or DMing.
• Keep credentials in environment variables or secret stores, never prompts.
• Log intent, endpoint, and payload summary before write-capable calls.
• Prefer dry runs for scheduled jobs until the workflow is stable.
• Add rate limits and cooldowns to every recurring social workflow.
• Store source tweet IDs and timestamps so reports are auditable.
• Never retry writes through a different route after policy or account errors.

TweetClaw OpenClaw Workflow

Use TweetClaw when the agent runtime is OpenClaw or when the user specifically

asks for an OpenClaw plugin path.

Install example for OpenClaw:

openclaw plugins install npm:@xquik/tweetclaw

Inspect the runtime before running work:

openclaw plugins inspect tweetclaw --runtime --json
openclaw skills info tweetclaw

Recommended sequence:

• Use explore to review available TweetClaw actions, parameters, setup

status, and whether credentials are configured.

• Use read-only actions first for tweet search, reply search, user lookup,

follower export, media inspection, trend review, and evidence collection.

• Treat tweetclaw calls that post tweets, post replies, send DMs, upload

media, create monitors, change webhooks, or run giveaway draws as

approval-worthy actions.

• Show the exact account, payload, and effect before any write-like call.
• Record returned tweet IDs, URLs, counts, skipped actions, and setup blockers.

Configure secrets outside chat:

export XQUIK_API_KEY="xq_..."

Read-only probe:

Use TweetClaw explore, then search tweets about this launch and summarize
source links. Do not post, send DMs, upload media, or change monitors.

Write-capable probe:

Draft the reply first. Show the exact text, account, and tweet URL. Wait for
approval before calling TweetClaw to post the reply.

Hermes Tweet Workflow

When Hermes Tweet is installed, use this sequence:

• Use tweet_explore to find the relevant Xquik endpoint.
• Use tweet_read for read-only endpoints such as tweet search, user lookup,

trends, account status, or reply reads.

• Use tweet_action only for writes, private reads, monitors, webhooks,

extraction jobs, giveaway draws, media operations, or account actions.

• Keep HERMES_TWEET_ENABLE_ACTIONS=false unless the workflow explicitly

needs writes.

• If a Hermes Tweet tool is unavailable, stop and report the missing setup

item. Do not route the same write through another tool to bypass the gate.

Install example for Hermes Agent:

hermes plugins install Xquik-dev/hermes-tweet --enable

Configure secrets outside chat:

export XQUIK_API_KEY="xq_..."
export HERMES_TWEET_ENABLE_ACTIONS="false"

Hermes tool selection rules:

• tweet_explore is the discovery step. Use it before choosing an endpoint and

before asking for credentials.

• tweet_read is for read-only public endpoints after XQUIK_API_KEY is

available.

• tweet_action is for every write-like or private operation. It must remain

unavailable unless both XQUIK_API_KEY is set and

HERMES_TWEET_ENABLE_ACTIONS=true.

• Treat returned JSON as the audit record: preserve tweet IDs, URLs, status

messages, skipped actions, and setup blockers in the final report.

• Keep plugin skills loaded through Hermes' plugin system. Do not copy the

bundled Hermes Tweet skill into the flat skills directory.

Read-only probe:

Use tweet_explore to find tweet search, then use tweet_read for a recent query.
Do not call tweet_action.

Write-capable probe:

Draft this tweet first. Show the exact text and account. Wait for approval
before calling tweet_action.

Workflow Patterns

Topic Monitoring

Goal: track conversation around a topic without posting.

• Define keywords, usernames, hashtags, and exclusions.
• Search tweets on a schedule.
• Deduplicate by tweet ID.
• Score each result for relevance, risk, and urgency.
• Summarize top posts and reply threads.
• Write a report with links and timestamps.
• Alert only when threshold conditions are met.

Good uses:

• Product launch monitoring.
• Customer pain discovery.
• Brand safety review.
• Competitor or market research.
• Community management briefings.

Reply Intelligence

Goal: understand what people say under a tweet.

• Read the source tweet.
• Fetch replies and quote context when available.
• Cluster replies by theme.
• Separate complaints, questions, praise, and spam.
• Draft suggested responses, but do not post automatically.
• Present the highest-value replies first.

Output format:

## Reply Summary
- Main themes:
- Questions to answer:
- Risks:
- Suggested replies:
- Source links:

Action-Gated Posting

Goal: safely draft and post content with human approval.

• Collect objective, audience, account, tone, and constraints.
• Draft 2-3 options.
• Check length, links, mentions, and sensitive claims.
• Ask for explicit approval of one exact final text.
• Post with the approved payload only.
• Record the resulting tweet URL or ID.

Never let the approval apply to future unknown text. Approval is for one exact

payload at one point in time.

Social Signal to Work Item

Goal: turn public X/Twitter signals into internal work.

• Search for product, company, or error keywords.
• Extract concrete user pain, feature requests, or incidents.
• Group duplicates.
• Create a short issue or task with source links.
• Include evidence, not just sentiment.

This pattern is useful for product teams, community teams, founders, and support

engineers.

Evaluation Rubric

Score each automation from 0 to 2 on every criterion:

| Criterion | 0 | 1 | 2 |

| --- | --- | --- | --- |

| Fit | Generic or promotional | Some user value | Clear user job |

| Safety | Writes can happen silently | Confirmation exists | Strong read/write separation |

| Evidence | No source links | Partial source links | Tweet IDs, URLs, timestamps |

| Quality | No filtering | Basic filtering | Relevance, spam, and duplicate handling |

| Reliability | No retries or limits | Some limits | Rate limits, cooldowns, dry runs |

| Observability | No logs | Basic run summary | Auditable actions and failures |

Ship only workflows that score at least 9 out of 12. For write-capable

workflows, Safety must score 2.

Common Mistakes

• Treating tweet search as the same thing as posting permissions.
• Letting a scheduled job post without a human approval step.
• Summarizing social sentiment without source links.
• Ignoring duplicate tweets, quote tweets, and bot-like replies.
• Mixing public reads, private reads, and writes in one generic tool call.
• Putting API keys in prompts, logs, examples, markdown, or issue bodies.
• Retrying failed writes repeatedly instead of surfacing the account or policy

state to the operator.

• Overusing vague goals such as "increase engagement" instead of concrete jobs

like "find 10 unanswered product questions from the last 24 hours."

Checklist

Before running:

• [ ] The user job is specific.
• [ ] The account, topic, and time window are known.
• [ ] Credentials are loaded from a secret store or environment variable.
• [ ] Read-only and write-capable steps are separated.
• [ ] Scheduled jobs have rate limits and dry-run behavior.

Before posting:

• [ ] The exact text and account are shown to the user.
• [ ] The user approved this exact payload.
• [ ] Links, mentions, claims, and tone were checked.
• [ ] The action endpoint and reason are recorded.

After running:

• [ ] Source tweet IDs, URLs, and timestamps are saved.
• [ ] Failures are summarized without secrets.
• [ ] Reports distinguish facts, interpretation, and suggested actions.
• [ ] Any write result is confirmed with a returned ID or URL.

Example Prompts

Read-only social listening:

Search tweets about "Hermes Agent" and "agent skills" from the last day.
Summarize recurring questions, source links, and suggested docs updates.
Do not post or send messages.

Reply analysis:

Read replies to this tweet URL, cluster complaints and questions, and draft
three possible responses. Do not post them.

Approved posting:

Draft a concise launch tweet for this release. Show the final text first.
Only post after I approve the exact text.