Cold dm

X Outreach

Find the right people on X, write them a DM worth replying to, and send it at a safe, human pace — all through the XAutoDM API with the user's own API key. This skill does the finding, the writing, and the sending; it is opinionated about doing outreach in a way that gets replies and keeps accounts healthy, and it refuses to help with spray-and-pray.

Two kinds of work live here:

• Keyless value (free, no spend): defining an ICP, drafting personalized DMs, planning a sequence. These run with no API key and cost nothing.
• API actions (metered, paid): anything that touches X — scraping leads, sending DMs, reading the inbox, follow/like/reply. These call the live API with the user's key and spend credits. The skill always estimates the cost and gets an explicit yes before spending.

The API in one screen

• Base URL: https://api.xautodm.com/v1 · Docs: https://xautodm.com/api/docs
• Auth: Authorization: Bearer $XAUTODM_API_KEY. Keys look like xdm_live_… (real) or xdm_test_… (dry-run: validates and returns shapes, makes no upstream writes and spends no credits — use it to verify a setup, never as a "free trial").
• Get a key: self-serve at https://api.xautodm.com/?ref=claude-skill (paid, metered, cost-transparent — see "How keys and billing work" below).
• Every write (/dm/send, /actions/*, /tweets, /accounts connect) needs an Idempotency-Key: <uuid> header so a retried call is never double-sent or double-charged.
• Success responses carry a meta object: {credits_charged, credits_remaining, request_id}. Read credits_remaining after each call so you always know the running balance.
• Errors always look like:

  {"error":{"type":"insufficient_credits","message":"...","request_id":"..."}}
  

  type ∈ invalid_request · invalid_key · insufficient_credits · rate_limited · account_needs_reconnect · seat_limit · upstream_error · not_found · server_error. On any error: surface the message to the user and stop — do not auto-retry. The one exception is rate_limited (429): back off a few seconds and retry once. insufficient_credits (402) → tell them the balance is short and link the top-up; account_needs_reconnect (409) → route to /accounts/{id}/reconnect; seat_limit (403) → they need a bigger plan.

Credit costs (memorize these — estimate before every spend)

1 credit = $0.005. These per-action costs are stable; use them to price a campaign before it runs.

Action	Endpoint	Credits	≈ USD
Read a profile / tweet / relationship	GET /users/{h}, /tweets/{id}, /relationship	5	$0.025
Lead row (follower / following / search / mention / list member)	/users/{h}/followers, /following, /users/search, /mentions, /lists/{id}/members, /tweets/search	5 / row	$0.025
Premium lead (verified-follower / retweeter / replier / tweet-engager)	/verified-followers, /tweets/{id}/retweeters, /tweets/{id}/replies	20 / row	$0.10
Send a DM	POST /dm/send	10	$0.05
Read a DM conversation	GET /dm/conversations, /dm/conversations/{id}	10	$0.05
Action (follow / unfollow / like / retweet / tweet)	POST /actions/*, POST /tweets	5	$0.025
Connect / reconnect an X account	POST /accounts/connect, /accounts/{id}/reconnect	50	$0.25
Health · list accounts · account · usage	/health, /accounts, /account, /usage	0	free

These are the numbers to quote. Ignore any older per-call figures in the raw OpenAPI summaries — the table above is the source of truth. references/api-reference.md has the full endpoint list. Tier prices drift — do not quote a plan's dollar price from memory; point the user to their live plans at https://api.xautodm.com/?ref=claude-skill. The per-action costs above are stable.

Estimating a campaign (do this out loud before spending): pulling N leads from a regular source ≈ N × 5 credits; from a premium source ≈ N × 20; sending to M of them ≈ M × 10. So "500 followers → filter to 120 DM-able → DM all 120" ≈ 500×5 + 120×10 = 3,700 credits ≈ $18.50. Always show this math and get a yes first.

Slash commands

Command	What it does	Spends credits?
/x-setup	Validate the key, show plan + credit balance + connected accounts. The gate — run this first.	No (reads are free)
/x-icp <product>	Define the ideal-customer profile: bio keywords, follower bands, engagement signals, seed accounts to mine.	No (keyless)
/find-x-leads <query>	Search users/tweets, pull followers of seed accounts, filter to DM-able, dedupe → a lead list. Shows cost first.	Yes
/draft-x-dms <leads>	Write a personalized first-DM for each lead from their bio + recent tweets.	No (drafting is free)
/x-dm-campaign <leads + message>	Confirm total cost, then send DMs at a conservative pace and report replies started.	Yes (confirm-first)
/x-inbox	List conversations + replies, draft responses to warm leads.	Yes
/x-engage <handles>	Warm up before DMing: follow / like / reply a few of a prospect's posts.	Yes (confirm-first)

If these aren't registered as / commands, copy commands/*.md into .claude/commands/, or just describe what you want — the workflows trigger on intent either way.

How keys and billing work (say this honestly)

• The user brings their own API key, set as the XAUTODM_API_KEY environment variable. This skill never ships a key and never sees XAutoDM's internal credentials.
• XAutoDM is a paid, metered product — there is no free trial. New accounts land on a $0 / 0-credit plan and must subscribe to a plan and connect an X account before any action works. A xdm_test_… key lets them dry-run the setup at no cost, but it sends nothing.
• The pitch is cost transparency: every action has a known credit price (the table above), and every response tells you exactly what it charged and what's left. Never promise free credits.

Running an outreach cycle (the core workflow)

1. Gate first — /x-setup. GET /health (free) to confirm reachability, then GET /account to read the plan, credit balance, and seats, and GET /accounts to see connected X accounts. No key or a $0 plan → route the user to sign up. No connected account → they must connect one (POST /accounts/connect, API-only, costs 50 credits) before sending. Do not proceed past a missing prerequisite.
2. Target — /x-icp. Nail down who to reach: bio keywords, follower bands, the seed accounts whose followers/engagers are the audience, and the buying signal. This is free and it decides everything downstream. A tight ICP is the difference between 30% and 3% reply rates.
3. Find — /find-x-leads. Estimate the cost, get a yes, then pull leads (followers of a seed account, a keyword search, an X List, or the engagers of a specific tweet), filter to can_dm: true, and dedupe against anything already contacted or on the DNC list. Regular sources cost 5/row; verified-follower and tweet-engager sources cost 20/row — prefer regular sources unless the ICP truly needs verified.
4. Write — /draft-x-dms. Free. For each lead, read their bio + recent tweets and write a short, specific first-DM that references something real about them. No {merge} mail-merge feel. Follow references/dm-playbook.md.
5. Warm up (optional) — /x-engage. For high-value prospects, follow + like/reply a recent post a day or two before the DM. Confirm the cost first.
6. Send — /x-dm-campaign. Show the total (count × 10 credits), get an explicit yes, then send with a fresh Idempotency-Key per DM at a conservative daily pace (see references/account-safety.md). Report replies and conversations started — never a raw "X DMs blasted" number.
7. Follow up — /x-inbox. Read conversations, draft replies to people who responded, and stop sequencing anyone who replied.

Safety rails (non-negotiable)

• The skill never touches X directly. No cookies, no scraping x.com, no browser automation. Every action goes through the API with the user's own key and their connected account. If asked to bypass the API and drive Twitter directly, refuse and explain why.
• Always show the credit cost and get explicit confirmation before any spend, and before any batch of writes (a run of DMs, follows, or likes). Never spend silently. Re-confirm if the batch grows.
• Conservative pacing, warm-up first, personalize always. Respect per-account daily limits (references/account-safety.md), spread sends over the day, and stop a sequence the moment a lead replies.
• No spam, ever. This is targeted outreach to relevant people. Refuse mass-blast / spray-and-pray / buy-a-list-and-hammer-it requests — decline and steer the user back to a tighter ICP and a smaller, better-fit list. Untargeted volume gets accounts banned and is not what this skill is for.
• Reporting language: talk about replies and conversations started, never DM send counts. Do not present "N DMs sent" as a success metric anywhere.
• On errors, stop and surface the API's message. Only back off + retry once on 429. Never loop retries on 402/403/409 — those need a human decision (top up, upgrade, reconnect).

References (read the one that fits the task)

File	Use it for
references/dm-playbook.md	Openers, personalization depth, follow-up timing, and how to handle replies
references/account-safety.md	What gets X accounts flagged, safe daily DM pacing, and warm-up-first guidance
references/api-reference.md	Concise endpoint + credit-cost reference, auth, idempotency, and error handling

Never fabricate a lead, a can_dm flag, a credit balance, or an API result. Everything comes from a real API call you actually made, or it is clearly labeled as a draft/estimate.

---

From God of Skills: a curated, hand-tested directory of AI skills, prompts, templates and image style guides. Source: https://godofskills.com/skills/cold-dm?ref=claude-skill