SDR daily

One day of a Sales Development loop. You do the mechanical work (read, classify, send what was already approved), surface the few human decisions, and hand back a short briefing. You run from the campaign folder the cold-email-pipeline skill produced: out/campaign.csv, out/sent-log.json (one object per send: run_id, message_id, to, subject, stage, sent_at), out/replied.json (one object per suppressed address: email, reason, date, note). The briefing goes to out/briefing-<date>.md.

Before you start

• Load the core jell skill (https://jell.run/SKILL.md) if it is not loaded. Confirm access with the free balance tool or jell balance.
• Establish, once per campaign and reuse after: CAMPAIGN_CSV (the approved rows with a stage column: J+0, J+4, J+10), INBOX_ID (the From address in directives/email-copy.md, confirmed against the free email.inboxes), DAILY_CAP (default 30, all stages together), LAST_RUN (the newest sent_at in the sent log; on the first run there is none, so read the inbox without after).
• Check that Gate 2 of the pipeline happened: the sent log carries the test send to the user's own address. If it does not, the drip does not start; send the test, ask the user to confirm it landed, and stop there for today.
• If out/replied.json does not exist, create it as [].

The one rule

Nothing new goes out without approval. The drip and the follow-ups are not new: their copy and targeting were approved at Gate 1 of the pipeline, and the domain passed the test send at Gate 2. So, without asking, you may: drip first-touch from the approved queue inside the cap, send due follow-ups, read and classify replies, draft responses, update the suppression file, run diagnostics. You may not: send a reply to a prospect (draft it, the user sends), start a queue whose copy was not approved, exceed the cap, or email anyone in replied.json.

Run order

1. Triage replies

jell run -c email.messages -i '{"inbox_id":"<id>","labels":["received"],"after":"<LAST_RUN>","limit":100}' --max-cost 0.01 --wait 30 -o out/inbox.json

The inbox holds the campaign's own sent mail too (label sent); labels: ["received"] keeps the read to what came in. Each row carries message_id, thread_id, from (a display string, Name <address>: parse the address), to, subject, preview (the first 200 characters, cut mid-word), labels, received_at, direction and bounced. A read costs $0.002. Match a reply to a campaign email on thread_id (the send and its replies share one) or on the sender's address against the sent log. When the preview is not enough to classify, fetch that one message in full with email.messages and message_id (returns text), and only mark the thread ambiguous when the full text still does not decide it. Classify:

Bounces are not notices: a hard bounce shows as the label bounced (and bounced: true) on the campaign's own sent row, so read the sent rows since LAST_RUN once with labels: ["sent"] and put every bounced to in replied.json with reason: bounce. Out-of-office replies arrive from [email protected] with the original subject after "Re:" and the auto-reply text in the preview: they are auto-replies, never bounces, do not suppress them.

2. Hygiene

Count bounces since LAST_RUN against sends since LAST_RUN (the whole campaign on the first run). Any spam complaint, or bounces above 5% of those sends, is the kill switch: recommend pausing the drip until the list and copy are reviewed, and do not drip today.

3. Follow-ups due

From out/sent-log.json, a J+0 sent 4 or more days ago with no reply and no J+4 is due for J+4; a J+4 sent 6 or more days ago with no reply is due for J+10. Skip anyone in replied.json. Send the due rows, oldest first, threaded on the original:

jell run -c email.send -i '{"inbox_id":"<id>","to":["[email protected]"],"subject":"Re: <original subject>","text":"<approved J+4 body>","in_reply_to":"<message_id of the J+0 from the sent log>","reply_to":"[email protected]"}' --max-cost 0.01 --wait 30

in_reply_to takes the message_id the J+0 send returned (that is why the sent log keeps it); unsubscribe_url is optional, the opt-out sentence in the body is the floor. Count them as F. Follow-ups are time-sensitive and take the first claim on the cap.

4. First-touch drip

Budget: DRIP = DAILY_CAP - F. If positive, send the next DRIP J+0 rows that are not in the sent log and not in replied.json. Before the send, the free jell history --file today.txt on today's addresses confirms nobody was contacted from another campaign in this workspace.

Spread the sends across the day rather than in one burst: on a scheduled run, send the batch due for this hour; on a manual run, space the sends with a pause between them and say what was sent when. On a fresh domain, ramp the cap: 5 to 10 a day for days one to three, 15 through day seven, then 25 to 30; the domain's age is not in any run output, so count from the first sent row in the inbox or ask. Log every send with its run ID, message_id, to, subject, stage and timestamp.

When the queue is empty, say so and suggest running cold-email-pipeline for the next batch.

5. Briefing

SDR briefing, <date>, <campaign>

Interested (<n>): drafts below, your approval to send
  - <company> (<name>): "<gist>"
Not interested (<n>) suppressed. Redirects (<n>). Unsubscribes (<n>).

Sent today: <F + DRIP> of <DAILY_CAP>
  - Follow-ups <F>, first-touch <DRIP>, queue remaining <n> (about <days> days)

Hygiene: bounces <n>, complaints <n>, domain OK or AT RISK

Your decisions today
  1. Approve or edit <n> reply drafts (below)
  2. <replenish the queue, pause, anything else human>

Charged today: $<total> across <n> runs (the sum of each run's billing.charged)

Then the reply drafts, one per interested thread.

Rules

• One daily cap across drip and follow-ups. Never exceed it.
• Suppression is sacred: anyone who replied (other than an auto-reply) is never contacted again by this campaign.
• Reply responses are drafted, never auto-sent, even on a scheduled run.
• Honesty on signal: zero replies in week one is normal; a booking that predates the campaign is not a result.
• max_cost on every run, --wait 30 on every send and read. Every send logged with its run ID and message_id.

Scheduling

Once the user trusts the drafts (a week or two of manual runs), this skill can run each weekday morning on a schedule. The scheduled run drips, follows up, triages and briefs on its own, and still only drafts responses.