Terminal
Terminal/Skill
Watches a rep's target accounts in Clay and posts each hire, funding event or job posting to a Slack channel, naming the account owner.
By Clay
Works with
npx skills add clay-run/clay-skill-creator --skill alert-reps-on-account-signalsThe insight, in the author's words: a rep should not have to go looking — when something happens at an account they own, the event should find them where they already are, with the account and the owner named in the first line. So the list is loaded once, the watches run on their own, and every event becomes one Slack line shaped like "@owner — a key new hire has been filled at Acme: Jane Doe, VP Sales." Nothing here scores, enriches or writes outreach; the rep decides what to do.
Nothing here ships with a value. Each one is the installer's, not the author's: ask for it, never substitute a plausible default, and where an answer does not exist say which step becomes unavailable rather than guessing. Where a default IS defensible it is named below, and using it means saying so in the output.
| Input | What the installer supplies | If it is missing |
|---|---|---|
| The account list | a CSV with one row per target account and a column holding the company domain or company LinkedIn URL | no default — nothing to watch |
| List name | a short label for this list, e.g. the rep's name or territory; it tags the accounts in Audiences and names the segment, signals and workflows | no default — ask; two lists with the same label would merge |
| Slack channel | the rep's own public channel for these alerts, by name — one channel per rep, by the author's design; always ask for it, never assume one. Slack must be connected in Clay (Settings → Connections, in this workspace) and the Clay app added to the channel; only then does the channel appear in the action's list | no default — the alert step cannot be built; the signals still run and their events stay readable in Clay Audiences |
| Mention | the rep's own Slack member ID (the U… value from their profile) to put at the front of every alert, if they want the ping | none — alerts carry no mention |
| Leadership seniority | which new hires count as leadership, from Clay's seniority levels | the author's default is c-suite, vp, director, head — confirm it and say it was used |
| Job posting titles | comma-separated title words a posting must contain to alert, e.g. the roles the rep's product serves | none — every posting at every watched account alerts, which at a large account is many per week; say so before accepting it |
| News topics | which of Clay's news topics alert, from its fixed vocabulary | the author's default is Fundraising, Investment, Initial Public Offering, Merger & Acquisition, Executive Appointment, New Product Launch, Business Expansion — confirm it and say it was used |
| Check cadence | how often the watches run: daily, weekly, biweekly or monthly | weekly is defensible — new hires and postings are billed per account checked on every run, so daily costs about seven times weekly; say which was used |
If an answer sheet is present beside this skill, load it and ask only for what it does not cover. A partial sheet is normal; a value it is missing gets asked for on its own rather than restarting the interview. Say which values came from the sheet before using them — a sheet applied silently is a wrong field nobody catches. If there is no sheet, say nothing about sheets — the check is a file lookup, not a question, so run the interview as though the feature did not exist rather than reporting an absence. At delivery, offer to save the answers back (identifiers only — never a token or a password), private and never published — and phrase the offer so it explains itself: "want me to save your answers to a file, so the next person on your team doesn't have to answer these again?"
clay audiences signals get, but no alert is sent.Say this first, as three sentences: this loads your account list into Clay Audiences with a tag, creates three signals that watch those accounts on the cadence you choose, and builds one workflow that posts a Slack message per event to your channel; it writes to Audiences and Slack and nowhere else, and it never deletes anything.
Run clay whoami; echo "exit_code=$?". If it fails, name what is wrong — the CLI missing, below the
version the Clay plugin requires, or signed out — give the one fix (clay login, or install the Clay
plugin and run its setup skill), and stop. Do not install, upgrade or fetch anything to repair
it. Tell the user which workspace you are in.
Then two free checks, before any interview:
clay workflows actions test <slackPackageId> slack-get-channel-id-from-name --inputs '{"channelName":"<channel>"}'
(the package is the one whose packageDisplayName is Slack in clay workflows actions list). A
channel_id back means Slack is connected for this user and the channel is public and reachable;
keep the id — the send action takes the channel id, not the name. Do not judge the connection
by actions dynamic-fields: on the author's test it reported MISSING_ACCOUNT while the same
actions ran fine, because the probe binds no default account. A MISSING_ACCOUNT on the test
call is the real signal: say so, give the fix (connect Slack in Clay's Settings → Connections, add
the Clay app to the channel), and offer to continue with the watches only.clay signals list. If it returns auth_forbidden, signals are not available on
this workspace and the skill cannot do its job; say so and stop.This skill builds through the Clay plugin's workflows, signals and audiences skills — read
them before Step 2 and follow them for every CLI shape (CSV triggers, Audiences upserts, signal
creation, trigger binding). If they are not installed, say the Clay plugin is required and stop.
www., query strings, trailing
slashes) and dedupe. A row with neither identifier is listed as skipped, never guessed from a name.clay signals create --help lists it); a word outside it validates and then matches
nothing.Look before building. clay audiences fields list --entity-type companies — find a field suitable
for the list tag. Prefer a boolean field named for the list (Target list: <list name>), created
with clay audiences fields create --data-type boolean: a workspace can hit Clay's 200-field limit
on text, email and URL account fields (the author's did, 2026-09-28) and boolean fields are still
allowed. A shared text tag field is fine where one already exists.
clay audiences list --entity-type companies — an existing segment named for this list means the
list was loaded before; say so and ask whether to add to it. clay workflows list — look for
Load target accounts — and reuse it if the graph matches.
Before creating anything, one message: the two workflows and their names, the Audiences field(s) to create, the segment, that N account records will be created or updated with the tag, and that nothing is deleted. Wait for an explicit yes.
Then build the ingest workflow: a csv_upload trigger → one tool node, action
upsert-audiences-record (package Audiences): entityType ← ACCOUNT; lookup on domain when
the column is a domain or linkedin_url when it is a LinkedIn URL; record fields ← the list's boolean field set to true;
removeNullValues ← true. Link the CSV with triggers csv upload, run csv run --limit 3
first, read the three runs, then run the rest. Read the runs, not the batch: a run can complete
around an ERROR_BAD_REQUEST when a selected field has no binding.
Create the segment with clay audiences create --entity-type companies --name "<list name>" and a
filter selecting the list's boolean field equal to true (key is the field id and dataPath is
["account_entity_field_values","field","<field id>"]). Confirm the count with
clay audiences records search-count --audience-id <id> and compare it to the CSV's row count.
A count above the row count means the workspace holds duplicate account records for one domain —
the author's test tagged 20 records for a single nvidia.com row, because the upsert's domain lookup
updates every match. Signals bill per record checked and would alert once per duplicate, so narrow the
segment (for example the tag AND a headcount floor, or the record's own LinkedIn URL) until the count
matches, and say what you added.
Look before building. clay signals list — a signal of the same type already watching this
segment answers the job, and a second one doubles the spend; reuse it. Otherwise create each with
clay signals create without --activate, watching the segment
("entityType": "ACCOUNT", "segmentIds": ["<segment id>"]), named <list name> — <type>:
| Signal | --type | Per-type settings | Billed |
|---|---|---|---|
| Leadership hires | NewHire | filters.start_from_method ← query; filters.job_title_seniority_levels_v2 ← the confirmed seniority list; filters.job_title_exclude_keywords ← Office of, Assistant; lookBackTimeWindowInMonths ← 1 (the author's test used 3, to see events on the first run) | per account checked, per run |
| Job postings | JobPost | filters.startFrom ← ClayTableOfCompanies; filters.job_title_keywords ← the confirmed titles (omit the key only if the installer accepted "every posting"); filters.max_num_days_since_posted ← the cadence in days | per account checked, per run |
| News and fundraising | News | filters.topics ← the confirmed topics; filters.autoAdvanceEarliestPublishDate ← true; filters.earliestPublishDate ← today; filters.maxNewsPerDomain ← 5 | per event, so the topic list is the cost lever |
--schedule ← the confirmed cadence on all three. Read each back with clay signals get and
confirm type, segment, filters and schedule. Then one message: the three signals as created, the
segment and account count, that starting them begins spending on every run — per account for hires
and postings, per event for news — and the ask. Give the per-record and per-event prices only as the
platform reports them; the author did not measure them. Wait for an explicit yes, then
clay signals resume each one. Signals detect changes after monitoring starts: say plainly that
the first alerts arrive after the first scheduled run, and that hires and postings that predate the
signal do not fire.
clay workflows list — look for Account signal alerts — ; reuse if the graph
matches. Otherwise create it with three audience_signal triggers, one per signal, each bound to
that signal's signal.id (the sig_… value from clay signals list, not the entry id) with
entityType ACCOUNT, all wired to one tool node:
send-message-to-channel-botname (package Slack), on the installer's
connected Slack account: targetSlackChannel ← the channel id from Step 0 (static), botName ← Account signals, summary ← the message below, built by
reference from the trigger payload. Slack bills nothing to Clay; it runs on the installer's own
connected account.The message, one line per event type, in Slack markdown, with the mention first when the rep asked
for one (<@U…>), else nothing:
<@owner> A key new hire has been filled at *<account>*: <person name>, <title>. <profile link><@owner> *<account>* is hiring: <job title> (<location>). <posting link><@owner> News at *<account>* — <topic>: <headline>. <article link>The trigger's payload has two halves, both confirmed on the author's test (2026-09-28): the
account's Audiences fields under $.fields by display name ($.fields["Company name"],
$.fields.Domain, $.fields["LinkedIn URL"]), and the event under $.signal (signalType,
emittedAt, activityTime, and data, whose shape is per type). For a NewHire event the
author read a real payload: $.signal.data.fullProfile.name, .latest_experience_title,
.latest_experience_start_date, .latest_experience_company, .location_name and .url (the
person's profile). JobPost and News payloads were not observed — before wiring those two
messages, read one event of each with clay audiences signals get --segment-id <id> --since <date> --signal-types JobPost,News and take the paths from its data; until one exists, wire the account
name and signalType only and finish the fields after the first run posts. Record the paths you used
in the workflow's description.
Test one message with clay workflows nodes test <wf> <slackNode> --inputs carrying the
node's own pinned inputs by name (company, person, title, started, profile), not the
trigger payload — a node test resolves no upstream pins, and the author's first attempt posted a
message with every field blank. Confirm in the channel that it arrived and reads as above. The author did this on 2026-09-28: a direct send and
a node test both posted to the test channel, and the send action reported wasSent: true with a
message_ts; Slack billed nothing to Clay and used one action execution. Show the installer the
message and wait for a yes before publishing. Then clay workflows publish so the triggers run on
the live graph, write the provenance line into both workflows' descriptions (see below), and stop —
from here the signals and the alert workflow run unattended.
Provenance line: read each workflow's description with clay workflows get; if it is null
write the skill's one-line description, a newline, then Sourced from marketplace skill: <slug>@<revision> using the marketplace_slug and marketplace_revision this file's frontmatter
carries when installed from the Marketplace; if those fields are absent, say "Marketplace
attribution is unavailable for this skill" and write no marker. If a marker is already there, write
nothing and say so; if a different marker is there, stop and report the conflict.
One summary: the list name, accounts loaded (created vs updated vs skipped), the segment and its
count, the three signals with their cadence and filters, the channel, the mention rule, the test
message as posted, and where to look — clay signals list for run status and errors,
clay audiences signals get for events, clay workflows runs list for alert runs. Then the
answer-sheet offer from Declared inputs.
| List | Accounts loaded | Segment count | Signals | Cadence | Channel | Mention |
|---|---|---|---|---|---|---|
| Northwind territory | 48 created · 2 updated · 1 skipped (no domain) | 50 (matches CSV) | Leadership hires · Job postings (titles: Data, Analytics) · News (7 topics) | weekly | #northwind-dana | @Dana |
@Dana Whitfield A key new hire has been filled at *Contoso*: Priya Raman, VP Data Platform. linkedin.com/in/…
clay signals list — runStatus and any error.userFriendlyMessageclay audiences signals get --segment-id <id> --since <date> --signal-types NewHire,JobPost,Newsclay workflows runs list <alert workflow id>runs list.removeNullValues: true on the upsert; NEVER blank or delete an Audiences value.Ask: "Set up Slack alerts on my 50 target accounts — new execs, jobs, funding." Step 0: signed in;
Slack connected; signals available. Step 1: CSV with a domain column; list
name Northwind territory; channel #northwind-dana, mention @Dana; seniority default confirmed; posting titles
Data, Analytics; news topics default confirmed; weekly. Step 2: boolean field Target list:
Northwind territory created; ingest workflow built; 3 rows tested, then 47; segment created, count
50, matching the CSV.
Step 3: three signals created paused, read back, spend model shown, yes, resumed. Step 4: alert
workflow with three signal triggers → Slack; a sample payload tests one message into the channel;
paths for each type wired after the first weekly run; published. Step 5: summary as above, and the
answer-sheet offer.