wannalead
Use Wannalead safely across LinkedIn, Instagram, and X. Start with an identity check and saved commercial context, then continue the user's task.
How do I install this agent skill?
npx skills add https://wannalead.co --skill wannaleadIs this agent skill safe to install?
No partner audit is available yet. Read the source before installing.
What does this agent skill do?
Wannalead
Use the CLI only as a thin HTTPS API client for https://wannalead.co. Prefer
the installed task skills over guessing workflows from flag help alone.
First request: identity, then saved context
On the first Wannalead use for a user or workspace, invoke the
wannalead-onboarding skill. The first command must be:
wannalead me
Use the returned LinkedIn public identifier to load any saved context from
~/.wannalead/workspaces/<linkedin-public-identifier>/ (ONBOARDING.md and the
durable files it points to). Then continue with the user's request, whether or
not onboarding is complete. Incomplete onboarding is not a prerequisite for
endpoint discovery, reads, or explicitly approved writes: take task context
from the request and ask only for details this task needs. Offer the
commercial interview when the user has not named a task or wants a reusable
profile; do not start it when the requested task can proceed.
If wannalead me reports missing authentication, run wannalead auth login
and retry it. If it reports a missing LinkedIn connection and the task needs
LinkedIn, use wannalead auth connect, then retry the same identity check. An
Instagram or X task does not require a LinkedIn connection. Do not run a
business endpoint while identity remains unresolved. Resume saved interview
progress instead of restarting it.
Workspace not subscribed yet
wannalead me works without a subscription and returns data.billing. When
billing.active is false (or any call answers 402 with
subscription_required), the workspace has not subscribed to the LinkedIn plan yet.
This state applies to LinkedIn endpoints only. Check data.network_access
separately before drawing a conclusion about Instagram or X:
- The onboarding interview still works: it only needs
wannalead meand the local workspace files. Do not call LinkedIn data endpoints during it, and do not stop the interview because of billing. - When the user asks for a first LinkedIn result, or right after the onboarding
synthesis, say plainly that Wannalead needs the LinkedIn plan subscription and give
billing.billing_url(https://wannalead.co/dashboard/billing?checkout=pro). That link opens Stripe checkout directly. Ask the user to tell you once it is done. Do not describe it as a trial that ended, and do not quote dates. - When a LinkedIn CLI or local MCP call answers
402 subscription_required, the CLI has already opened that checkout page in the user's browser (at most once every 10 minutes). Tell the user the payment page just opened, give the same link in case it did not, and wait for them to say it is done. Do not retry the call in a loop. - When the user says it is done, run
wannalead meagain. Ifbilling.activeistrue, resume exactly where you stopped (the pending request, or the use-case menu). If it is stillfalse, wait a few seconds and check once more before asking the user to verify the payment page. - For
billing_action_required, ask the user to fix the payment at the same URL (it also opens on its own after a402). Forworkspace_unavailable, stop and say the workspace cannot be used.
An Instagram or X add-on has its own entitlement. When its add-on is active but
network_access.<network>.read_permission is false, ask the user to reconnect
Wannalead and grant that network's read scope. A 403 insufficient_scope is a
permission problem, not a subscription problem. If an add-on is inactive,
describe that network's billing requirement without claiming the other
networks are blocked. Continue on any network with an active entitlement and
the needed permission.
Choose a use case
The six use cases below cover LinkedIn. Offer this menu, in the user's language, right after onboarding completes and at the start of any session where the user has not named a task. Ask which one to start with instead of guessing; then load the named skill (or, for a use case without a dedicated skill, run endpoint discovery for the listed chain). Every item runs from the user's own LinkedIn account through Wannalead endpoints only.
- Lookalikes with signals. Describe the ICP; find every lookalike
profile with the signals that matter (tools in use, posts, job changes).
Skill:
icp-lookalikes, plusjob-change-signalsfor the trigger side. - Campaigns. From the durable context (offer, customers, writing style)
draft the messages, then run sends and follow-ups. Skills:
campaign-automation,warm-outreach,humanizer. - Dormant conversations. Revisit LinkedIn conversations left without a
reply and follow up on leads that went quiet. No dedicated skill: chain
messages.conversations→messages.history→warm-outreachfor each approved follow-up. - Dormant network. Surface first-degree connections the user never
contacted and propose an approach. No dedicated skill: chain
me.connections→messages.conversations(to exclude existing threads) →profiles.enrich→warm-outreach. - Saved post → ICP. A post whose reactions or comments look like the
target audience: the user saves it on LinkedIn, the agent pulls the
engagers and keeps those matching the ICP. Skill:
engagement-prospectingfrom the post URL; readme.saved-postsfirst to find the saved posts. - Messy list → contacts. A rough CSV (for example family offices):
reconcile each row with the company that exists on LinkedIn, then list the
people to contact in each. No dedicated skill: chain
companies.search/companies.resolve→companies.employees, thenaccount-mappingfor the buying committee.
Present the menu as a short numbered list, not as documentation. When the user picks one, restate the concrete outcome in one sentence and start.
The list is not capped at six. Load the wannalead-use-cases skill for the
full routing, including scraping a competitor's own audience, registering for a
LinkedIn event and mining its attendees, reconciling any roster (CSV / Luma /
Slack) to LinkedIn, mining best customers' enrichment and recommendations for
ICP patterns, and batch-replying to the inbox through a shared review artifact
that then feeds OUTREACH-STYLE.md.
Mandatory live endpoint discovery
After the identity and context check, run wannalead endpoints
and scan the complete current catalog before any business API call. Search the
relevant domain with wannalead routes <filter>. For every candidate endpoint,
run both:
wannalead endpoints <name> --examples --json-schema
wannalead openapi path <exact-/v1-route>
Read the complete returned contract before building a request: method, path, parameters, body schema, examples, responses, permissions, write guards, idempotency, pagination, deprecation, and replacement metadata. The live OpenAPI contract overrides examples or remembered endpoint shapes in any skill. Never claim that a capability is unavailable or impossible until the full catalog has been searched using the user's terms and relevant synonyms, and every plausible endpoint and replacement has been inspected. If nothing matches, state which catalog searches and candidate endpoints were checked.
Setup
Start with wannalead me. Authenticate only if that command reports missing
authentication, then retry it. Run endpoint discovery after the identity and
saved-context check; incomplete onboarding does not block it.
wannalead me
# Only when the previous command reports missing authentication:
wannalead auth login
wannalead me
# After the identity and saved-context check:
wannalead endpoints
If the API returns linkedin_connection_required, run:
wannalead auth connect
Wait until wannalead me reports the workspace connection as ready, then retry
the exact read once. A linkedin_connection_required response creates no job
and consumes no product unit. For a guarded write, preserve the original body
and idempotency key instead of constructing a replacement request.
For a new account, or when durable commercial context is missing, the
wannalead-onboarding skill can build the user's persistent local ME.md,
COMPANY.md, ICPS.md, and OUTREACH-STYLE.md files through a resumable
interview tracked in ONBOARDING.md. Offer it; do not make the requested task
wait for it.
For later prospecting, audience, account-mapping, campaign, or outreach tasks,
use the progressive context rules in wannalead-onboarding: load only the
relevant durable files, read current operational state from its authoritative
system, and persist a durable change only after the user confirms it. Do not
store CRM records, prospects, connections, conversations, campaigns, or run
logs in the local Markdown workspace.
MCP clients can launch command wannalead with args ["mcp"]. The MCP and
CLI expose the same public API capabilities.
Core workflow
wannalead endpoints companies.search
wannalead routes events
wannalead openapi path /v1/linkedin/events/register
wannalead call companies.search -body '{"keywords":"legaltech","company_size":["51-200"]}'
Prefer dedicated commands when available:
wannalead profiles enrich -profiles alice,bob --poll
wannalead profile-resolve -public-identifier alice
Enrich lists in bulk
When there is more than one profile to enrich, send the whole list in one
profiles.enrich call (up to 99 profiles per call; split longer lists into
calls of 99). Never loop profiles.resolve or profiles.enrich one profile at a
time: a 270-profile list enriched one by one takes hours, the same list in bulk
takes minutes. Pass profileAcoaa values when you have them, so the whole list
is read in a single bulk request; URLs and slugs work in the same call but are
read one by one.
wannalead profiles enrich -profiles ACoAA...,ACoAA...,ACoAA... --poll
wannalead call profiles.enrich -body '{"acoaa":["ACoAA...","ACoAA..."]}'
Use wannalead call <endpoint-name-or-path> for catalogued product
capabilities. Use wannalead create -endpoint-key ... only as a last-resort
escape hatch when no product alias exists.
For stored LinkedIn conversation history, use the named mirror reads:
wannalead call mirror.conversations -param limit=100
wannalead call mirror.messages \
-param conversation_id=<conversation-id> \
-param limit=100
The API key selects the associated LinkedIn account. Do not ask the user for an
account identifier or expose local configuration, raw HTTP commands, refresh
cycles, polling, or synchronization internals. Return the stored business result
directly. Inspect mirror.status only when missing or incomplete data would
materially change the answer, and describe that limitation without narrating
the backend workflow.
For a multi-step LinkedIn campaign with profile visits, delayed invitations,
acceptance-aware messages, per-recipient copy approval, and lifecycle tracking,
use the installed campaign-automation skill.
For agent chaining, prefer the stable identifiers already projected by the API:
messages.conversationsreturns exact conversation URNs plus participants withprofileAcoaa/profileUrn/publicIdentifierwhen exposed.messages.threadandmessages.seen-receiptsretain sender/actor ACoAA.events.attendeesretainsprofileAcoaafor directprofiles.enrichcalls.me.notificationsreturns recent inbound cards with post/profile/company identifiers for follow-up reads.
Reuse profile identifiers exactly as returned. Prefer profileAcoaa or a
profile URN for enrichment; otherwise preserve the exact Unicode or
percent-encoded /in/{publicIdentifier} value. Never construct a slug from a
display name or strip its accents: LinkedIn can return 403 for the ASCII form
while the canonical Unicode profile remains accessible.
Response handling
- Job creation can return completed data or
job_id/job_ids. - Use
--pollfor local polling (five-minute default deadline) orwannalead job -id <job_id>. - Use
-wait -timeout-ms <n>for bounded API waits, at most 25000 ms. - Keep
-client-timeout-msabove the bounded API wait. - A poll deadline with a queued/running job is a pending result, not a failed
LinkedIn call. Keep the same
job_id; do not create a duplicate job. - Re-check a pending job at most once per advised delay (
Retry-After,poll_after_ms, or the MCPretry_after_seconds). Never re-check a job that is already terminal, and never script or schedule a loop over job ids: a finished job never changes. - A confirmed
invitations.sendcan return202with statusinvitation_profile_resolve_pendingorrelationship_preflight_pendingbefore any invitation job exists. Pollpreflight_job_id, then retry the exact request with the same idempotency key after it succeeds. The dedicated CLI andwannalead call invitations.senddo this automatically when--pollis set. The key is reserved for that canonical target and message while pending. Stale relationship evidence is refreshed under that same key. Internal prerequisite reads do not consume product or LinkedIn account quota; quota starts only if the invitation job is created. - Invitations are capped per LinkedIn account at 50 a UTC day and 200 a week,
shared between direct calls and campaigns. A newly connected account ramps
up first: 20 on the UTC day it was connected to Wannalead, then 30, 35, 40,
and 50 from its fifth day. The ramp follows the calendar, so idle days still
count. The workspace
dailyInvitationLimitis the 50 ceiling, not the account's budget for today. An invitation over the budget is deferred to the next UTC day (account_budget), not failed: tell the user, do not retry. - An invitation job with terminal status
failednever changes. Reusing itsIdempotency-Keyreturns409 invitation_attempt_failedwith the original job, not a new send. Inspectdata.errorfor the cause. For a confirmedLINKEDIN_RATE_LIMITEDresponse, stop; LinkedIn does not give a reliable reset time. After any Wannalead account cooldown, a separately approved new attempt needs a new key and a fresh relationship check. Never treat HTTP 200 alone as proof an invitation was sent; requiredata.status=succeeded. LINKEDIN_TARGET_UNAVAILABLEis scoped to that read target. If the input was manually constructed or ASCII-normalized, retry once with the exact sourceprofileAcoaa, profile URN, or profile URL. Otherwise skip or report it without treating the connected account as unavailable. Never brute-force possible accents.- Browser-backed read success is trusted for 24 hours while the same account binding and healthy extension session remain active.
- Stop on rate-limit, restriction, authentication, or connection errors; do not retry-loop.
- A CLI command using
--pollexits non-zero when the terminal job isfailed,queue_failed,parse_failed,timed_out, orcanceled, after printing the final job response. Inspect both the process exit code and the returnedstatus. - Stdout is agent-safe, not identifier-free: URNs and ACoAA values may remain when they are useful for chaining calls. Provider-domain URLs, credentials, browser routing, and unprocessed response bodies must not appear.
Daily agent journals
Create one stable run, then append stable events. Replays return the same
run/event; reusing a key with a different body returns 409.
wannalead agent-runs start \
-client-run-id "daily-outreach:2026-09-01" \
-channel daily-outreach \
-objective "Process the daily prospect queue" \
-idempotency-key "agent-run:daily-outreach:2026-09-01"
wannalead agent-runs append \
-run-id <run-id> \
-client-event-id "target-12:send" \
-event-type invitation.succeeded \
-status succeeded \
-job-id <job-id> \
-payload '{"processed":1,"succeeded":1}' \
-idempotency-key "agent-event:<run-id>:target-12:send"
Journal writes do not consume LinkedIn product units. The linked Wannalead jobs remain the source of truth for metering, quotas, execution, and polling.
Invitation campaigns
Campaigns remain extension-backed. Preparing a campaign does not authorize its LinkedIn writes; require explicit approval of the exact target set before launching, approving, or scheduling it.
For an invitation-only campaign, use invitations.send directly as the
sequence root. The server reuses cached identity, resolves a missing identity
as an internal prerequisite, and reruns profile.actions immediately before
dispatch. First-degree, pending, or unavailable relationships terminate as
non-billable skips.
Add visible profile.full_unary or profile.actions steps only when the user
explicitly wants a profile visit, a delay anchored to that visit, or separate
routing for existing first-degree connections. They are not invitation safety
prerequisites.
To message only existing first-degree connections without creating invitation
jobs, publish relationship.check as the sequence root and route
relationship.first_degree to messages.send_profile. Leave
relationship.not_connected without a branch so those targets complete
without an invitation or message. Inspect the full request with
wannalead endpoints automation.campaigns.sequences.create --examples --json-schema or use the campaign-automation skill.
The activation order is validate, launch, review, then schedule.
launch selects and binds the workspace's healthy LinkedIn account; clients
must not require a campaign binding before calling it. Reviews require the
campaign to be running. More than 100 targets require the review confirmation
token returned by the first call before scheduling.
Once scheduled, keep campaign and job ids durable. Poll existing state without
recreating writes. The server owns extension-offline deferral, account-binding
changes, pacing, quotas, cooldowns, retries, and target-level idempotency. After
an invitation obtains its dispatch slot, the automation worker reruns
profile.actions through the extension immediately before the write. Only a
fresh relationship.not_connected result proceeds; first-degree, pending, and
unavailable results terminate as non-billable skips. When the target has
already sent the account an invitation (relationship.invitation_received),
the worker accepts that invitation instead of inviting and the step finishes as
relationship.first_degree, so the message branch continues. The same
fallback runs inside every relationship.check step and before a
messages.send_profile send.
Product feedback
When Wannalead itself returns an incorrect or incomplete result, has unclear
endpoint guidance, or blocks a workflow reproducibly, preserve the endpoint
name plus the response request_id and job_id when available. Explain the
observed behavior, the expected behavior, and the impact without copying
credentials, cookies, raw provider payloads, or unrelated private content.
Submitting feedback is an external write. Show the exact report to the user and obtain approval before sending it. Then use the catalogued meta endpoint:
wannalead endpoints feedback.send --examples --json-schema
wannalead call feedback.send \
-body '{"message":"<observed behavior and expected behavior>","context":"endpoint=<name> request_id=<id> job_id=<id>"}'
feedback.send does not create a LinkedIn job or consume LinkedIn product
quota. Do not send automatically, include secrets, or retry a rejected report
in a loop. Reporting the issue does not replace returning the usable partial
result or explaining the current blocker to the user.
Writes
Request write scope and validate with dry-run before sending:
wannalead auth login --write
wannalead invitations send \
--dry-run \
--profile-url "<profile-url>" \
--message "Hello"
wannalead invitations send \
--idempotency-key invite-001 \
--profile-url "<profile-url>" \
--message "Hello" \
--poll \
--yes
Post engagement uses the same guarded contract. Wannalead resolves canonical thread/root identifiers server-side; do not derive them or call the provider directly:
wannalead posts comment --activity-id <id> --text '<final-text>' --dry-run
wannalead posts reply --parent-comment-urn '<urn>' --text '<final-text>' --dry-run
wannalead posts repost --activity-id <id> --dry-run
After explicit approval, repeat the chosen command with a stable
--idempotency-key and --yes. The write result may retain comment/repost URNs
for chaining, but never echoes the submitted text.
Use the warm-outreach workflow for every invitation, DM, follow-up, comment,
or reply. It requires the installed humanizer skill before approval, dry-run,
and send. Send only the exact humanized text the user approved.
Every write requires explicit user intent. Preserve idempotency and never turn a guarded one-recipient workflow into bulk automation.
Boundaries
Never read LinkedIn cookies, browser profiles, local LinkedIn state, or gateway WebSocket internals. Never assemble private LinkedIn identifiers or call LinkedIn endpoints directly. Pass user-facing product inputs to Wannalead and let the API resolve execution details.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/wannalead.co/wannalead">View wannalead on skillZs</a>