wannalead-onboarding
First-use identity check and progressive commercial context for Wannalead; identify the account, load saved context, and build it over time without blocking the user's task.
How do I install this agent skill?
npx skills add https://wannalead.co --skill wannalead-onboardingIs this agent skill safe to install?
No partner audit is available yet. Read the source before installing.
What does this agent skill do?
Wannalead onboarding
Build the small amount of durable commercial context that improves later Wannalead prospecting and outreach. Keep this context separate from operational state already owned by Wannalead, LinkedIn, or an authorized CRM.
This skill covers onboarding and later user-confirmed context updates. It does not authorize invitations, DMs, CRM writes, campaign actions, or any other external mutation.
First-use identity check
Every Wannalead operational skill invokes this skill on first use. The first Wannalead command must be:
wannalead me
This identity bootstrap is the only exception to the normal OpenAPI discovery
gate. Do not call wannalead endpoints, a data endpoint, or a write endpoint
first. If authentication is missing, run wannalead auth login and retry
wannalead me. If the LinkedIn connection is missing, run wannalead auth connect and retry the same identity check.
Use the returned public identifier to inspect the matching ONBOARDING.md
immediately and load any relevant durable context. Then return control to the
requested workflow, whether or not onboarding is complete. Incomplete
onboarding is not a prerequisite for endpoint discovery, reads, or explicitly
approved writes. When the requested task can proceed, do not start the
interview; ask only for details that task needs. Offer to start or resume the
interview when the user has not named a task or wants a reusable profile.
Authentication, the relevant network connection, billing, and endpoint guards
still apply.
Mandatory endpoint discovery after bootstrap
Before using any later Wannalead data endpoint, run:
wannalead endpoints
wannalead routes <filter>
wannalead endpoints <name> --examples --json-schema
wannalead openapi path <exact-/v1-route>
Scan the full catalog, search the domain and synonyms, then read every candidate's complete contract before building a payload. Never rely on remembered endpoint lists or say a capability is impossible until the live catalog, plausible candidates, deprecations, and replacements have been checked.
Operating modes
- First session without a named task: offer a resumable commercial interview, one focused question at a time.
- Progressive enrichment: load the existing context before a later Wannalead task and update it only with durable information confirmed by the user.
Do not rerun the interview when the existing files answer the current question.
Local workspace
Use a location explicitly chosen by the user when one exists. Otherwise use:
~/.wannalead/workspaces/<linkedin-public-identifier>/
Create or maintain only these files:
ONBOARDING.md
ME.md
COMPANY.md
ICPS.md
OUTREACH-STYLE.md
Their responsibilities are:
ONBOARDING.md: interview status, completed or deferred topics, file pointers, and the next question. Never duplicate commercial facts here.ME.md: the user's confirmed professional identity, role, commercial goals, responsibilities, and relevant personal constraints.COMPANY.md: the company's offer, outcomes, use cases, proof, geography, delivery constraints, positioning boundaries, and stable exclusion policies.ICPS.md: one or more named ICPs. Each ICP keeps company-fit criteria, buyer and prescriber personas, triggers, audience sources, exclusions, evidence, confidence, and status together so they cannot drift across separate files.OUTREACH-STYLE.md: user-approved language, tone, rhythm, length, openings, calls to action, phrases to keep or avoid, and channel-specific guidance.
Keep raw private messages, CRM data, API responses, and credentials out of these files and out of tracked repositories. Restrict local permissions where the platform supports it.
Existing workspace migration
Before starting or resuming the interview, inspect the existing workspace. If one or more of the four durable context files is missing but older top-level Markdown documents exist, run a one-time migration before deciding that context is absent:
- Read the existing decision documents and extract only durable, previously confirmed user, company, ICP, persona, audience, commercial-boundary, and outreach-style information.
- Consolidate those facts into
ME.md,COMPANY.md,ICPS.md, andOUTREACH-STYLE.mdunder the ownership rules below. Do not migrate contact rows, prospect queues, conversation details, campaign state, or run logs. - Record migration progress and unresolved conflicts in
ONBOARDING.md. - Show the user a short synthesis and ask them to confirm corrections before marking the migration complete.
Never delete or rewrite the older documents automatically. After the migration is confirmed, ignore them as context so later runs have one source of truth.
State ownership
The local Markdown files are durable decision context, not a second product database:
- Current connections and conversations remain in Wannalead's server-side account state and must be read through current public endpoints when needed.
- Customer, opportunity, lead, and do-not-contact records remain in the
authorized CRM.
COMPANY.mdmay name that source and keep stable global rules, but must not copy its entity state. - Prospect shortlists and campaign target statuses remain in the current task result or the server-side campaign target state.
- Durable execution journals use Wannalead
agent-runs; do not create a local run log.
Operational skills may read the four context files. They must not rewrite them merely because a search, profile, message, campaign, or CRM result changed. If a task suggests a durable change, present it as a proposed update and persist it only after the user confirms it.
Progressive update rules
Load the existing workspace before asking questions or calling Wannalead. Preserve user-confirmed facts and prior decisions. Never silently replace a confirmed fact with an inference.
For material context, record a source and observation date using one of:
user-confirmed, linkedin-observed, crm-confirmed, inferred, or
unknown. Organize evolving sections as:
- confirmed decisions;
- active hypotheses;
- dated evidence;
- open questions.
A user-confirmed correction can update the canonical summary immediately. A single search result, missing result, reply, company, or profile remains dated evidence. Promote it to an ICP, persona, audience, or style rule only after the user confirms it or repeated strong evidence supports it and the user accepts the proposed update.
First-session workflow
1. Resolve the user
Run wannalead me and verify that the returned LinkedIn identity is the
intended account. Stop on an identity mismatch. If Wannalead returns
linkedin_connection_required, send the user through wannalead auth connect.
Read data.billing from the same response. When billing.active is
false, the workspace has not subscribed to LinkedIn yet: run the whole interview anyway,
without any LinkedIn data call (no profile or mirror prefill), and handle the
subscription at the end as described in the wannalead skill ("Workspace not
subscribed yet"). data.network_access reports Instagram and X independently.
Never present a missing LinkedIn subscription as an ended trial or as a block
on those other networks.
When the identity is unambiguous, state the account name or public identifier and continue with the first interview question in the same response. Ask for a separate identity confirmation only when the result is ambiguous.
Initialize ONBOARDING.md before the first question. Keep at least the current
phase, migration status, completed topics, deferred topics, last update date,
and next question.
2. Ask the commercial interview progressively
Ask one focused question at a time. At most, group three tightly related questions when the interface makes that easier. Never present the whole questionnaire in one message.
On a new workspace, the first question should cover the offer and initial ICP in the user's language. For example:
Pour commencer, que vends-tu aujourd'hui, à quels types d'entreprises, et quel résultat concret leur apportes-tu ? Si tu as déjà un ICP, ajoute les secteurs, tailles, zones géographiques et exclusions que tu vises.
Then cover the remaining topics progressively:
- The user's role, responsibilities, commercial goal, and expected first useful result.
- For each ICP: company fit, buyer and prescriber roles, pains, triggers, objections, exclusions, and likely audience sources.
- Company proof and boundaries: customers or references, delivery constraints, and what the company does not sell.
- Outreach style: language, formality, length, rhythm, openings, calls to action, phrases to keep or avoid, and a few real sent examples when the user is willing to provide them.
- The authoritative CRM or customer source and stable exclusion rules. Store only the source and rules, not its current contacts or statuses.
Use profile or company observations to prefill answers when helpful. Label them
as observations and ask the user to correct or confirm them. Accept unknown
or later as a valid answer and keep it under deferred topics. Do not trap the
user in the interview or start a broad network, conversation, or CRM crawl
while waiting for an answer.
After each answer:
- update only the relevant durable file;
- update
ONBOARDING.mdwithout copying the answer into it; - ask the next unanswered question.
3. Confirm the initial context
Once the user, company, first ICP, outreach preferences, commercial boundaries, and goal are captured, show a short synthesis for correction. Mark the initial interview complete only after the user confirms it.
Full knowledge is not a prerequisite. Deferred questions remain visible and later Wannalead tasks can progressively enrich the four durable files.
If billing.active was false, first give the subscription link from
billing.billing_url (it opens Stripe checkout directly), ask the user to say
when it is done, and check wannalead me again before a LinkedIn data call.
An authorized Instagram or X workflow may continue under its separate add-on.
Right after the confirmation, present the "Choose a use case" menu from the
wannalead skill (lookalikes with signals, campaigns, dormant conversations,
dormant network, saved post → ICP, messy list → contacts) and ask which one
to start with. Do not wait for the user to discover the task skills alone.
Later Wannalead tasks
Before a prospecting, account-mapping, campaign, or outreach task, load only the context files relevant to that task. Fetch current profiles, relationships, conversations, CRM status, prospects, and campaign state from their authoritative systems instead of relying on a local snapshot.
After the task, keep operational results in the response or their server-side
objects. If the task produced a potentially durable insight, explain the
evidence and ask the user whether it should update ME.md, COMPANY.md,
ICPS.md, or OUTREACH-STYLE.md.
Completion criteria
The initial onboarding is ready when:
- the LinkedIn identity is unambiguous;
ME.mdstates the user's role and goal;COMPANY.mdstates the offer, proof or current evidence gap, and boundaries;ICPS.mdcontains at least one named ICP with company and person fit;OUTREACH-STYLE.mdcontains confirmed preferences or explicitly records that examples and preferences are deferred;ONBOARDING.mdpoints to those files, records unresolved topics, and has no duplicated commercial facts;- no external write occurred without separate authorization.
Guardrails
- Use public Wannalead endpoints only for LinkedIn data.
- Poll the same job ID until terminal; never recreate a pending job.
- Stop on authentication, connection, rate-limit, or restriction errors.
- Minimize retention of private messages and CRM fields.
- Never infer permission to invite, message, follow up, launch a campaign, or mutate the CRM.
- Never call LinkedIn directly, read cookies, or access browser or gateway state.
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-onboarding">View wannalead-onboarding on skillZs</a>