Skip to main content
Deepline Skills are the packaged operating instructions for common GTM work. Start with deepline-gtm for prospecting, enrichment, research, and outreach. Use a specialized skill when the job needs a deeper method, such as research planning before paid data work or a human review loop after a pilot. Skills do not replace the Deepline CLI. The skill plans and checks the job; the CLI searches tools, runs enrichment, executes plays, and exports results.

​
Start with Quickstart

The Quickstart is the one-request install. Ask your coding agent:
That setup installs the CLI and Deepline Skills for supported coding agents, authorizes your workspace, and verifies the result. Then return here to choose the right skill and start with a small request.
Prefer a Claude plugin or MCP instead? Open Use Deepline from an agent in the Get started navigation.

​
The default operating pattern

Use the same safety loop for every paid or customer-data workflow:
  1. Define the output, format, evidence, and acceptance checks.
  2. Inspect the input and preserve existing good data.
  3. Search and describe the tools or plays that may execute the job.
  4. Run a small pilot.
  5. Review coverage, evidence, failures, and the Deepline-facing spend estimate.
  6. Approve the full run.
  7. Export the result or save the accepted workflow as a reusable play.

​
Core skills

Choose a skill by the first decision you need to make:
  • You have a GTM outcome, CSV, or audience: start with deepline-gtm.
  • You need to find the evidence and cheapest route first: use deepline-pre-research before paid data work.
  • You need durable code, a schedule, or a webhook: use deepline-plays.
  • A pilot needs row-level approval or correction: use deepline-plays-review.
This section is generated from docs/customer-facing-skills.json and validated against the shipped .skills source. Run bun run docs:skills after changing the recommended set or its recipes.

​
deepline-gtm

The default skill for prospecting, enrichment, research, and outreach jobs. Start here when the request begins with a CSV, a target audience, or a GTM outcome. Use it for
  • Clean or enrich a CSV
  • Build account and contact lists
  • Find and validate LinkedIn URLs
  • Research and write outreach
Workflow
  1. Give the agent the input file or outcome and define the accepted result.
  2. Let the skill route to the relevant GTM guide or recipe.
  3. Run a small pilot and inspect coverage, evidence, and estimated spend.
  4. Approve the full run only after the pilot passes.
  5. Export the result and save the accepted workflow as a play when it should be reused.

​
deepline-engine

Builds, publishes, and verifies an orchestrator Play as a durable state machine, including a small paid pilot through its real transitions. Use it for
  • Turn a user-defined state machine into a Deepline Play
  • Define input and output tables for each state
  • Reuse workspace Plays or prebuilts for state transformations
  • Publish the checked engine and verify terminal, failed, invalid, and retried transitions with a bounded paid pilot
Workflow
  1. Describe the states, transitions, and transformations; the agent infers routine names and structural choices.
  2. Let the agent choose stable row identity, idempotency keys, table contracts, and an advancement model from the workflow.
  3. Find or create a transition Play for each state, then author an orchestrator Play that calls it, records the result, and hands off to the next state.
  4. Validate every transition and replay path, then publish the child Plays and orchestrator in dependency order.
  5. Run one synthetic record through the real paid transitions in an internal/test workspace, adding at most two for branch coverage under a disclosed aggregate cap, then show final outputs, spend, and replay safety; request exact traces only when needed.

​
deepline-plays-review

Reviews a play and its output with a human feedback loop. Use it after a pilot when rows need approval, corrections, or a repeatable review sheet. Use it for
  • Review pilot output row by row
  • Collect corrections in a Google Sheet
  • Revise checks after failures
  • Decide whether a play is ready to scale
Workflow
  1. Point the skill at the play, run, or output file.
  2. Open or resume the review loop and inspect the sample.
  3. Record row-level feedback and separate fixes from acceptable misses.
  4. Revise the play or contract, then rerun the affected rows.
  5. Continue until the acceptance checks pass or a clear blocker remains.

​
deepline-pre-research

Plans a research job before paid data work. It maps public and private sources, designs source-specific queries, finds material datasets, and recommends the Deepline route. Use it for
  • Research a market or buyer problem
  • Find public datasets and source coverage
  • Design queries before provider execution
  • Estimate research cost before approval
Workflow
  1. State the research objective, audience, time window, and required evidence.
  2. Fan out across public sources and synthesize what is already knowable.
  3. Map candidate Deepline tools and exact datasets for the remaining gaps.
  4. Describe tool contracts and estimate cost before paid execution.
  5. Return a sourced research report and a recommended execution workflow.

​
deepline-ads-audiences

Builds, enriches, audits, and uploads paid advertising audiences for Google, Meta, and LinkedIn. It is for audience matching, not cold outbound. Use it for
  • Increase customer-list match coverage
  • Build suppression audiences
  • Audit hashing and normalization
  • Upload and compare Google or Meta audiences
Workflow
  1. Confirm data rights, geography, targeting intent, and suppression requirements.
  2. Discover and confirm the destination ad accounts.
  3. Build a first-party baseline and an enriched audience using the approved coverage mode.
  4. Normalize, hash once, deduplicate, and audit the payload.
  5. Upload separate platform audiences, read back status, and report coverage lift.

​
deepline-analytics

Answers governed RevOps and GTM questions from a configured warehouse or semantic layer, with Snowflake as the primary analytics surface. Use it for
  • Answer funnel and pipeline questions
  • Compare periods using governed metrics
  • Find the right semantic object
  • Use raw SQL only when the semantic layer cannot answer
Workflow
  1. Translate the question into a metric, grain, filters, and comparison period.
  2. Choose the governed semantic object that owns the metric.
  3. Run the smallest query that can answer the question.
  4. Check totals, periods, and filter semantics before interpreting the result.
  5. Report the answer, query assumptions, and any data-quality limits.

​
niche-signal-discovery

Finds observable signals that distinguish closed-won from closed-lost accounts, then turns them into an evidence-backed scoring model and prospect list. Use it for
  • Discover ICP signals from won and lost accounts
  • Analyze jobs, websites, and technology evidence
  • Build an account scoring model
  • Return net-new prospects that fit the discovered pattern
Workflow
  1. Research the target company, buyers, competitors, and ecosystem.
  2. Prepare and deduplicate won and lost domain lists.
  3. Extract website, job, and technology evidence with approval gates.
  4. Run the quality gate and differential analysis.
  5. Interpret the signals and return ten deduplicated net-new prospects.

​
deepline-plays

Creates durable, reusable Deepline plays when a one-off enrichment command is not enough. Use it for multi-tool logic, schedules, webhooks, joins, or custom exports. Use it for
  • Compose multiple tools or plays
  • Add fallback logic and durable datasets
  • Build scheduled or webhook-triggered workflows
  • Create inspectable TypeScript automation
Workflow
  1. Freeze the job contract: input, output, checks, spend, and approval boundaries.
  2. Search and describe existing plays and tools before authoring new code.
  3. Choose direct execution, a wrapper, or a new custom play.
  4. Check the play without running it, then pilot one to three rows.
  5. Inspect the output, report the run and export, and publish only after the pilot passes.

​
Recipes included with deepline-gtm

The router loads these focused recipes when the request matches. You do not need to invoke them separately.

​
What is not in this list

deepline-feedback is a support utility, and deepline-quickstart is a guided demo. They still ship with Deepline, but they are not part of the recommended day-to-day customer workflow above.

​
Keep the docs in sync

The docs generator fails when a recommended skill no longer exists, its source name changes, its customer workflow is incomplete, or a deepline-gtm recipe has no title or summary. The skills build runs the generator automatically.