Skip to content

getaero-io/gtm-eng-skills / skills/deepline-research

deepline-research

Verified

Use when the user wants a last30days-style pre-research pass in Deepline: discover the critical public, private, CRM, workflow, social, and web data sources for a research/enrichment job; compare provider coverage; estimate Deepline credit cost; recommend the source plan before building or running the workflow; or build custom language/messaging from buyer, competitor, community, and CRM evidence. Triggers: pre-research, source discovery, provider strategy, research data sources, ScrapeCreators, X/Twitter data, Reddit comments, public and private datasets, CRM data, workflow data, custom language, messaging language, pain language.

references/source-map.md

View on GitHub
# Source Map

Use this reference to choose source families for `/deepline-research`. Tool ids drift; always confirm with `deepline search` and `deepline describe`.

## Required Dataset Inventory

To replicate `last30days`-style functionality inside Deepline, every pre-research plan must account for these dataset families. Some are already first-class Deepline providers, some can be reached through generic routes, and some are explicit integration gaps to add.

| Dataset family                              | Required evidence fields                                                                                                              | Deepline route today                                                                                                               | Parity status                                     |
| ------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------- |
| Reddit threads                              | post URL, subreddit, title/body, author when available, score/upvotes, comments count, created date                                   | social-data tool, found with `deepline search`                                                                                     | native                                            |
| Reddit comments                             | comment text, comment URL/permalink, author when available, upvotes, parent post, created date                                        | social-data tool, found with `deepline search`                                                                                     | native                                            |
| X/Twitter posts                             | post URL/id, handle, text, timestamp, likes, reposts, replies, quote count, optional entity handle search                             | social-data tool, found with `deepline search`                                                                                     | native                                            |
| YouTube search/transcripts                  | video URL/id, channel, title, publish date, views/likes, transcript excerpts                                                          | social-data tool, found with `deepline search`                                                                                     | native                                            |
| TikTok                                      | video URL/id, creator, caption/transcript, hashtags, timestamp, views/likes/comments                                                  | social-data tool, found with `deepline search`                                                                                     | native                                            |
| Instagram Reels/posts                       | URL/id, creator, caption/transcript, timestamp, views/likes/comments                                                                  | social-data tool, found with `deepline search`                                                                                     | native                                            |
| Hacker News                                 | story/comment URL, title/text, author, points, comments, created date                                                                 | Hacker News tool, found with `deepline search`                                                                                     | native                                            |
| Bluesky                                     | post URL/id, handle, text, timestamp, likes/reposts/replies                                                                           | social-data tool, found with `deepline search`                                                                                     | native                                            |
| Truth Social                                | post URL/id, handle, text, timestamp, likes/reposts/replies                                                                           | no native tool; generic web search and extraction                                                                                  | generic route or gap                              |
| LinkedIn posts                              | post URL, author, text, timestamp, reactions, comments                                                                                | LinkedIn post search, comments and reactions tools, found with `deepline search`                                                   | native                                            |
| Podcasts                                    | episode URL, show, title, publish date, transcript excerpts                                                                           | podcast episode and show search, found with `deepline search`                                                                      | native                                            |
| Web/news/blogs/docs                         | URL, title, snippet/full text, publish date, source domain, citations                                                                 | web search and page-extraction tools                                                                                               | native/generic route                              |
| Public registries and niche public datasets | registry/API URL, dataset title, row identifiers, entity names, addresses, status/taxonomy/license fields, update date, source agency | generic web/API discovery through web search, page extraction, and `generic_http`; native tool only when catalog finds one         | generic route unless catalog finds native support |
| Company/account datasets                    | domain, name, LinkedIn URL, firmographics, funding, industry, geo, employee count, source URL                                         | company search and enrichment tools                                                                                                | native                                            |
| People/contact datasets                     | full name, title, company, LinkedIn URL, email/phone when approved, confidence, source                                                | people search and contact enrichment tools                                                                                         | native                                            |
| Jobs/hiring signals                         | job title, company, location, posted date, description, URL                                                                           | job-posting search tools                                                                                                           | native                                            |
| Technographics/install base                 | domain/company, vendor/tool, category, detected date, confidence/source                                                               | technology-detection tools                                                                                                         | native                                            |
| Funding/company events/news                 | company, event type, date, amount/round when available, source URL                                                                    | funding and company-event tools, web search                                                                                        | native/generic route                              |
| CRM data                                    | account/contact/deal ids, owner, stage, lifecycle timestamps, activity fields, custom fields                                          | the CRM connected to the workspace, found with `deepline search crm`                                                               | private connector                                 |
| Warehouse/semantic metrics                  | metric name, dimensions, filters, result rows, rendered SQL/audit trail                                                               | the warehouse semantic-layer tools; read `deepline-analytics`                                                                      | private connector                                 |
| Product/workflow usage                      | org/user/account ids, event/run ids, timestamps, status, outputs, credits/run metadata                                                | `plays`, workflow/session/usage tools, warehouse/customer DB                                                                       | private connector                                 |
| Support/call/docs data                      | transcript/note URL/id, speaker/contact, timestamp, summary/snippet, related account                                                  | CRM, chat, docs, and call-transcript connectors where configured                                                                   | private connector                                 |
| Sheets/CSVs/customer-owned lists            | row ids, domains/emails, source file/sheet id, provenance columns                                                                     | user CSV, Google Sheets, Deepline Playground                                                                                       | private/customer source                           |
| Custom language corpus                      | exact quote/snippet, speaker/source, audience/persona, context, timestamp, engagement or CRM outcome when available                   | Reddit/X/LinkedIn/YouTube/TikTok/Instagram, sales calls, support tickets, CRM notes, win/loss notes, call transcripts, docs/sheets | native/generic/private mix                        |

Minimum useful parity for a last-30-days community research run is: web/news, Reddit threads, Reddit comments, X posts, YouTube transcripts, TikTok or Instagram when relevant, HN, and at least one private/customer context source when the job is GTM/customer-facing.

For custom language runs, minimum useful parity is: at least one community/social source, at least one long-form context source (YouTube transcript, podcast, blog, call transcript, support thread, or Reddit comments), and one business-context source such as CRM stage/outcome, persona, segment, deal notes, or target account list.

## What To Preserve From last30days

The useful pattern is not the local script itself. Preserve:

- broad source taxonomy: Reddit, X/Twitter, YouTube, TikTok, Instagram, HN, Bluesky, Truth Social, and web
- recency-first retrieval for trends and news
- community evidence with engagement stats
- source coverage stats and explicit missing-source nudges
- comparison mode that researches each side separately plus direct comparison
- synthesis weighted by source quality, engagement, and cross-platform agreement
- two-phase fanout: broad parallel source search, then supplemental searches from discovered handles, subreddits, domains, datasets, and entities
- common evidence rows before scoring, dedupe, clustering, and synthesis

Deepline adaptation:

- route through Deepline tools, plays, workflows, CRM connectors, and warehouse connectors
- quote Deepline credits only
- save outputs in CSVs/runs/playgrounds where agents and users can inspect them
- convert the source plan into a repeatable workflow when the user wants automation
- add private/source-of-truth joins, provider-cost checks, approval gates, and workflow activation paths that `last30days` does not own

## Public Source Families

| Need                               | Start With                                                                                                                                                    | Notes                                                                                                                                                                                                                                                                      |
| ---------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| General web/news/source discovery  | web search and page-extraction tools                                                                                                                          | Use search to discover URLs; use extraction for known pages or JS-rendered sources.                                                                                                                                                                                        |
| Public registries / niche datasets | web search, page extraction, `generic_http`                                                                                                                   | Treat public registries as materializable sources even without native Deepline tools. Example: NPI registry/provider taxonomy data can be found and pulled through generic web/API search/extraction, then joined by NPI, organization name, address, phone, and taxonomy. |
| Reddit threads/comments            | social-data tool, found with `deepline search`                                                                                                                | Search returns posts with score and comment counts; pull the comment tree per post URL.                                                                                                                                                                                    |
| X/Twitter posts                    | X/Twitter tool, found with `deepline search`                                                                                                                  | Search with query operators, a handle's recent posts, replies, quotes and mentions. Browser cookies are not an acceptable backend Deepline integration pattern.                                                                                                            |
| YouTube search/transcripts         | social-data tool, found with `deepline search`                                                                                                                | Preserve channel, URL, views, publish date, and transcript excerpts.                                                                                                                                                                                                       |
| TikTok/Instagram                   | social-data tool, found with `deepline search`                                                                                                                | Use captions/transcripts plus engagement. For local-business contact workflows, also consider Instagram profile bio links/contact fields as candidate contact-data signals.                                                                                                |
| Facebook pages/profiles            | Discover public page/profile contact details when local businesses, restaurants, storefronts, or social-first companies may publish email/phone/website there | Candidate route through social profile tools or generic web extraction.                                                                                                                                                                                                    |
| Hacker News                        | Hacker News tool, found with `deepline search`                                                                                                                | Capture points, comments, author, URL, and story age.                                                                                                                                                                                                                      |
| Bluesky                            | Bluesky tool, found with `deepline search`                                                                                                                    | Search public posts by keyword, author, mention, tag or domain within a date range.                                                                                                                                                                                        |
| Truth Social                       | No native tool; treat as a gap unless catalog search finds a current route                                                                                    | Prefer API-backed access over brittle page scraping.                                                                                                                                                                                                                       |
| LinkedIn posts                     | LinkedIn post tools, found with `deepline search`                                                                                                             | Search posts by keyword, then pull comments and reactions for the posts that matter.                                                                                                                                                                                       |
| Podcasts                           | podcast search tool, found with `deepline search`                                                                                                             | Find episodes whose transcript mentions a keyword; preserve show, episode URL and publish date.                                                                                                                                                                            |
| Jobs/hiring signals                | job-posting search tools                                                                                                                                      | Jobs are high-intent and often better than marketing pages.                                                                                                                                                                                                                |
| Tech stack/install base            | technology-detection tools                                                                                                                                    | Confirm whether the tool supports domain lookup, vendor lookup, or customer list search.                                                                                                                                                                                   |
| Funding/company events             | funding and company-event tools, web search                                                                                                                   | Verify event date, source URL, and company domain.                                                                                                                                                                                                                         |
| Ads/audience signals               | search-demand tools and the ads-audience connectors                                                                                                           | Use only when the research question needs market demand or ad targeting.                                                                                                                                                                                                   |

## Private And Customer-Owned Sources

Private data often beats public data. Do not build a public-only plan when the user has relevant internal data.

| Dataset                    | Deepline Route                                                   | Good Join Keys                                                       | Guardrails                                                                                                                        |
| -------------------------- | ---------------------------------------------------------------- | -------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| CRM                        | the connected CRM's tools after catalog search/describe          | account or company id, domain, contact email, deal or opportunity id | Do not scan broad objects blindly. Describe the object model and list fields first; preserve object ids and lifecycle timestamps. |
| Warehouse / semantic layer | the warehouse semantic-layer tools after catalog search/describe | account id, domain, email, user id, date                             | Use semantic metrics before raw SQL. Read `deepline-analytics`.                                                                   |
| Customer DB                | `customer-db` / `query_customer_db` when available               | workspace/org/account ids, domain                                    | Query narrowly; avoid large scans.                                                                                                |
| Workflow/run data          | `plays`, workflow/run/session tools if present                   | play id, run id, org id, output dataset id                           | Billing belongs in metadata, not row output.                                                                                      |
| Product analytics          | analytics/browser or warehouse-backed tools if exposed           | user id, org id, account id, event date                              | Aggregate when possible; do not leak PII unnecessarily.                                                                           |
| Support/calls/meetings     | CRM, call-transcript, chat, and docs connectors where configured | contact email, company domain, meeting id                            | Quote only relevant snippets and cite source location.                                                                            |
| Sheets/CSVs                | user-provided CSV, Google Sheets connector, Deepline Playground  | stable ids, domain, email                                            | Never enrich source files in place; write derived outputs.                                                                        |

## Custom Language Workflow

Use this when the user needs language that sounds like the market, not generic AI copy.

Source hierarchy:

1. **Buyer/customer words**: sales calls, support tickets, call transcripts, CRM notes, win/loss notes, onboarding calls, email replies.
2. **Community words**: Reddit comments, X/LinkedIn posts, YouTube transcripts/comments, TikTok/Instagram captions, niche forums, Slack/community exports when provided.
3. **Competitor/category words**: competitor homepages, reviews, docs, pricing pages, G2/Capterra-like reviews when reachable, customer case studies.
4. **Internal desired framing**: product docs, positioning docs, ICP notes, campaign briefs, prior high-performing outbound.

Extract fields:

- exact phrase or quote
- source and timestamp
- persona/segment/account when known
- emotion or pain type
- objection or desired outcome
- product/category term
- proof point or evidence source
- suggested reuse: subject line, opener, ad hook, landing-page section, sales-call talk track, CRM personalization field

Guardrails:

- Keep exact quotes separate from rewritten copy.
- Do not treat high-engagement social language as a factual claim.
- Preserve audience context; do not reuse SMB language for enterprise personas without checking fit.
- For outbound at scale, output structured columns a Play can read, not just prose.

## CRM Method

1. Identify the business question before touching CRM data.
2. List candidate CRM objects and fields; inspect schema/describe endpoints first.
3. Choose stable join keys: CRM object ids first, normalized domains/emails second.
4. Pull the smallest useful slice. Prefer recent, scoped records over whole-object exports.
5. Preserve provenance columns: object id, source system, field name, timestamp, and source URL when available.
6. Separate facts from inferences. CRM activity fields often reflect sales effort, not buyer fit.
7. Join private and public sources only after each side has passed a quality gate.

## Provider Gaps Worth Considering

These are the remaining gaps against `last30days` parity:

- Truth Social: no native tool.
- TikTok and Instagram transcripts: the native tools return captions and engagement; a transcript needs a general scraping tool.

For each proposed provider, require:

- test endpoint or tool action for agent validation
- pricing in Deepline credits
- sample payload and sample output fixture
- evidence fields: URL, author/channel, timestamp, engagement, text excerpt
- failure modes and rate-limit behavior

## Cost Method

Use current tool metadata, not provider websites, for Deepline-facing estimates:

```bash
deepline describe <tool-id>
deepline billing balance
```

Estimate separately:

- discovery cost: queries/searches
- extraction cost: pages/posts/comments/transcripts
- enrichment cost: per row or per successful result
- synthesis cost: `deeplineagent` or model-backed steps
- rerun/freshness cost: what changes when run daily/weekly

If pricing is missing, say "unknown until describe/probe" and do not scale.

## Output Quality Bar

A good pre-research answer is opinionated and operational:

- A good O&P report should surface the actual research learning: official taxonomy over-includes wig shops and generic DME suppliers, Google reviews are a patient-volume proxy but not an absolute estimate, and the useful public source is the registry plus storefront validation, not generic company databases.
- "Use web search to find source URLs, page extraction to read them, a job-posting tool for hiring signals, the CRM for customer context, and note the gap for Reddit comments."
- "Pilot should cost about X Deepline credits under these assumptions; full run is unknown until we know result count."
- "This source is weak because it lacks comments/transcripts/author timestamps."
- "After the public research pass, this private dataset should be joined because it tells us which accounts actually converted."

A bad answer is a generic list of providers without contracts, costs, join keys, or probe plan.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall