getaero-io/gtm-eng-skills / skills/deepline-research
deepline-research
VerifiedUse 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.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 24.5 KB |
| THIRD_PARTY_NOTICES.md | markdown | 1.6 KB |
| evals/last30days-public-private-corpus.json | config | 15.0 KB |
| references/fanout-consolidation.md | markdown | 7.1 KB |
| references/last30days-gtm-corpus.md | markdown | 11.7 KB |
| references/query-design.md | markdown | 4.0 KB |
| references/source-map.md | markdown | 26.6 KB |
| scripts/evaluate_examples.py | script | 19.7 KB |
| scripts/evaluate_public_private_corpus.py | script | 5.8 KB |
| scripts/query_design.py | script | 23.5 KB |
| skill-metadata.json | config | 2.7 KB |
references/query-design.md
View on GitHub# Query Design
`/deepline-research` should not send the user's raw prompt to every source. The query-design layer is the core behavior to preserve from `last30days`.
Attribution: the query-cleaning, query-type, source-tiering, and supplemental extraction patterns are adapted from `mvanhorn/last30days-skill` under the MIT License. See `../THIRD_PARTY_NOTICES.md`.
Use `scripts/query_design.py` to generate the query plan:
```bash
python3 .skills/deepline-research/scripts/query_design.py "best GTM data sources for SMB consumer services companies" --depth deep
```
In Deepline runtime, use the native API route only after the public-source fanout has produced a first synthesis and source map. The API translates findings into Deepline routes and costs; it must not replace the research pass:
```bash
curl -s "$DEEPLINE_API_BASE_URL/api/v2/pre-research/plan" \
-H "Authorization: Bearer $DEEPLINE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"objective":"best GTM data sources for SMB consumer services companies","depth":"deep"}'
```
The script emits JSON with:
- `query_type`: product, concept, opinion, how_to, comparison, breaking_news, prediction, gtm_dataset, private_workflow, or custom_language
- `core_subject`: a cleaned version of the prompt with question/meta/noise words removed
- `enabled_sources`: tiered source defaults for the detected query type
- `variants`: source-specific broad and fallback queries
- `supplemental_templates`: phase-two templates for discovered handles, subreddits, domains, accounts, CRM ids, datasets, and language phrases
- `extraction_keys`: what to pull from phase-one results before supplemental fanout
- `scoring_notes`: how to rank results before synthesis
## What Was Ported
Preserve these behaviors:
- Strip common prompt wrappers and research/noise words before platform searches.
- Keep different core subjects by platform. X should be aggressive because keyword search is literal; YouTube/TikTok/Instagram should keep content-type terms like tutorial, review, and tips.
- Detect query type before choosing source defaults.
- Expand Reddit into core, original, review/opinion, and problem/issue variants depending on depth and query type.
- Expand X into the core query with a recency filter, quoted compound-term OR queries, shorter keyword fallback, and strongest-token fallback.
- Run video/social caption searches separately from web searches.
- Treat private/workflow/custom-language queries as first-class query types, not post-processing.
- Extract phase-one handles, hashtags, subreddits, domains, datasets, CRM ids, workflow ids, personas, pain phrases, objections, competitors, and category terms for supplemental fanout.
## Deepline Additions
The Deepline query planner extends `last30days` in four ways:
- `gtm_dataset` query type for provider strategy, public records, account data, contact data, intent data, and enrichment waterfalls.
- `private_workflow` query type for CRM, warehouse, product usage, workflow runs, and RevOps evidence.
- `custom_language` query type for buyer language, objections, competitor/category phrasing, and campaign hooks.
- Provider-catalog and cost-awareness hooks so synthesized research findings can become Deepline source plans with approval gates after the public pass.
## Implementation Contract
Any Deepline runtime implementation should follow this order:
1. Build query plan with `query_design.py` logic or equivalent TypeScript port.
2. Execute public broad variants across the selected source families.
3. Extract supplemental keys from phase-one evidence.
4. Execute targeted supplemental variants.
5. Normalize every result to the evidence schema in `fanout-consolidation.md`.
6. Score, dedupe, cluster, and emit coverage nudges.
7. Synthesize the `last30days`-style GTM research report.
8. Search/describe Deepline tools for the useful source families, private joins, costs, probes, and provider gaps.
Do not collapse this into one generic web search. The source-specific query variants are the quality lever.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| 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: 198k | Updated | Claude CodeCodex | Verified | Install |