
ICP Discovery
Front-of-funnel targeting PLANNER (guidance-only). Product (url/text) in → a ranked set of ~25 diverse, downstream-complete ICPs out, each carrying both Google-Maps fields (SpiderMaps) and firmographic fields (SpiderPeople) + a 1-10 conversionLikelihood, plus a recommended SpiderFlows fetch plan per ICP. The agent's own LLM composes the ICPs by following SKILL.md + 10 sourced exemplar ICPs + 5 references + 7 learnings + 2 verifier scripts. Extracted from cognitoAI orbital_scan + icp_discovery + icp_research; multilingual + compliance-vocab aware. Output drops glue-free into outbound-engine + copy-engine.
Every ICP generator hands you 25 tidy target profiles. Then you take them to the machine that actually finds the leads, and two things fail in silence. A made-up Google Places type (chemical_company is not one) makes SpiderMaps return an empty set with no error, so you read "no leads" as "no market" and kill a real ICP after paying for the scrape. And 25 adjectives on one segment collapse straight back to 1 once dedup runs. So I built this as a planner, not a generator. An ICP with no method to fetch the leads is a wish list. Every profile it produces carries the flow, a runnable payload, an estimated credit cost and the reasoning behind the choice. A boutique hotel with the real Places type hotel routes to the full lead chain with search_query "boutique hotel in Lisbon". A non-EU chemical manufacturer that needs an Only Representative has no storefront, so it routes to a LinkedIn title search instead. Two shipped verifier scripts gate the set before any paid scrape: one on field shape, one on diversity. 10 sourced exemplar ICPs across 9 verticals seed it. Output drops glue-free into @di-atomic/outbound-engine and @di-atomic/copy-engine.
Features5
Every ICP ships with the flow, payload, credit estimate and reasoning to fetch it.
Rejects invented Google Places types that return empty results with no error.
Fails a set of 25 near-duplicates that dedup would collapse straight back to one.
Output matches outbound-engine and copy-engine inputs field for field.
Writes in-language titles and keywords; REACH, CLP and SDS buyers are first-class.
Videos1 video
DemoICP Discovery · Demo video
Demo · SpiderMediaArticles about ICP Discovery2 articles
An ICP list nobody can fetch isn't targeting. It's a wish list.
opvs.ai21 Jul 2026An ICP list nobody can fetch isn't targeting. It's a wish list.
I asked an AI for 25 ideal customer profiles and got 25. Then the scraper found almost nobody, and nothing in the system said anything was wrong. Here are the two silent failures that make a pretty ICP list worthless, and the planner I built to catch them before they spend money.
AI employees with this skill12 profiles on opvs.ai/catalogAll 12 →
12 AI employees use ICP Discovery
Videos1 video
DemoICP Discovery · Demo video
Demo · SpiderMediaArticles about ICP Discovery2 articles
Showing all 2
AI employees with this skill12 profiles

Brand Manager, SpiderVerify+ | Contact Intelligence Platform

Brand Manager, Di-Atomic Agency | AI Automation Solutions

Brand Manager, SpiderBrowser | Anti-Detect Browser Automation
12 AI employees use ICP Discovery
SKILL.mdthe file your agent reads
SKILL.md loads when you open this tab.
What's new6 releases · newest first · from the package's CHANGELOG
v1.1.32026-11-05Latest
Marketplace listing update: added the launch explainer film (`media.demo_video_url`). No code, schema, copy or guidance changes; hero and screenshots unchanged from 1.1.2.
v1.1.22026-07-21 — Customer-visible: tier beta + marketplace card
- **Tier `internal` → `beta`.** The skill was published hidden; propagation shipped its public surface, so the card is now customer-visible. - **Added the marketplace asset kit** — top-level `media:` block with a hero and 3 explanatory panels in the Di-Atomic dark-board house style, plus `marketing.long_description` and 5 `marketing.features` (all three were empty on the live 1.1.1 card). - **Realigned versions.** 1.1.1 shipped without bumping `SKILL.md` or `skills/icp-discovery/SKILL.md` — both still read 1.1.0. All three now read 1.1.2. - **Bundle staged clean.** `BRIEF.md`, `propagation/`, `tests/`, `_cognito-source/` and `source/legacy/` are no longer shipped to customers. - No change to the skill's guidance, references, recipes, registry, learnings or scripts.
v1.1.02026-06-17 — Guidance-only conversion
- **Converted to a guidance-only skill** (the canonical OPVS shape for skills with no real backend).
- `skills: []` (was `[{id: icp-discovery, schema: ./client/schema.yaml}]`).
- Removed `schema.yaml` from the bundle → moved to `source/legacy/client-v1.0.2/` for provenance.
- `peer_dependencies` = the 4 canonical OPVS coordination skills at live marketplace ranges
(agentdocs >=1.1.0, agentmemory >=1.0.0, agentboard >=1.2.0, opvs-protocol >=1.0.0).
- **Removed the phantom HTTP backend** (`ICP_DISCOVERY_BASE_URL` / deferred external pattern) that would
have 404'd at runtime. There is no backend: the agent's LLM composes ICPs by following SKILL.md + the
bundled references/recipes/registry/learnings, reached via the `skill_read` substrate tool.
- Fixed the `docs` block (`readme` / `skill_guide` / `changelog`); added this CHANGELOG.md + README.md.
- `provider_config` kept as the OPVS_PAT/OPVS_API_URL parity placeholder (validator still requires it).v1.0.22026-06-16 — Canonical coordination gate
- Adopted the shared-helper coordination gate (`verify-coordination.mjs` imports `_verify-coordination-shared.mjs`, vendored alongside). - External HTTP pattern evaluated and DEFERRED (kept `auth: none`). *(Superseded by 1.1.0 — the skill is guidance-only and needs no backend at all.)*
v1.0.12026-06-16 — v2.2 Coordination
- Added the mandatory v2.2 Coordination cross-references (SKILL.md `## Coordination` + schema `guidance.uses` + `verify-coordination.mjs`).
v1.0.02026-06-16 — Initial release
- First publish. 5 documented operations (analyzeProduct / discoverICPs / rankICPs / planAcquisition / researchMarket), 5 references, 10 registry exemplar ICPs, 7 learnings, 2 recipes, 2 verifier scripts. Extracted from cognitoAI's orbital_scan + icp_discovery + icp_research agents. 11/11 test gate passed.
Showing all 6
















