17 AI skills for finding customers and making content
account-research
turn any company into a one-page brief in ten minutes
How the two work together
Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.
Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.
No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.
Prompt for Claude
---
name: account-research
description: One-page, sourced sales brief on a named company before a first message or call.
---
# Turn any company into a one-page brief in ten minutes
Every fact dated and sourced, then the angle, who to contact, what to ask.
## Claude does
1. Get {COMPANY}, {DOMAIN}, {OFFER}, {ICP} with disqualifiers, {GOAL}, any CRM export. No goal or ICP: ask once, do not start.
2. Check `research-ledger.md`; reuse a brief under 7 days old and say so.
3. Seven fixed searches: "{COMPANY}" (home, about), "{COMPANY} news", "{COMPANY} funding", "{COMPANY} careers", "[person] {COMPANY} LinkedIn", "{COMPANY} product", "{COMPANY} customers".
4. One dated fact with URL per line, plain notes; skip pages not about this company. Fails a disqualifier: stop, say why.
5. Tag notes by brief section; hiring: open roles by department and the problem behind each; news: why it matters for this outreach.
6. Two people, likely buyer and starter: name, title, LinkedIn URL, one professional and one personal hook from public posts.
7. Reflect: any missing, uncertain or placeholder section gets one more targeted search; still unknown becomes a discovery question, never a blank.
8. Decide: positive signals with evidence; concerns and what to watch; unknowns; the angle, one honest sentence from their gap to {OFFER}, or stop, not ready; entry point and why; plain-words opening hook; three discovery questions (situation, pain, decision process); do-not-mention list.
9. Save `notes-<company>-<date>.md` (Facts, People, Decisions), fill CODEX.md, hand off.
## Then Codex does
Codex files the notes as brief, Clay outreach lines, sources list and ledger row: bulk formatting, no judgement.
```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```
Fill first: {COMPANY}, {DOMAIN}, {GOAL}, {RESEARCH_DATE}, {NOTES_FILE}.
## Claude checks
1. Headings exact, in order, nothing before or after; one fact per bullet; Quick Take the only prose, opening "{COMPANY} is a ... that ... for ...".
2. No placeholders or "no information found"; gaps only under Unknown (Ask in Discovery).
3. Every fact, date, number and URL matches the notes; funding and news rules kept; sources lines equal brief facts.
4. Angle, hook, questions verbatim; hook has no jargon or dates; do-not-mention items nowhere; ledger row appended.
Failures go back to Codex with the line and the rule; never patch by hand.
## Rules
- Date every fact; never invent a number: "roughly 50 staff (LinkedIn)" yes, "52 staff" unsourced no.
- No goal and ICP: no research. Fails a disqualifier: stop, say so.
- Public information only; no guessed emails or phones; nothing that reads as surveillance (personal posts, litigation, layoffs) unless they raised it publicly.
- One fact per bullet, no paragraphs, padding, jargon or commentary; never date the hook; the brief is not the outreach.
- CRM and enrichment outrank the web; funding round once, never a range; news 90 days, newest first.
- Notes first, schema second, reflect before shipping; lead with their gap, not the pitch; calibrate on a known account first.
## If Codex is not installed
Claude does that half and says so.
## Built from
- Anthropic knowledge-work-plugins: https://github.com/anthropics/knowledge-work-plugins/blob/main/sales/skills/account-research/SKILL.md (23,876 stars, api.github.com)
- ericosiu/ai-marketing-skills: https://github.com/ericosiu/ai-marketing-skills/blob/main/lead-dossier/SKILL.md (3,482 stars, api.github.com)
- guy-hartstein/company-research-agent: https://github.com/guy-hartstein/company-research-agent/blob/main/backend/prompts.py (2,272 stars, api.github.com)
- Clay, "11 AI prompts for sales prospecting research": https://www.clay.com/blog/11-ai-prompts-for-sales-prospecting-research (Clay article)
- langchain-ai/company-researcher: https://github.com/langchain-ai/company-researcher/blob/main/src/agent/prompts.py (215 stars, api.github.com, archived)
Prompt for Codex
# Turn any company into a one-page brief in ten minutes - Codex task
Research and decisions are done. You file and format, adding nothing.
## You are given
- {COMPANY}, {DOMAIN}, {GOAL}, {RESEARCH_DATE} (YYYY-MM-DD).
- {NOTES_FILE}, three parts. Facts: one per line, `[section] fact, date, URL`. People: two, each with name, title, LinkedIn URL, professional and personal hook. Decisions: positive signals, concerns, unknowns, angle, entry point, opening hook, three discovery questions, do-not-mention list.
Slug: company name, lower-case, hyphenated; {date} is {RESEARCH_DATE}.
## Produce
1. `brief-{slug}-{date}.md`, under 450 words before Sources, these headings only, in order:
- `## Quick Take`: 2-3 sentences opening "{COMPANY} is a [what] that [does what] for [whom]", then the angle verbatim from Decisions.
- `## Company Profile`: table: founded, HQ, size, funding, model, customers; present facts only.
- `## Recent News`: `- **[Headline]**, [date], [why it matters]`, newest first.
- `## Hiring Signals`: open roles by department, the problem behind each.
- `## Key People`: name, title, LinkedIn URL, both hooks.
- `## Tech Stack`: only if in Facts.
- `## Qualification Signals`: `### Positive Signals`, `### Potential Concerns`, `### Unknown (Ask in Discovery)`, bullets from Decisions.
- `## Recommended Approach`: `**Best Entry Point:**`, `**Opening Hook:**`, `**Discovery Questions:**` 1-3, verbatim from Decisions.
- `## Sources`: `- [page](URL)`, one per distinct URL.
2. `lines-{slug}-{date}.md`, each with source fact in brackets: mission under 6 words; who they sell to, up to three job titles; B2B or B2C, nothing else; per open role, up to five: "I saw [short company name] is hiring a [title]. In my experience this means you're trying to improve [the problem behind the posting]."
3. `sources-{slug}-{date}.md`: one line per brief fact, in brief order: fact, date, URL as given.
4. Append to `research-ledger.md` (create if absent): `{RESEARCH_DATE} | {COMPANY} | {DOMAIN} | {GOAL} | brief path`.
British English.
## Rules
- One fact per bullet; Quick Take the only prose; no commentary before or after.
- No "no information found", "N/A" or placeholders; omit empty sections; Qualification lists take Decisions only.
- Copy dates, numbers, names and URLs exactly, never rounded or completed; flag undated facts.
- Each funding round once (same month and amount is one round), never a range; news over 90 days before {RESEARCH_DATE}: left out, flagged.
- Lines: no jargon, no reference to when an article ran, company name as a person says it.
- Do-not-mention items appear nowhere; a matching Fact is dropped and flagged.
- Two people; any other count filed as given, flagged.
## Return
Print last: four file paths; brief word count before Sources; fact count and sources line count (must match); people; empty sections; undated, out-of-window or dropped facts; flags or "none".
Built from the best public work on this
Sources: account-research
Built from
- Anthropic knowledge-work-plugins, sales/account-research SKILL.md: https://github.com/anthropics/knowledge-work-plugins/blob/main/sales/skills/account-research/SKILL.md (23,876 stars, api.github.com, pushed 4 Sep 2026; official Anthropic repo; Anthropic's own BD team runs this pattern nightly)
- ericosiu/ai-marketing-skills, lead-dossier and sales-playbook SKILL.md: https://github.com/ericosiu/ai-marketing-skills/blob/main/lead-dossier/SKILL.md (3,482 stars, api.github.com, pushed 2 Sep 2026)
- guy-hartstein/company-research-agent, backend/prompts.py: https://github.com/guy-hartstein/company-research-agent/blob/main/backend/prompts.py (2,272 stars, api.github.com, pushed 3 Sep 2026; live demo at companyresearcher.tavily.com)
- Clay, "11 AI prompts for sales prospecting research" by Eric Nowoslawski: https://www.clay.com/blog/11-ai-prompts-for-sales-prospecting-research (article on the Clay GTM platform, read in full; the prompts Clay users run at scale over enrichment fields)
- langchain-ai/company-researcher, src/agent/prompts.py: https://github.com/langchain-ai/company-researcher/blob/main/src/agent/prompts.py (215 stars, api.github.com; official LangChain org; repository archived March 2026, used as a pattern reference)
All star counts were read directly from api.github.com on 4 Sep 2026 and matched exactly. Also checked and not used: exa-labs/company-researcher (1,491 stars, source checklist only) and amplemarket/skills (5 stars, requires the Amplemarket MCP server).
Best public prompt we found for this job
From https://github.com/anthropics/knowledge-work-plugins/blob/main/sales/skills/account-research/SKILL.md, quoted verbatim:
```
Quick Take
[2-3 sentences: Who they are, why they might need you, best angle for outreach]
[...]
Recent News
- **[Headline]** — [Date] — [Why it matters for your outreach]
[...]
Qualification Signals
Positive Signals
- ✅ [Signal and evidence]
Potential Concerns
- ⚠️ [Concern and what to watch for]
Unknown (Ask in Discovery)
- ❓ [Gap in understanding]
Recommended Approach
**Best Entry Point:** [Person and why]
**Opening Hook:** [What to lead with based on research]
**Discovery Questions:**
1. [Question about their situation]
2. [Question about pain points]
3. [Question about decision process]
Sources
- [Source 1](URL)
[...]
Step 2: Web Search (Always)
Run these searches:
1. "[Company name]" → Homepage, about page
2. "[Company name] news" → Recent announcements
3. "[Company name] funding" → Investment history
4. "[Company name] careers" → Hiring signals
5. "[Person name] [Company] LinkedIn" → Profile info
6. "[Company name] product" → What they sell
7. "[Company name] customers" → Who they serve
**Extract:**
- Company description and positioning
- Recent news (last 90 days)
- Leadership team
- Open job postings
- Technology mentions
- Customer base
[...]
Step 5: Synthesize
1. Combine all sources
2. Prioritize enrichment data over web (more accurate)
3. Add CRM context if exists
4. Identify qualification signals
5. Generate talking points
6. Recommend approach
```
Want this running in your business?
I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.
Foxera