What are Agent Skills?
The sixtyfour-ai/skills repo packages Sixtyfour as an Agent Skill — a single directory of instructions that teaches an AI coding assistant how to use the Sixtyfour API correctly.Prerequisites
- Have an AI coding assistant installed (Claude Code, Cursor, etc.).
- Sign up at app.sixtyfour.ai.
- Generate a key by following Get an API Key.
- Set
SIXTYFOUR_API_KEYin your agent’s shell.
Install
The Sixtyfour Agent Skills are not yet available in the Cursor and Claude Code marketplaces. For now, follow the instructions below to install manually.
Cmd+Shift+P on macOS, Ctrl+Shift+P on Windows/Linux) and run Developer: Reload Window. Then open Cursor Settings → Plugins and confirm Sixtyfour appears with a Local badge and 1 skill.
Usage
Once installed, your agent can:- Investigate people — “Tell me everything about Jane Doe at Acme Corp — title, background, LinkedIn, career trajectory”
- Research companies — “Research Ramp — revenue, headcount, funding, engineering leadership”
- Search at scale — “Find 500 VPs of Sales at Series B SaaS startups in NYC”
- Build pipelines — “Search for AI startups, find their CTOs, get contact info, and export to CSV”
- Due diligence — “Deep investigation on this person for compliance review” (medium/high tier)
Troubleshooting
API key not set
Check that the variable is exported in the same shell your agent runs in:Environment variables don’t persist in Cursor or VS Code
Cursor and VS Code don’t always inherit shell environment variables — especially on macOS when launched from the Dock or Spotlight. Fix it one of three ways:- Launch from a terminal that already has the variable set:
cursor . - Add the export to
~/.zshrcor~/.bashrcso it’s always available - Add it to
.claude/settings.jsonunderenv
Locating poll_job.py
The polling script ships inside the skill’s plugin cache. The agent knows where to find it, but if you need to run it manually:
429: too many concurrent searches
Deep search allows a maximum of 5 concurrent searches per org. If you hit a 429, wait for running searches to finish before starting new ones. Check the status of previous tasks withGET /search/status/{task_id}.
403 on high tier requests
Thehigh research tier requires enterprise access. If you get a 403, book a call at cal.com/team/sixtyfour/discovery. The low and medium tiers work on all plans.
Results are empty or incomplete
- Bump
tierfrom"low"to"medium". The default single-pass lookup misses people with limited online presence or common names. - For people intelligence, pass as many identifiers as possible in
lead_info:full_name,company,linkedin_url, andemailall improve matching.
Deep search appears stuck
Searches take 30–120+ seconds while the agent runs. Thepoll_job.py script prints progress messages like Analyzing your query... and Found X profiles as it goes. Don’t kill it early. If it hangs past 10 minutes, re-run with a fresh start-deep-search call.
Output file is CSV, not JSON
Deep search results download as CSV. Parse them withcsv.DictReader, not json.load. Columns include fullName, currentPosition, currentCompany, linkedinUrl, skills, locationName, and others.
Signed download URLs expire
After a search completes, the download URL expires in 15 minutes. If you wait too long, re-poll the status endpoint — it issues a freshresource_handle_id you can use to get a new signed URL.