Skip to content

Templates

A template is a pre-built mission brief. It defines the objective, the strategy, and the expected output — combined with your mode and effort, it determines AIBA's entire behavior.

You pick a template at launch (Step 2). AIBA takes your extra context from (Step 4), and injects them into the template. The result is a detailed, structured prompt that becomes the agent's marching orders.

Three templates ship out of the box. Each built for a different class of work.

Available Templates

default

General-purpose. No pre-defined strategy. What you type in Step 4 is exactly what the agent receives.

Good for research, browsing, QA testing, open-ended exploration — anything where you know what you want and don't need a structured playbook.

Output: Whatever the agent produces. No fixed format.

Job discovery with direct contact extraction. Built for one purpose: find jobs that match your skills, then hunt down recruiter emails for each one.

What it does:

  1. Analyzes your USER_PROFILE and derives 6–10 role titles from your skills
  2. Searches job sites e.g. LinkedIn Jobs and Indeed for each title, exhaustively — every page, every listing

Warning

It is recommended to share your browser session in .playwright-mcp/cookies.json in order for sites like LinkedIn Jobs and Indeed to allow the agent without facing a sign-in page. Learn more about browser sessions and cookies (covered in How-To Guides).

  1. For each matching job, runs a contact-hunting sequence: inspects the posting, visits the company page, searches for recruiter emails across the corporate domain
  2. Discards jobs without verified contact info. Only confirmed matches make the report.

Inputs that matter: USER_PROFILE is required to be set in .env for job_search template to work at its full potential, else it will only have the context given in the extra context prompt (step 4).

Output: A markdown table of verified jobs with company, role, URL, contact email, contact name, and verification method. Sorted alphabetically by company.

osint

Deep open-source intelligence investigation. Maps a person or company's digital footprint across the open web and produces a structured dossier.

What it does:

  1. Surface mapping — discovers the target's presence across LinkedIn, Twitter, GitHub, Crunchbase, and news sources
  2. Deep-dive per platform — opens each profile, captures structured data, screenshots, and meta tags
  3. Cross-verification — checks findings against corporate registries, patent databases, academic publications, and news archives
  4. Dossier output — compiles everything into a structured intelligence report

Inputs that matter: Extra context (Step 4) is the investigation target. USER_PROFILE is included as supplementary notes.

Output: A full dossier with executive summary, digital footprint map, professional timeline, contact & attribution, network & affiliations, and raw data appendix.

Inputs at a Glance

Template Uses USER_PROFILE? Uses extra context?
default No Yes — passed verbatim as the prompt
job_search Yes — drives role derivation and job matching Yes — appended as additional instructions
osint Yes — included as supplementary notes Yes — becomes the investigation target

How to Choose

Pick default when:

  • You know exactly what you want and just need an agent to go do it
  • You're exploring, researching, or QA testing
  • You want a conversational REPL session

Pick job_search when:

  • You're job hunting and want verified contact emails, not just links
  • Your USER_PROFILE is filled in with your skills and experience
  • You want volume — the template pushes for 50+ listings

Pick osint when:

  • You're investigating a person, company, or organization
  • You need a structured, cross-verified dossier — not a summary
  • You want breadth — the template maps presence across multiple platforms

Building Your Own

Templates are Python dataclasses with a generate_prompt function. You can register your own in src/prompts/templates.py.