How to Customize Prompt Templates in Hiring-Agent: 3 Methods Explained

Yes, you can fully customize prompt templates in InterviewStreet's hiring-agent by editing the Jinja2 files in prompts/templates/, registering new templates in the TemplateManager class, or supplying a custom template directory at runtime.

The hiring-agent repository uses Jinja2 templating to construct LLM prompts for each resume section. According to the source code in prompts/template_manager.py, the TemplateManager class orchestrates template loading and rendering, allowing you to modify extraction behavior, JSON output schemas, and system instructions without changing the core extraction logic.

How Template Management Works

The TemplateManager Class

The orchestration lives in prompts/template_manager.py. The TemplateManager class is instantiated (for example, in pdf.py) with a default template directory of prompts/templates.

During construction, it calls _load_templates(), which builds a hard-coded dictionary mapping section names to Jinja file names:

# From prompts/template_manager.py - conceptual structure

"basics": "basics.jinja"
"work": "work.jinja"
"education": "education.jinja"

When processing a section, PDFHandler requests a rendered prompt via render_template(section_name, text_content=...), which returns the final prompt string ready for the LLM.

Default Template Locations

All default templates reside under prompts/templates/:

• basics.jinja – Extracts name, email, phone, location, and profiles
• work.jinja – Extracts job history and employment details
• education.jinja – Extracts academic background
• skills.jinja – Extracts technical and soft skills
• projects.jinja – Extracts project descriptions
• awards.jinja – Extracts awards and recognitions
• system_message.jinja – Wrapper template used for every section that receives the section_name_param variable

Three Ways to Customize Prompt Templates

Method 1: Edit Existing Jinja Files

The simplest approach is to modify the existing template files directly. Open any file under prompts/templates/ (e.g., basics.jinja) and change the wording, add extra instructions, or modify the JSON schema that the LLM must return.

The next time render_template() is called for that section, it will use your revised wording immediately.

Method 2: Add New Section Templates

To extract custom data points, create entirely new templates:

1. Create a new file certifications.jinja inside prompts/templates/

2. Add an entry to the template_files dictionary in TemplateManager._load_templates():

   "certifications": "certifications.jinja"

3. Call render_template("certifications", text_content=...) from your code, or extend PDFHandler with a new extractor method

This approach allows you to invoke custom extractions using the same infrastructure as built-in sections.

Method 3: Change the Template Directory

To keep templates separate from the source tree (e.g., for user-specific overrides), pass a different path when constructing TemplateManager:

from prompts.template_manager import TemplateManager

# Use custom template directory

tm = TemplateManager(template_dir="./my_custom_templates")

You can implement this by subclassing PDFHandler or modifying the instantiation in pdf.py.

Practical Implementation Examples

Rendering a Template Manually

from prompts.template_manager import TemplateManager

tm = TemplateManager()  # Uses default prompts/templates directory

prompt = tm.render_template(
    "basics", 
    text_content="John Doe\njohn@example.com\nSan Francisco, CA"
)
print(prompt)  # Final Jinja-rendered prompt ready for LLM

Adding a Runtime Template Without Source Modification

# Create my_templates/custom_section.jinja

"""
Extract certification data from the resume text.
{{ text_content }}
Return JSON with fields: "name", "issuer", "date".
"""

# Runtime usage

tm = TemplateManager(template_dir="./my_templates")

# Manually register the mapping

tm._templates["custom"] = tm.env.get_template("custom_section.jinja")
prompt = tm.render_template("custom", text_content=resume_text)

Extending PDFHandler for a New Section

# In pdf.py or your subclass

def extract_certifications_section(self, resume_text: str):
    prompt = self.template_manager.render_template(
        "certifications", text_content=resume_text
    )
    if not prompt:
        return None
    return self._call_llm_for_section(
        "certifications", 
        resume_text, 
        prompt, 
        CertificationsSection  # Your Pydantic model

    )

Summary

• Edit existing templates by modifying Jinja files in prompts/templates/ to change wording or output schemas for standard sections like basics, work, and education
• Add new templates by creating .jinja files and registering them in TemplateManager._load_templates() to support custom extraction sections
• Override template directories by passing template_dir to the TemplateManager constructor to maintain user-specific template collections outside the repository
• Access templates at runtime via render_template(section_name, text_content=...) which returns fully-rendered prompts ready for LLM submission

Frequently Asked Questions

Where are the default prompt templates stored in hiring-agent?

The default templates are located in the prompts/templates/ directory. This includes basics.jinja, work.jinja, education.jinja, skills.jinja, projects.jinja, awards.jinja, and system_message.jinja. The TemplateManager class in prompts/template_manager.py loads these files by default when instantiated without a custom template_dir parameter.

Can I add custom resume sections without modifying the source code?

Yes, you can add custom sections at runtime by placing new Jinja files in a custom directory and using TemplateManager(template_dir="your/path"). Alternatively, you can monkey-patch the TemplateManager._templates dictionary at runtime to register new templates without editing the source files in the repository.

What templating engine does hiring-agent use for prompts?

Hiring-agent uses Jinja2 for template rendering. The TemplateManager class maintains a Jinja environment (self.env) that loads templates from the specified directory and renders them with variables like text_content and section_name_param.

How do I change the system message for all resume sections?

Edit the system_message.jinja file in prompts/templates/. This template acts as a wrapper for every section and receives the section_name_param variable (as seen in line 79-80 of pdf.py). Modifying this file changes the overall tone, instruction set, or formatting requirements for all extractions simultaneously.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}