How Development Mode Affects Caching and CSV Export in the InterviewStreet Hiring Agent

When DEVELOPMENT_MODE is set to True in config.py, the application caches resume and GitHub data to local JSON files and automatically appends evaluation results to resume_evaluations.csv; when set to False, it skips both caching and CSV generation to ensure stateless execution.

The InterviewStreet hiring-agent repository uses a single global boolean flag to toggle between iterative development convenience and production-grade statelessness. Understanding how this flag manipulates file I/O behavior is essential for anyone running the evaluation pipeline locally versus deploying it to production environments.

Configuration Source for Development Mode

The DEVELOPMENT_MODE flag lives in config.py at the repository root. This module-level constant acts as the central authority that downstream modules check before executing file-system operations.

# config.py

DEVELOPMENT_MODE = True  # Toggle this for production runs

All conditional logic in score.py references this flag to decide whether to persist intermediate artifacts or operate entirely in memory.

Resume and GitHub Caching Behavior

The caching mechanism prevents redundant PDF parsing and GitHub API calls across multiple runs. The implementation resides in the main function of score.py, with distinct blocks handling resume extraction (lines 26–45) and GitHub profile fetching (lines 70–95).

When Development Mode Is Enabled

If DEVELOPMENT_MODE is True and a cache file exists, the script bypasses expensive operations:

• Resume caching: After extracting text from a PDF, the resulting JSON is written to cache/resumecache_<basename>.json. On subsequent executions, the script checks for this file first and loads it directly instead of re-invoking the parser.
• GitHub caching: Fetched profile data is stored in cache/githubcache_<basename>.json. The script loads this cached version on future runs, eliminating redundant API requests.

# Conceptual flow from score.py main function (lines 26-45, 70-95)

if DEVELOPMENT_MODE and os.path.exists(cache_path):
    data = json.load(open(cache_path))
else:
    data = expensive_operation()
    if DEVELOPMENT_MODE:
        json.dump(data, open(cache_path, 'w'))

When Development Mode Is Disabled

Setting the flag to False causes the conditional guards if DEVELOPMENT_MODE and os.path.exists(...) to evaluate as falsy. Consequently:

• The PDF is parsed fresh on every execution.
• The GitHub API is queried every time.
• No files are written to the cache/ directory.

This ensures deterministic, stateless behavior suitable for production pipelines where reproducibility and data freshness take priority over speed.

CSV Export Control in Evaluation Pipeline

Beyond caching, the development mode flag governs whether the system generates a persistent audit trail. This logic appears in score.py lines 41–63 and delegates row formatting to transform.py.

Automatic CSV Generation

When DEVELOPMENT_MODE is active, the script:

1. Calls transform_evaluation_response from transform.py to convert the evaluation object into a dictionary representing one CSV row.
2. Appends that row to resume_evaluations.csv, creating the file and writing the header on the first run.

# From score.py (lines 41-63)

if DEVELOPMENT_MODE:
    csv_row = transform_evaluation_response(evaluation_data)
    with open('resume_evaluations.csv', 'a', newline='') as f:
        writer = csv.DictWriter(f, fieldnames=csv_row.keys())
        if f.tell() == 0:
            writer.writeheader()
        writer.writerow(csv_row)

Suppressing Output for Production

With DEVELOPMENT_MODE = False, the entire CSV-generation block is skipped. No file is created or updated, preventing disk pollution in production environments and keeping the execution strictly in-memory.

Practical Configuration Examples

Toggle the flag via direct modification or runtime monkey-patching depending on your deployment strategy.

Enable full debugging (default):

# Edit config.py to set DEVELOPMENT_MODE = True

python score.py path/to/resume.pdf

Disable for production:

# Option 1: Edit config.py

# DEVELOPMENT_MODE = False

# Option 2: Runtime override

import config
config.DEVELOPMENT_MODE = False
from score import main
main("path/to/resume.pdf")

Summary

• config.py houses the global DEVELOPMENT_MODE switch that dictates I/O behavior across the codebase.
• Caching: When enabled, JSON files are written to cache/ for resumes and GitHub data; when disabled, the main function in score.py bypasses these checks entirely.
• CSV Export: Enabled mode appends to resume_evaluations.csv via transform_evaluation_response in transform.py; disabled mode skips this block to ensure stateless execution.
• Use True for iterative local development to speed up debugging; use False for production to guarantee fresh data and no side effects.

Frequently Asked Questions

How do I completely disable caching in the hiring-agent?

Set DEVELOPMENT_MODE = False in config.py. This causes the existence checks in score.py (lines 26–45 and 70–95) to evaluate to False, forcing the script to parse the PDF and query the GitHub API on every run without writing cache files.

Where are the cache files stored and what naming convention do they use?

Cache files are stored in a cache/ directory relative to the execution path. Resume data is saved as resumecache_<basename>.json and GitHub data as githubcache_<basename>.json, where <basename> is derived from the input PDF filename.

Does disabling development mode affect the evaluation logic or just the output?

Disabling development mode affects only the persistence layer. It suppresses cache reads/writes and CSV generation, but it does not alter the core scoring algorithm or the evaluation criteria applied to the candidate data.

Can I generate CSV output without enabling the cache?

No. The current implementation in score.py guards both features behind the same if DEVELOPMENT_MODE: block. There is no separate flag to decouple CSV generation from caching behavior; both are toggled simultaneously by the global flag in config.py.

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