agentleFS
rahuljallirahuljalli/job-application-agent/CLAUDE.md

job-application-agent

CLAUDE.md0 starsChanged 3 months agoMITView on GitHub

The core pipeline is built and runs end to end: scrape → score → tailor → validate → save → log (scripts/, .claude/skills/tailor-resume/, .claude/hooks/resume_hook.py). prd.md remains the source of truth for scope and the done-contract per stage. Not yet built: the Modal scheduled deployment (modal_app.py) and the Google Docs formatting path. Candidate: the repo ships a fictional sample resume (resume/base_resume.md, "Alex Morgan") so the pipeline can be demonstrated without exposing a real CV. Replace it with your own before real…

  • Reads credentials
  • In this repositoryOne other file it gives its agents.

The file

We split this CLAUDE.md into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use these as the done-contract when implementing each stage:

    line 73
  2. Distinctive:

    Every scored job has a numeric score.

    line 75

≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project status

The core pipeline is built and runs end to end: scrape → score → tailor → validate → save → log (scripts/, .claude/skills/tailor-resume/, .claude/hooks/resume_hook.py). prd.md remains the source of truth for scope and the done-contract per stage. Not yet built: the Modal scheduled deployment (modal_app.py) and the Google Docs formatting path.

Candidate: the repo ships a fictional sample resume (resume/base_resume.md, "Alex Morgan") so the pipeline can be demonstrated without exposing a real CV. Replace it with your own before real use. The pipeline targets AI Engineer postings throughout (scrape niche, scoring, out-of-scope note all agree).

What this project is (per prd.md)

An unattended agent that, once daily:

  1. Scrapes AI Engineer job postings (Apify, LinkedIn actor), remote or hybrid, full-time.
  2. Scores each against the base resume and keeps only 8+/10 fits, extracting matched/missing requirements in the same scoring call.
  3. Tailors the resume per qualifying job via a Claude Code skill — reorder/reword only, never invent experience, employers, tools, or metrics.
  4. Validates the tailored resume via a hook before it's saved/logged (required sections present, no leftover placeholder text); failed validation flags the row instead of writing it.
  5. Runs a company-research script (plain Python + OpenRouter, not a subagent — no interactive session exists on the automated path).
  6. Logs one row per qualifying job to a Google Sheet (title, company, link, score, resume link, status, timestamp).
  7. Saves each tailored resume into a per-job folder {company}-{job-title-slug}/ — Desktop locally, Google Drive on Modal.
  8. Deploys as a Modal scheduled function (modal.Cron), with secrets injected via Modal secrets (Apify, OpenRouter, Google, Telegram) and never committed.

Explicitly out of scope: auto-submitting applications, niches other than AI Engineer, resume visual/formatting polish.

Architecture (planned)

Two parallel execution paths share the same pipeline stages but differ in how the "tailor" step reasons:

  • Interactive path — run inside Claude Code directly; the tailor-resume step is a skill (.claude/skills/tailor-resume/SKILL.md) invoked with live Claude reasoning. Output goes to the Desktop.
  • Automated/headless path — scripts/tailor_job.py and scripts/company_research.py call OpenRouter instead of live Claude, so the pipeline can run with no Anthropic API key and no interactive session (needed for the Modal cron job). Output goes to Google Drive.

Both paths funnel through the same validation logic (scripts/validate_resume.py, also wrapped as .claude/hooks/resume_hook.py for the interactive path) and the same Google Sheets writer (scripts/write_sheet.py).

Target layout from prd.md Section 5:

job-apply-agent/
  CLAUDE.md
  PRD.md
  README.md
  GWS_SETUP.md
  .env.example
  .claude/skills/tailor-resume/SKILL.md
  .claude/hooks/resume_hook.py
  resume/base_resume.md
  scripts/
    scrape_jobs.py         # Apify
    score_jobs.py           # scores + extracts matched/missing requirements
    tailor_job.py            # automated tailoring path (OpenRouter, for Modal)
    company_research.py
    write_sheet.py          # Google Sheets
    format_resume_doc.py     # markdown -> real Google Docs formatting
    validate_resume.py       # shared validation logic
  output/                    # gitignored — raw/scored/tailored job data
  modal_app.py                # scheduled entrypoint
  .env                        # gitignored — Apify key, OpenRouter key, Google creds, Telegram bot token

Hard rules from the PRD

  • Score cutoff is fixed at 8/10 — jobs below that never proceed past scoring. Log reject counts alongside pass counts as proof the filter works.
  • Resume tailoring is reorder/reword only. Fabricating experience, employers, tools, or metrics is a spec violation, not a style choice.
  • Validation happens before a write — on failure, flag instead of writing the Sheet row.
  • Failures in the scheduled Modal run must not fail silently: wrap in try/except, send a Telegram alert titled JOB-APPLY-AGENT — WHAT BROKE, then re-raise.
  • Secrets (Apify key, OpenRouter key, Google credentials, Telegram bot token) go through Modal secrets / .env (gitignored) — never committed.

Verification checklist (per prd.md Section 6)

Use these as the done-contract when implementing each stage:

  • Scrape returns N jobs with title/company/link/description.
  • Every scored job has a numeric score; sub-8 jobs are excluded from all downstream steps.
  • Tailored output has no placeholder text and passes the validation hook.
  • Sheet row count matches qualifying-job count; every row has a working job link and resume reference.
  • Desktop folder count matches qualifying-job count; each folder has exactly one resume file.
  • claude -p headless run produces the same output as the interactive run, restricted to allowed tools.
  • Modal: scheduled function deploys, manual trigger runs end to end, a forced failure produces the named Telegram alert.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.