jobsmith

JobSmith

JobSmith turns your AI coding agent into a structured job-search partner while keeping you in charge of the decisions that matter.

It helps you turn messy resumes, job boards, job descriptions, interview notes, and application history into a repeatable workflow: define the job you want, find a few strong matches, build a master resume, tailor application packages, prep for interviews, debrief what happened, and improve the next run. Unlike one-click AI resume builders, JobSmith is built around human decision points. The agent does the scanning, sorting, drafting, checking, and remembering, but you choose the target, the positioning, the proof, the boundaries, and the final edits.

It is designed for people using Claude Code, Codex, Claude Cowork-style skills, Cursor, or any agent that can read markdown files.

Why It Is Different

Most AI job tools optimize for speed: paste a resume, paste a job description, get a generated document. JobSmith optimizes for judgment. It gives the agent enough structure to do the heavy lifting, then pauses where your taste, truth, and strategy matter most.

The built-in human decision points are:

• Onboarding: You approve the source-of-truth career story before it becomes the master resume.
• Job discovery: You approve the dream-job target, hard filters, source access, and daily or twice-daily cadence. Every recommended job waits for your pursue, watch, or pass decision.
• Before employer-facing documents: The Minimal Grill Checkpoint asks only the missing questions that can change the outcome, such as which proof point should lead, whether there is a referral path, and what motivation is actually true.
• Personal ReadMe decision: JobSmith recommends whether this optional document helps or distracts, but you decide whether to include it.
• Expert panel checkpoint: The agent runs ATS, recruiter, hiring manager, peer, AI-detector, and personal-calibration reviews, then asks whether to apply all fixes, only critical fixes, or review changes first.
• Interview prep: A short grill captures the interview format, known agenda, concerns, and constraints before prep is generated.
• Interview debriefs and retrospectives: You bring the notes, transcript, or memory dump, then decide which lessons should update future applications.
• Privacy and release boundaries: Anything involving private career material, examples, or public sharing requires explicit review.

Start Here

This is the shortest path from "I have some old resumes" to "I know what to focus on and can apply with a tailored package." Put your existing materials in the private my-data/ folder, ask your agent to run onboarding, and JobSmith will extract what it can before asking a small number of high-value questions about your target and where to find it.

1. Clone or download this repo.

   git clone https://github.com/0miee/jobsmith.git
   cd jobsmith
   

2. Create your private workspace.

   scripts/init-jobsmith.sh
   

3. Drop 1-3 existing resumes, LinkedIn exports, bios, or old application materials into my-data/inbox/.

4. Open the repo in your agent and say:

   Onboard me into JobSmith. Follow workflows/00-onboarding/WORKFLOW.md.
   

5. Approve your target, saved searches or job-board sources, and a daily or twice-daily scan. Then say:

   Scan my approved job sources. Follow workflows/10-job-discovery/WORKFLOW.md.
   

6. When you choose a job to pursue, paste the job description and say:

   Help me apply to this job. Follow workflows/02-applications/WORKFLOW.md.
   

What It Creates

JobSmith keeps your private job-search data in my-data/, which is ignored by git. The public repo contains workflows and templates; your actual career history, application folders, interview notes, and learnings stay local unless you choose otherwise.

During onboarding, JobSmith creates:

• my-data/master-resume.md: the full private source-of-truth resume
• my-data/story-bank.md: reusable STAR stories for applications and interviews
• my-data/positioning.md: target roles, differentiators, boundaries, and proof points
• my-data/user-learnings.md: your preferences, feedback, patterns, and standing instructions
• my-data/job-search/search-profile.md: your approved target, sources, filters, and cadence

During job discovery, JobSmith creates:

• a deduplicated ledger of previously seen jobs and your pursue/watch/pass choices
• daily or twice-daily digests with no more than three strong new matches
• one number-one focus that considers jobs, follow-ups, interviews, and unfinished applications

During applications, JobSmith creates:

• job-description archive
• gap analysis and keyword map
• workflow progress file for resumability
• tailored resume
• tailored cover letter
• optional Personal ReadMe recommendation
• expert panel review
• submission checklist

Compatibility

JobSmith is repo-template-first. That means the markdown workflows are the source of truth, and platform-specific skills are only thin adapters that route agents to those workflows.

Tool	How to Use
Claude Code	Reads CLAUDE.md; optional skill adapter in .claude/skills/jobsmith/
Claude Cowork-style skills	Import or reference .claude/skills/jobsmith/SKILL.md
Codex	Reads AGENTS.md; optional adapter in .codex/skills/jobsmith/
Cursor / other agents	Ask the agent to read the relevant workflows/*/WORKFLOW.md file
Plain markdown	Follow the workflows manually

Project Structure

The repo separates public reusable instructions from private user data. Public files teach the workflow. Private files under my-data/ hold the user's real career materials and should not be committed.

jobsmith/
├── README.md
├── CLAUDE.md
├── AGENTS.md
├── CONTEXT.md
├── _shared/
│   ├── writing-rules.md
│   └── resume-cover-letter-best-practices.md
├── workflows/
│   ├── 00-onboarding/
│   ├── 01-master-resume/
│   ├── 02-applications/
│   ├── 03-networking/
│   ├── 04-interview-prep/
│   ├── 05-tracker/
│   ├── 06-interview-debrief/
│   ├── 07-retrospective/
│   ├── 08-rejection-analysis/
│   ├── 09-linkedin-optimization/
│   ├── 10-job-discovery/
│   └── 99-release-audit/
├── prompts/
├── templates/
├── examples/
├── .claude/skills/jobsmith/
├── .codex/skills/jobsmith/
└── my-data/              # private, gitignored

Find Fewer, Better Jobs

JobSmith begins discovery with a concrete example of the job you actually want. It uses that target to shape LinkedIn recommendations, saved searches, filters, and ranking, then scans only the sources you approve. A digest can contain up to three strong new matches, but it is allowed to return none.

During onboarding, the agent asks one compact question about where useful jobs appear for you now. You can provide saved job-board searches, company career pages, communities, or email alerts and choose a daily or twice-daily cadence. Authenticated pages and email alerts require explicit permission, and JobSmith never stores credentials or bypasses a site's controls.

When your agent supports recurring tasks, it can schedule the approved scan times. Otherwise, use native job-board alerts and run:

Scan my approved job sources. Follow workflows/10-job-discovery/WORKFLOW.md.

Say today, or /today where custom slash commands are supported, for one highest-value action across new matches, follow-ups, interviews, and unfinished applications. Say wrap or /wrap to record decisions and leave the next session ready.

The Human-Led Application Loop

The application workflow is designed to let the agent do most of the work while still involving you where your judgment matters. This is the core difference from a generic AI resume builder: JobSmith does not silently decide your story for you. It pauses before the choices that affect truth, tone, positioning, and submission strategy.

It has two mandatory application checkpoints: one before employer-facing documents are created, and one after the expert panel finds fixes.

The loop:

1. Save the job description.
2. Research the company.
3. Extract keywords and subtext.
4. Match the JD against your master resume.
5. Run a Minimal Grill Checkpoint so you choose the right proof, emphasis, motivation, referral strategy, and optional Personal ReadMe path.
6. Generate the tailored resume and cover letter.
7. Recommend whether to include the optional Personal ReadMe.
8. Scrub for truth, voice, AI tells, domain context, and consistency.
9. Run the 11+ expert panel.
10. Ask how you want to apply fixes: all, critical only, or review first.
11. Prepare the final submission checklist.

The Personal ReadMe

The Personal ReadMe is optional, but JobSmith makes the choice obvious because it can be a real differentiator. It explains how you think, collaborate, make decisions, and work with tools.

It is often useful for senior, product, engineering, design, leadership, consulting, startup, and remote roles. It may be worth skipping for rigid portals, highly traditional hiring processes, or applications where extra material would distract. When a portal only allows one upload, JobSmith can recommend a combined package.

Interview Prep and Debrief

JobSmith treats interviews as part of the same learning system, not a separate task. It creates company-specific prep, asks a short interview-prep grill, and captures what happened afterward so future applications and interviews improve.

Before an interview, say:

Prep me for my interview at [Company]. Follow workflows/04-interview-prep/WORKFLOW.md.

After an interview, bring the richest artifact you have:

• audio recording, only if lawful, allowed, and consented to where required
• transcript
• rough notes
• memory dump

Place it in my-data/interviews/inbox/ and say:

Debrief this interview. Follow workflows/06-interview-debrief/WORKFLOW.md.

Retrospectives

JobSmith should get better as you use it. Retrospectives look across applications, interviews, rejections, referrals, and expert-panel results to suggest concrete improvements.

Run a retrospective after 5 applications, after 2-3 interviews, monthly during an active search, or whenever you ask "what's working?"

Run a JobSmith retrospective. Follow workflows/07-retrospective/WORKFLOW.md.

The output should update my-data/user-learnings.md so future packages improve.

Rejections and LinkedIn

JobSmith treats rejection analysis and LinkedIn updates as part of the same improvement loop. Rejections become targeting lessons, and LinkedIn should stay aligned with the master resume as positioning changes.

When rejected, say:

Analyze this rejection. Follow workflows/08-rejection-analysis/WORKFLOW.md.

To update LinkedIn, export or paste your profile content, then say:

Optimize my LinkedIn. Follow workflows/09-linkedin-optimization/WORKFLOW.md.

Writing Philosophy

JobSmith is strict about human, specific, truthful writing. It never fabricates experience, and it avoids the polished language patterns that make AI-generated job materials easy to spot.

Core rules live in _shared/writing-rules.md:

• concrete evidence over vague claims
• no fabricated experience
• no AI-sounding filler
• honest gap bridging
• domain context when readers may not know your field
• cover letters that read like thoughtful emails

Examples

The public repo should include only synthetic or anonymized examples. Private job-search data belongs in my-data/.

Example types:

• synthetic starter example for a quick trial
• synthetic interview debrief and retrospective examples
• anonymized golden path later, only after manual privacy review

See examples/synthetic-happy-path/ for a complete synthetic demo with expected outputs.

For maintainers, docs/first-usable-demo.md defines the end-to-end demo checklist.

You can smoke-check the synthetic demo files with:

scripts/demo-check.sh

Release Safety

Before publishing or syncing to GitHub, run the release audit. This is a hard privacy gate.

Run the release audit. Follow workflows/99-release-audit/WORKFLOW.md.

The audit checks that no private my-data/ content, real contact details, private application folders, or unanonymized source material are included.

Before public release, use docs/github-deployment-checklist.md.

Publishing steps live in docs/github-publish.md.

Contributing

Improvements are welcome. The best contributions make the workflow clearer, safer, more truthful, or easier for job seekers to use.

Good contribution areas:

• better templates
• clearer workflow steps
• anonymized examples
• stronger release-audit checks
• agent adapter improvements that stay thin

Read CONTRIBUTING.md and PRIVACY.md before adding examples or release-related changes.

For security or privacy concerns, see SECURITY.md.

See ROADMAP.md for planned improvements and CHANGELOG.md for release notes.

License

MIT.