What is an AI job-search Agent?
See the complete workflow for profiling, job discovery, complete decisions, platform delivery, and human takeover.
Read the guide →A 36-dimension profile first decides what's worth applying to, and selected jobs are then applied to automatically; an AI voice interviewer practices with you until showtime, with a dual-track report that even measures how you say it; 8-stage tracking, offer comparison, and negotiation practice close it out. The whole job-search flow in one browser workbench.
Activates automatically the first time you use any commercial product within 7 days of signing up. Shared across all AgentMesh360 products — no card required, no automatic renewal. The 30 credits cover a resume profile plus your first custom question bank, or 2 platform Discovers.
Copy the command below and send it to your AI Agent to set up Job Agent and connect your AgentMesh360 subscription.
$ curl -fsSL https://raw.githubusercontent.com/jiyangnan/AgentMesh-JobAgent/main/scripts/install.sh | bash
Your AI Agent runs the installation: it sets up the Job Agent Skill and CLI and connects your subscription.
US$4 for 30 days, including 1000 credits shared across the AgentMesh360 ecosystem. No automatic renewal.
Open the workbench in your browser; once your subscription is verified, your complete job-search flow loads.
Installed and signed in — so what's inside the workbench? Three moves: decide what's worth applying to, apply automatically, and practice until you land it. One recruiter-side, 36-dimension profile runs through everything, and every stage gets a concrete tool, so judging, executing, practicing, following up, and deciding all beat going by feel.

| Stage | The problem you hit | The tool Job Agent gives you |
|---|---|---|
| Before applying | No idea what's worth applying to | 36-dimension profile × job decisions across four platforms, up to 100 deduplicated jobs judged per run, selected / review / rejected with evidence of fit and verifiable signatures |
| Before the interview | Nervous, no feedback, no idea where you're weak | 5-dimension custom question bank + live voice practice + dual-track score report |
| Following up | Applications forgotten, follow-ups missed | An 8-stage progress line per job; stalled items are grouped into one account status digest |
| After the offer | Choosing between offers by feel, no negotiation experience | Weighted four-dimension comparison + read-aloud negotiation scripts + practice with an AI HR |
| Throughout | Scattered tools, scattered accounts, unclear costs | One AgentMesh360 pass works across the ecosystem; credits are billed per action, with no automatic renewal |
Chrome on your device connects to Boss 直聘, 猎聘, 智联招聘, and 51Job, while the cloud's 36-dimension profile judges every job one by one and returns complete decisions in all three categories — selected jobs are applied to automatically in platform order, and you only step in for sign-ins and CAPTCHAs. It shares the same account and the same pool of credits as the web workbench.
Boss 直聘, 猎聘, 智联招聘, and 51Job each fetch, deduplicate, and judge on their own, so one platform acting up doesn't sink the whole round.
Starting a round authorizes applying to selected jobs in order; review is excluded by default and rejected is never applied to — the boundaries are spelled out before you start.
CAPTCHAs, sign-ins, and network drops all pick up from where they stopped; failures before the decision aren't charged, and failed cloud decisions are refunded.
✓ candidate profile 36 recruiter-side dimensions ready ✓ candidate jobs 86 normalized and deduplicated ✓ decision selected 18 review 11 rejected 57 ✓ personalized greetings 18 grounded in resume evidence ✓ official result signature verified
From pasting the install command to a verifiable AgentMesh360 cloud decision, then sequential execution on Boss 直聘, 猎聘, 智联招聘, and 51Job. A real workflow recording, about 3 minutes.
Before the real thing, spend 10 credits to generate a written question bank — 5 dimensions, 10+ questions, charged once per JD — and read it through before you go on mic. The AI interviewer talks with you in real time using neural voice: it has read your resume profile and target JD, usually opens with a project you've worked on, and follows up on your answers instead of moving straight to the next question.

Your resume profile + target JD generate a 5-dimension bank — self-introduction, project deep-dives, technical checks, behavioral questions, and your questions for them — each with follow-up directions. Not a generic interview-prep list.
Mention an architecture choice and it asks "why did you pick that at the time?" The gap between finishing a sentence and hearing the next one is about 1.3 seconds — close to the rhythm of a real conversation.
About 12–15 fully structured exchanges. A dual-track report is generated automatically when it ends; you can also generate the question bank on its own and review it in writing first.
Generated automatically when the interview ends. The content track scores every question and gives actionable advice; the delivery track is computed from word-level speech-recognition timestamps, not a model's subjective impression. Scores have explicit anchors.

It skips filler like "delivery could improve" — "you said 'you know' 6 times in 15 seconds on question 3" is feedback you can act on.
Filler-word frequency, pause distribution, net speaking rate, repeated restarts — measured values that come out identical if you recompute the same interview.
What each score means is spelled out in the report. Vague answers come with "a better answer" examples.
▎Content track (per-question scores) Technical depth 4/5 solid project detail Project experience 3/5 missing quantified results STAR structure 3/5 results section runs thin ▎Delivery track (measured) Pace 4.8 words/s △ a bit fast — aim for about 3.5 Fillers 8.2 /min ⚠ 6 "you know" in 15 seconds on Q3 Pauses longest 2.3 s (02:15) ▎Improvement suggestions Q2 ran 3 minutes in circles → lead with the conclusion, keep it under 90 seconds The two vague answers include "a better answer" examples
Every job gets its own progress line with 8 stages: Applied → Replied → Interviewing → Negotiating → Waiting for offer → Offer received (plus two terminal states, Declined and Closed). You tap once to update a status; the system records the time and computes how long it has been sitting.

Each job is still evaluated independently, but every reminder cycle sends one account digest with the full active-stage summary and only the jobs that need attention expanded. Each stage can trigger at most twice, with quiet hours you can set.
You get a "one last practice" nudge with a one-click jump straight into the mock interview.
Accept the offer and close all other open applications in bulk. Three replies on Boss 直聘 doesn't mean five on 猎聘 — every job is tracked independently.
With two or three offers on the table, Job Agent helps you run the numbers and find the right words — and guarantees it never invents any numbers.


Weighted comparison across four dimensions: total package, growth, commute, and stability. Recommendations come with explanations; you set the weights and the conclusion follows them. Guaranteed at the code level: only the numbers you provide are used, and anything missing is marked "not provided".
Generates a three-piece negotiation script — anchoring lines, pushback handling, and a concession ladder down to your floor — then spars with you in text as "an HR with a budget cap". It lowballs, it stalls, it counters with non-salary compensation.
Comparison results are stored for later review; one click moves the journey stage from "Offer received" to "Negotiating".
The cloud delivers a complete job-search decision chain, not a context-free model call.
Interprets your resume from a recruiter-screening perspective and builds 36 recruiting dimensions. Explicit facts remain separate from reasonable inferences, so search, matching, and communication all use the same candidate profile.
Normalizes candidate jobs across platforms and evaluates their meaning. Every deduplicated job is classified as selected, review, or rejected, with evidence of fit, risks, and reasons for the decision.
Combines job requirements with real evidence from your resume to create job-specific greetings for Boss 直聘 and 猎聘. The copy avoids generic templates and never invents candidate experience.
One platform Discover processes up to 100 deduplicated jobs. The cloud does not return just a handful of recommendations; it explains which jobs are worth applying to, which need review, and which should be excluded.
Every deduplicated job must receive one clear outcome with supporting reasons.
After a round starts, all four platforms run in the product-defined order without concurrent browser use.
Your host AI Agent reads the public repository and standard commands, then continues, pauses, or asks for your help based on the state returned by the CLI. You do not need to stitch together a multi-step job workflow by hand.
Register for AgentMesh360 with email, Google, or GitHub, generate an API Key in your account, and get the Monthly Pass that provides cloud credits.
Open your AgentMesh360 account →
The public repository provides standard onboarding documentation. Claude Code, OpenClaw, Cursor, or Codex reads the current commands and verifies the signed version policy after installation.
View Agent onboarding →
Job Agent uses a dedicated Chrome profile and runs in the order Boss 直聘, 猎聘, 智联招聘, and 51Job. If it detects a sign-out, a CAPTCHA, or another manual step, the Agent pauses immediately and asks you to handle it.
The cloud returns complete, signed decisions in all three categories. Starting a job-search round authorizes automatic applications to selected jobs; review is excluded by default and can be promoted only when the user explicitly chooses; rejected is never considered for application.
$ curl -fsSL https://raw.githubusercontent.com/jiyangnan/AgentMesh-JobAgent/main/scripts/install.sh | bash $ jobagent init --key <AgentMesh API Key> $ jobagent resume analyze --file resume.pdf $ jobagent round start
Job decisions and automatic applications, custom question banks, voice interviews with dual-track reports, offer comparison, negotiation practice — all drawing on one pool of credits, billed per action and only when it completes. Two typical ways to spend them:
Calculation: scenario one = initial resume profile 5 credits + 99 platform Discovers (10 credits each, up to 100 deduplicated jobs); scenario two = profile (5) + 5 custom question banks (50) + 4 voice mock interviews with dual-track reports (800) + offer comparison (10) + negotiation practice (10) = 875 credits, leaving 125 credits for about 12 more Discovers (1,200 job decisions). About 2,970 jobs at the 30% selected estimate enter the automatic application flow; this does not guarantee successful platform submission, replies, interviews, or offers. Credits can also be used with other AgentMesh360 cloud products.
| Action | credits | Details |
|---|---|---|
| Resume profile (36 dimensions) | 5 | Cloud resume analysis via CLI; manual profile creation in the workbench is free |
| Custom interview question bank (5 dimensions) | 10 | Customized to your profile + target JD, charged once per JD |
| Voice mock interview (dual-track report included) | 200 | Billing starts only once the microphone is ready; the report is included at no extra charge |
| Offer comparison | 10 | Explainable weighted scoring across four dimensions, using only the numbers you provide |
| Negotiation strategy + AI HR practice | 10 | Charged when the strategy script is generated; practice rounds are free |
| Platform Discover (CLI, up to 100 deduplicated jobs) | 10 | Shares the same pool of credits as the workbench |
One typical use of 1000 credits: 1 profile (5) + 3 question banks (30) + 3 interviews with reports (600) + 1 comparison (10) + 1 negotiation (10) ≈ 655, leaving 345 for 34 Discovers. These are usage examples, not a guarantee of interviews or offers.
The API Key is an account credential; the Monthly Pass provides usable cloud credits. One purchase is shared across AgentMesh360 cloud products and never renews automatically.
Activates immediately after purchase and remains valid for 30 days.
Charges apply when cloud operations complete, not according to the number of selected jobs.
Recruiting-platform login and browser actions happen on your device. The cloud receives the data needed to generate the candidate profile, search plan, and job decisions, then returns official results that the client can verify.
Your device handles platform identity and execution; the cloud handles consistent decisions.
A signed protocol binds both sides to the same Discover.Create an AgentMesh360 account, generate an API Key in your account, and get a Monthly Pass when needed. The API Key identifies your account; the pass provides the shared credits required for Job Agent cloud capabilities.
US$4 for 30 days, including 1000 credits shared across AgentMesh360 cloud products, with no automatic renewal.
Open your account →Found a bug or want a new feature? Open an Issue; replies stay in the Issue and are publicly indexable.
Open an Issue →Want to discuss a use case or collaboration? Email us directly; we will reply by email.
Learn where an AI job-search Agent fits and how resume-to-job matching should begin with evidence, risk, and authorization to apply.
See the complete workflow for profiling, job discovery, complete decisions, platform delivery, and human takeover.
Read the guide →Understand job matching through required qualifications, evidence from experience, career direction, risk flags, and actionable guidance.
Read the guide →From handing the install command to your Agent, through signing in to the platforms, judging 100 jobs one by one, and automatic selected applications — your first full round, hands-on.
Read the guide →Create your profile, generate a custom question bank, practice out loud, and read the dual-track report — one complete interview rehearsal for 200 credits.
Read the guide →install.ps1 (PowerShell) and install.sh (Bash / Git Bash / WSL). It works on Mac, Linux, and Windows.
selected jobs in the order Boss 直聘, 猎聘, 智联招聘, and 51Job. review is excluded by default and can be promoted only when the user explicitly chooses; rejected is never applied to. At login, CAPTCHA, or resume selection, the Agent pauses and asks you to intervene.