Career Architect Harness

A single-user, multi-agent AI harness built for Claude Code — your personal career team. Nine specialist agents produce tailored resumes, cover letters, job search strategy, interview preparation, and an AI leverage plan for your current job.

This is a template repository. Click Use this template to create your own copy — make it private, because it will hold your personal career data.

---

Why (the problem this solves)

Generic resumes lose. Most candidates are filtered out by Applicant Tracking Systems (ATS) before a human ever reads their resume, and most people underestimate their own transferable skills. Doing this well requires several distinct disciplines — skill excavation, keyword intelligence, ATS-compliant writing, storytelling, quality review, search strategy, and interview coaching — that rarely live in one person.

This harness encodes each discipline as a dedicated agent with its own playbook, and wires them into a gated workflow so that:

• Every output is tailored to you and a specific job description — never generic
• Every claim is evidence-backed — the team philosophy is "User First, Evidence Always"; fabrication is a hard stop
• Nothing skips the line — a resume can't be built before the skill inventory exists; nothing ships without QA approval

And it works both directions: the job-search pipeline optimizes your next role, while /amplify optimizes your current one — mapping your job to AI leverage opportunities whose measured wins feed back into your resume.

Who (the team)

Nine specialist agents, each with a defined role, model tier, and an exit state it must reach before the next agent proceeds.

Agent	Role	Model	Exit State
Intake Consultant	Initial consultation, profile creation	sonnet	Profile Complete
Skills Analyst	Skill inventory, latent & transferable skill mapping	sonnet	Skill Inventory Complete
Keyword Researcher	Job description analysis, 4-tier keyword extraction	sonnet	Keywords Extracted
Resume Architect	Resume construction, ATS optimization	opus	Draft Ready for Review
Narrative Crafter	Professional summary, cover letter, LinkedIn About	opus	Narrative Complete
QA Reviewer	Final quality gate before delivery	opus	Approved for Delivery
Career Strategist	Job search strategy, networking, salary guidance	opus	Strategy Delivered
Interview Strategist	Interview prep, mock interviews, STAR coaching	opus	Interview Preparation Complete
Amplify Strategist	AI leverage mapping for your current role	opus	Amplify Plan Delivered

Every agent names its model explicitly in frontmatter — heavy-judgment roles (writing, QA, strategy, coaching) run on Opus; conversational/extraction roles run fast on Sonnet. The QA Reviewer runs on the same tier as the Resume Architect so the final gate is never weaker than the builder. The Career, Interview, and Amplify Strategists carry WebSearch/WebFetch for live research.

flowchart LR
    subgraph Discovery
        IC[Intake Consultant]
        SA[Skills Analyst]
    end
    subgraph Intelligence
        KR[Keyword Researcher]
    end
    subgraph Production
        RA[Resume Architect]
        NC[Narrative Crafter]
        QA[QA Reviewer]
    end
    subgraph Strategy
        CS[Career Strategist]
        IS[Interview Strategist]
    end
    subgraph "Current Role"
        AS[Amplify Strategist]
    end
    IC --> SA --> KR --> RA --> NC --> QA
    QA -->|revisions| RA
    QA -->|revisions| NC
    QA -->|approved| CS --> IS
    IC -.-> AS
    AS -.->|measured wins| SA

Loading

What (the pieces)

The harness is a collection of Claude Code primitives wired together:

• Agents (.claude/agents/) — specialist sub-agent definitions with role playbooks, output templates, and handoff rules
• Commands (.claude/commands/) — slash commands that run pre-flight prerequisite checks, then invoke the right agent
• Skills (.claude/skills/) — domain expertise modules (ATS rules, bullet formulas, negotiation scripts, AI leverage patterns) that agents load as needed; not invoked directly by users
• Team config (.claude/team-config.json) — the roster, workflow order, and stop-the-line conditions in machine-readable form

Command	What It Does
/consult	Start here — intake interview, saves your profile
/skill-inventory	Full skill inventory: hard, soft, latent, and transferable skills
/analyze-job	Analyze a job description — 4-tier ATS keywords, gap analysis
/build-resume [job-slug]	Tailored resume + cover letter + LinkedIn About, QA-gated
/review-resume	Critical review of your existing resume with example rewrites
/career-strategy	Job search channels, target companies, networking, 30-day plan
/interview-prep [job-slug]	Prep plan, STAR story bank, scored mock interviews, research brief
/amplify	Map your current job to AI leverage — quick wins, quarter plays, measured impact

Skill	Used By	Purpose
ats-optimization	Resume Architect, QA Reviewer	ATS formatting rules, keyword placement, anti-patterns
resume-patterns	Resume Architect, Narrative Crafter, QA Reviewer	Bullet formulas, summary patterns, section templates, length rules
skill-translation	Skills Analyst	Latent/transferable skill identification frameworks
career-strategy-patterns	Career Strategist, Interview Strategist	Search tactics, outreach templates, negotiation scripts
interview-frameworks	Interview Strategist, Career Strategist	Question banks, STAR story bank template, research protocol
linkedin-optimization	Narrative Crafter, Career Strategist	Recruiter search ranking, profile optimization
amplify-patterns	Amplify Strategist	Task taxonomy, automate/augment rubric, measurement, governance

Each fact lives in exactly one skill — agents reference the skill files rather than embedding copies, so updates land in one place and never drift.

How (the workflow)

Each slash command runs pre-flight checks against the files earlier steps produced. If a prerequisite is missing, the command halts and routes you to the correct prior step — this is the gating mechanism. /amplify is the exception: it runs standalone at any time, though it reads your profile when it exists.

Each analyzed job gets its own job-targets/{slug}/ directory (slug = {company}-{title}, kebab-case). /analyze-job creates it; every downstream artifact for that job lands inside it.

flowchart TD
    A["/consult"] -->|"profile/profile.md"| B["/skill-inventory"]
    B -->|"profile/skills.md"| C["/analyze-job"]
    JD[Job posting] --> C

    subgraph JT["job-targets/{slug}/  (one directory per job)"]
        K["job-description.txt<br/>keywords.md"]
        O["resume.md<br/>cover-letter.md<br/>qa-report.md"]
        P["interview-prep.md"]
    end

    C --> K
    K --> D["/build-resume"]
    D --> O
    D -->|"profile/linkedin-about.md"| E["/career-strategy"]
    E -->|"profile/job-search-strategy.md"| F["/interview-prep"]
    K -.-> F
    F --> P
    R["/review-resume"] -.->|existing resume feedback| D

    A -.-> M["/amplify"]
    M -->|"profile/amplify-plan.md"| W[Measured AI wins]
    W -.->|new skills + resume bullets| B

    D -. missing profile .-> A
    D -. missing skills .-> B
    D -. missing keywords .-> C

Loading

/build-resume itself is a three-phase pipeline with a QA revision loop:

sequenceDiagram
    participant U as You
    participant RA as Resume Architect
    participant NC as Narrative Crafter
    participant QA as QA Reviewer
    U->>RA: /build-resume job-slug
    RA->>RA: Load profile + skills + job-targets/{slug}/keywords.md
    RA->>NC: Draft Ready for Review (job-targets/{slug}/resume.md)
    NC->>NC: Polish summary, write cover-letter.md + profile/linkedin-about.md
    NC->>QA: Narrative Complete
    QA->>QA: ATS, keyword coverage, ethics, consistency checks
    alt Revision required
        QA->>RA: Route back with job-targets/{slug}/qa-report.md
        RA->>QA: Revised draft
    end
    QA->>U: Approved for Delivery

Loading

Every file the harness produces:

File	Contents
job-targets/{slug}/cover-letter.md	Targeted cover letter
job-targets/{slug}/interview-prep.md	Prep plan, STAR bank, mock scores, action items
job-targets/{slug}/job-description.txt	The raw job posting
job-targets/{slug}/keywords.md	Tier 1–4 keywords, gap analysis, title recommendation
job-targets/{slug}/qa-report.md	QA review with APPROVED/REVISION status
job-targets/{slug}/resume.md	ATS-optimized tailored resume
profile/amplify-plan.md	Task inventory, AI plays, measurement log, guardrails
profile/job-search-strategy.md	Channels, company tiers, 30-day plan
profile/linkedin-about.md	LinkedIn About section (one profile, updated per target role)
profile/profile.md	Your career goals, background, target roles
profile/skills.md	Full skill inventory with gap analysis

When (which command, which moment)

Situation	Start Here
Just created your copy of this template	/consult — everything begins with a profile
You have a resume and want honest feedback	/review-resume — critique first, rebuild after
Found a job posting to target	/analyze-job — keywords before any writing
Ready to apply to a specific role	/build-resume — requires profile + skills + keywords
Application is out, search feels scattered	/career-strategy — channels, targets, weekly plan
Interview scheduled	/interview-prep — works with ≤3 days (triage), 1 week, or 2+ weeks
Happy in your job, want to perform better	/amplify — AI leverage plan for your current role
30 days after starting your Amplify Plan	/amplify again — measure wins, expand plays

Repeat /analyze-job + /build-resume for each target role — your profile and skill inventory are reused.

Getting Started

1. Click Use this template on GitHub → create a private repository (it will hold your personal career data)
2. Clone your copy and open it in Claude Code
3. Run /consult and follow the intake interview
4. Follow the recommended next step each command gives you

/consult → /skill-inventory → /analyze-job → /build-resume → /career-strategy → /interview-prep
                └──────────────── /amplify (current role, any time) ────────────────┘

Stop-the-Line Rules

Any agent will halt work and route back if:

• Career goals are undefined — return to /consult
• No target job description — required before /analyze-job can run
• Skill inventory incomplete — /skill-inventory must complete before /build-resume
• Fabricated credentials detected — work stops immediately
• An Amplify play would put confidential data into unsanctioned AI tools — halted and redesigned

Repository Structure

.claude/
├── agents/               # Specialist agent definitions
├── commands/             # Slash command workflows
├── skills/               # Domain expertise modules
├── team-config.json      # Agent roster and workflow config
profile/                  # Your profile, skills, strategies, amplify plan
job-targets/              # One {slug}/ directory per job: description, keywords,
                          # resume, cover letter, QA report, interview prep
patterns_library/         # Reusable resume patterns and templates
CLAUDE.md                 # Team philosophy, workflow rules, stop-the-line conditions

Ethics

• No fabricated accomplishments, credentials, employers, or dates
• Your career data stays in your private repository
• Embellished content is flagged, not silently accepted
• Every recommendation is grounded in your actual experience and the job's real requirements
• Amplify plays respect your company's AI policy and never expose confidential data