resume-architect

Generated source view for the actual executable personal/resume-architect skill. The durable routing article is Writing and Content Skills. Source: skills/personal/resume-architect/SKILL.md

Runtime Source

Bundled Resources

These files are part of the executable skill folder and must be preserved with the skill source.

Description

Make a resume un-rejectable for a specific role. A 5-stage pipeline — Profile, Diagnoser (ATS scan /100), Researcher (live-benchmark real JDs + profiles of people who got hired), Rewriter (XYZ bullets + full ruleset), Hiring Manager (mock interview). Use when the user says "tailor my resume", "review my resume", "make my resume un-rejectable", "ATS scan", "why am I getting rejected", "resume for [company/role]", "rewrite my resume bullets", "resume gap analysis", "score my resume", "mock interview from my resume", "beat ATS", or pastes a JD + resume. Owns resume craft; career-ops owns job-search orchestration.

Skill Source

---
name: resume-architect
description: Make a resume un-rejectable for a specific role. A 5-stage pipeline — Profile, Diagnoser (ATS scan /100), Researcher (live-benchmark real JDs + profiles of people who got hired), Rewriter (XYZ bullets + full ruleset), Hiring Manager (mock interview). Use when the user says "tailor my resume", "review my resume", "make my resume un-rejectable", "ATS scan", "why am I getting rejected", "resume for [company/role]", "rewrite my resume bullets", "resume gap analysis", "score my resume", "mock interview from my resume", "beat ATS", or pastes a JD + resume. Owns resume craft; career-ops owns job-search orchestration.
origin: personal
source_slug: resume-architect
---

# Resume Architect

The single owner of **resume craft**: turn a real resume into one that gets an
interview for a *specific* role. Synthesizes four sources — the Diagnoser/Recruiter/
Rewriter/Hiring-Manager persona strategy, the proficiently job-search skills
(work-history profile, strict accuracy, mandatory self-critique), an expert
tech-resume ruleset, and Kevin's own [[application-standout-protocol]].

**The reader scans for under 10 seconds.** Every decision serves one goal: get the
interview. Not document a full work history.

## Scope (MECE — stay in your lane)

| This skill owns | Hand off to |
|---|---|
| Diagnose a resume vs a JD (ATS + gaps) | — |
| Research what gets people hired at the target | `last30days`, `agent-reach` (tools) |
| Rewrite bullets, sections, ordering | calls `humanizer` + `kevin-voice` for polish |
| Mock-interview stress test | — |
| Build/maintain the work-history profile | — |
| Find roles, score portals, track apps, negotiate, outreach | **`career-ops`** (orchestration) |
| Cover letters | `career-ops` / proficiently `cover-letter` |

For Kevin specifically, the Rewriter output **must pass** [[application-standout-protocol]]
(11 rules + Pre-Submit Checklist) before it is presented. That page is the policy gate;
this skill is the engine.

## Inputs to gather first

Before anything, confirm you have (ask only for what's missing):

1. **Target JD or role** — the specific posting. *No JD = no tailoring.* If absent, say
   so explicitly and offer a generic strengthening pass, but push for a JD: it is the
   single biggest lever on interview rate.
2. **Current resume** — file, paste, or `raw/career/resume-kevin-liu.tex` for Kevin.
3. **Career level** — new grad / early / mid / senior / tech lead / eng manager. Drives
   section order and emphasis (see `references/ruleset.md`).
4. **Special context** — career change, break, bootcamp, visa/work-auth, remote-only.

## The Pipeline

Run in order. Each stage names the reference to load and produces an artifact the next
stage consumes. Stages 1–4 are the four personas; Stage 0 feeds them.

### Stage 0 — Profile (build once, reuse forever)

A resume is a lossy summary. Tailoring needs the full source. Conduct (or load) a deep
work-history profile that is *richer than the resume*: per role, capture company context,
mandate, team/scope, accomplishments **with real metrics**, technical depth, and
challenges. Interview guide + template: `references/personas.md` § Profile.

- Persist it. For Kevin: `wiki/career/career-profile.md` + `wiki/career/resume.md` already
  hold this; deepen them rather than starting fresh.
- **Learn from corrections.** Every time the user corrects a fact, write it back to the
  profile so the same mistake never recurs.

### Stage 1 — Diagnoser (ATS scan + gap analysis)

Act as the Applicant Tracking System and a 10-second human scan. Compare resume ↔ JD.
Output: **score /100**, missing keywords, weakly-demonstrated skills, sections that would
get auto-rejected, and the 5 first-glance items (years of experience, JD-named tech,
quantified impact, work-auth if relevant, standout credential). Rubric in
`references/personas.md` § Diagnoser.

### Stage 2 — Researcher + Recruiter (evidence-backed benchmark) ← the differentiator

Most tools tailor to one JD's text and *guess* what matters. This stage pulls **real
signal** for the target company + level, then benchmarks the resume against it. Full
method: `references/research-playbook.md`. In short:

- **Requirements** — pull 3–5 real JDs for the same role/level at the company (and close
  peers) to find the true must-haves, not one posting's wording.
- **People who got in** — sample public profiles of people currently in that role at that
  company: what they led with, prior companies, project types, seniority signals.
- **Field reports** — `levels.fyi`, Reddit/Blind/X "how I got into X" writeups for the bar.

Then act as a recruiter who screens 200/day: deliver the **top-20 skills/success signals**
the resume must hit and the **top-3 red flags** — each tagged with the evidence it came
from. Use `last30days` and `agent-reach` (per `wiki/meta/capability-routing-map.md`); never invent the
benchmark from memory.

### Stage 3 — Rewriter (achievement formatting)

Rewrite experience into achievement bullets. **Formula:** accomplished **[X]** as measured
by **[Y]** by doing **[Z]** → in practice, *Action verb + what you did + tool/how + measurable
result*. Apply the full ruleset (formatting non-negotiables, content rules, section order by
level, special cases) from `references/ruleset.md`.

Then run the **mandatory self-critique pass** (`references/personas.md` § Rewriter) — kill
AI-sounding writing line by line, then call `humanizer` (AI-tell removal) and, for Kevin's
self-positioning lines, `kevin-voice`. Plain, sharp, human. No em dashes.

### Stage 4 — Hiring Manager (mock interview)

Act as the decision-maker for the role. From the *rewritten* resume, ask the 5 hardest
questions the candidate will face, score answers /10, and suggest improvements. This both
preps the candidate and surfaces any bullet that can't survive a follow-up (a bullet you
can't defend is a bullet to cut or fix). Rubric in `references/personas.md` § Hiring Manager.

## Accuracy (non-negotiable — highest priority)

Resolve conflicts in this order: **accuracy > user corrections > workflow > writing quality
> format > tone.**

- **Only enhance and reframe what the user provides.** Never fabricate numbers, companies,
  titles, technologies, scope, or business model. Reorder, reword, mirror JD language, and
  surface buried wins — but every claim traces to a real fact from the resume or profile.
- When something is ambiguous, use conservative language or omit it. Understate over overstate.
- Gaps in the JD's requirements are found via *adjacent real experience*, never invented.

## Output

When rewriting or creating a resume, return:

1. **The full resume** — clean, copy-paste-ready markdown/plain text, following the
   candidate's existing section structure (don't invent sections).
2. **Diagnosis** — the /100 score, missing keywords, and what would have gotten rejected.
3. **Benchmark** — top-20 target signals + top-3 red flags, with their evidence.
4. **Changes made** — a short list of key edits and *why* (what was reordered/rewritten/added,
   and the 2–3 sentence narrative for why this candidate is the obvious choice).
5. **Mock interview** — the 5 questions (offer to run them interactively).
6. **Flags** — any bullet needing a real number or detail only the user can supply.

If no JD was given, remind the user that tailoring to a specific JD significantly improves
results. **For Kevin:** also run the [[application-standout-protocol]] Pre-Submit Checklist,
flag any AI-slop bullet (Rule 9), keep it one page, and present for voice review before any
submission — agents propose, Kevin rewrites in his own voice.

## Formatting non-negotiables (quick list; full set in `references/ruleset.md`)

PDF only · ≤2 pages (1 for new grad/career-changer) · reverse-chronological · one column ·
bullets not paragraphs · every bullet has ≥1 number · "June 2021 – July 2022" date format ·
clickable links, no raw URLs · bold only titles/companies/dates · no photo/personal data ·
no self-rated skill bars/stars · no "references available on request" · no "we" (write the
candidate's own contribution) · no internal acronyms · no em dashes.

## Modes

- **Kevin mode** (default in this repo): source = `raw/career/resume-kevin-liu.tex`; gate =
  [[application-standout-protocol]]; voice = `kevin-voice` + `humanizer`; targets in
  `wiki/career/career-profile.md`. Two variants (startup vs big-tech) per Standout Rule 5.
- **Generic mode**: works for anyone — gather inputs, run the same 5 stages, skip the
  Kevin-specific gate and sources.

## Test prompts

- "Here's my resume and the JD for a Stripe backend role — make it un-rejectable." → full pipeline
- "ATS scan my resume against this posting, score it." → Stage 1 only
- "What do people who got hired as PMs at Google actually put on their resume?" → Stage 2
- "Rewrite these 3 bullets with the XYZ formula." → Stage 3
- Near-miss (do NOT trigger): "find me open backend roles" → that's `career-ops`.

## Related

- **Skills:** `career-ops` (job-search orchestration), `kevin-voice` (positioning lines),
  `humanizer` (AI-tell removal), `last30days` + `agent-reach` (Stage 2 research).
- **Wiki:** [[application-standout-protocol]] (the gate), [[resume]], [[career-profile]].
- **References:** `references/ruleset.md`, `references/personas.md`, `references/research-playbook.md`.

---

Timeline

• 2026-07-15 | Generated a browseable source page from the actual executable skill file. Source: skills/personal/resume-architect/SKILL.md