Resume Tailor

---
name: resume-tailor
description: Maintain a master resume and generate tailored 1-page application packages from job descriptions. Domain-siloed — job-search specific, local-only (master resume never leaves the machine). Use for /tailor, /add-to-master, /merge-resume, /parse-transcript.
tags: [intent:resume, intent:tailor, intent:document, type:skill]
---

# Resume Tailor Skill

**Triggers:** `/tailor`, `/add-to-master`, `/merge-resume`, `/parse-transcript`  
**Purpose:** Maintain a master resume and generate tailored 1-page application packages from job descriptions.

---

## Constants

- **Master resume:** `~/.claude/resume/master.md`
- **Output base:** `~/.claude/resume/tailored/`
- **Output pattern:** `~/.claude/resume/tailored/[company-slug]-[YYYY-MM-DD]/`
- **Scripts:** `~/.agents/skills/resume-tailor/scripts/`
- **Python:** `~/.agents/venv/bin/python`

---

## Command Routing

| Invocation | Handler |
|---|---|
| `/tailor [url] [--cover-letter]` | [TAILOR] |
| `/add-to-master "..."` | [ADD-TO-MASTER] |
| `/merge-resume [path]` | [MERGE-RESUME] |
| `/parse-transcript [path]` | [PARSE-TRANSCRIPT] |

---

## [TAILOR] — Full Application Package

### Phase 1: Intake

1. If a URL was provided: run `~/.agents/venv/bin/python ~/.agents/skills/resume-tailor/scripts/fetch_jd.py <url>`
   - If it succeeds: use stdout as the JD text
   - If it fails (non-zero exit): say "Couldn't fetch that URL — please paste the job description directly."
2. If no URL: ask "Please paste the job description."
3. If `--cover-letter` was NOT in the invocation: ask "Would you like a cover letter with this application? (yes/no)"

### Phase 2: Research

Search for:
- **Company:** mission, values, product, recent news (last 12 months), tech stack, team size/stage, culture signals (startup vs. enterprise)
- **Role:** seniority signals, team context, reporting structure if visible, tech signals
- **Contact:** hiring manager name, recruiter, team lead — check company site, LinkedIn, job post

Determine output folder name: slugify company name (lowercase, hyphens, no special chars) + today's date.  
Write `~/.claude/resume/tailored/[slug]-[date]/research.md` with all findings.

### Phase 3: Tailoring

Read `~/.claude/resume/master.md` in full.

**Suppressed entries:** an entry whose heading is followed by `<!-- suppress: ... -->` is kept in the master for the record only. Never include it, mention it, or draw facts from it in the resume, cover letter or summary, however relevant it looks. (Added 2026-10-07: Snorkel AI and Nourished Mobile App.)

**Extract from JD:**
- Job title → use verbatim as the header title line (all-caps). If ambiguous, infer the best match.
- Semantic concepts: domain areas, responsibility types, tech clusters, cultural signals
- Aerospace/defense signals: check for keywords — systems, engineering, simulation, physics, aerospace, hardware, robotics, defence, aviation, space, dynamics, structural, fluid, orbital, DoD, government, clearance

**Score every master resume entry:**
```
combined = (relevance × 0.7) + (recency × 0.3)

recency: 1.0 = current | 0.85 = <1yr | 0.65 = 1–3yr | 0.40 = 3–6yr | 0.20 = >6yr
relevance: 1.0 = direct match | 0.6 = transferable | 0.3 = adjacent
```
**Revised 2026-07-02 — no hard item caps.** Use the score to rank and curate, not to cut at a fixed count. Required sections (always include): Summary, Professional Experience, Projects, Education. Optional sections (include only if genuinely strong/JD-relevant; cut before padding): Technical Skills, Soft Skills, Relevant Training, Achievements. Typical good curation lands around 4 roles and however many projects are genuinely strong — that's a description of what good output looks like, not a count to hit. `render_pdf.py` now auto-shrinks to fit 1 page (Phase 5), so content length is no longer what keeps the resume on one page — curation quality is. Ties broken by recency.

**Preserve exact quantities when condensing master.md bullets.** Known failure mode from 2026-07-02: "14 charts, 2 tabs" (of one dashboard) got compressed into "14 dashboards" while tailoring — same number, false claim. Before writing a shortened bullet, check every number/unit in it against the master.md source sentence. If a number can't survive compression without becoming misleading, cut the number rather than let it drift.

**Aerospace crossover signal:** If JD has any aerospace/defense/systems keyword → surface the aerospace background in:
- The summary (explicitly mention dual-discipline background)
- The Engineering & Domain Knowledge skills category
- The cover letter unique angle paragraph

**Security clearance signal:** If JD is for a government/defense/cleared role → surface "Security Clearance: Secret (Reinstatable)" prominently in the header contact block.

**Mirror company language:** Use their terminology, not generic buzzwords.

### Phase 4: Build Resume

Write `~/.claude/resume/tailored/[slug]-[date]/resume.md` using this exact template structure.

**CRITICAL FORMATTING RULE:** Every paragraph element (bullets, project titles, achievement lines) MUST be separated by a blank line. Items within the same markdown block merge into a single visual line. One blank line = one visual paragraph break.

```markdown
# GIL ESKAYO

[JD-DERIVED TITLE — ALL CAPS]

gileskayo.me  |  geskayo@gmail.com  |  linkedin.com/in/gileskayo

(650) 867-2527  |  github.com/G-Eskayo  |  U.S. Citizen[  |  Security Clearance: Secret (Reinstatable) — if defense/gov role]

[SUMMARY — 2–3 sentences. Tailored to the specific role. No "I am a..." opener. Lead with what you build, then the crossover if relevant, then the current work.]

## PROFESSIONAL EXPERIENCE

**[Company Name]** *[Title, Dates]*

[ONE synthesized overview paragraph, 2–4 sentences: role scope, tasks, and value delivered — action verbs, quantified impact where available, mirrors JD language. Do NOT split this into multiple bullets that each repeat the bold company name — that's the Projects section's shape, not this one. Combine what master.md lists as separate bullets for this role into one flowing paragraph.]

**[Company Name]** *[Title, Dates]*

[overview paragraph for the next role]

[Required section. Curate to the strongest ~4 roles — not a hard cap, a description of good curation. One entry per company/role, not per achievement. If a company genuinely has two distinct stints (e.g. promoted, or left and came back), those are two entries; multiple bullets for one continuous role are not.]

## TECHNICAL SKILLS

• **[Category] –** [skill], [skill], [skill], [skill — max 4 items per category]

• **[Category] –** [skill], [skill], [skill], [skill]

• **[Category] –** [skill], [skill], [skill], [skill]

• **[Category] –** [skill], [skill], [skill], [skill]

[Optional section — include only if genuinely strong/JD-relevant. Typically ~3–4 categories. Include Cybersecurity if JD is security-relevant. Include Engineering & Domain Knowledge if aerospace signal detected.]

## SOFT SKILLS

[skill]  ·  [skill]  ·  [skill]  ·  [skill]  ·  [skill]  ·  [skill — up to ~6, JD-matched, on one line]

## PROJECTS

**[Project Name]** *([tech stack] — [rough date, e.g. "2026" or "Q2 2026"])*

• [bullet 1 — what was built, impact, scale]

• [bullet 2 — optional, max 2 per project]

**[Project Name]** *([tech stack] — [rough date])*

• [bullet 1]

[Required section, and the resume's highest-visual-priority one (render_pdf.py gives project names a larger/bolder treatment than other bold text — see design.md §7.1). Include the strongest, JD-relevant projects — no fixed count, curated by strength. Always include a rough date in the italic parenthetical alongside tech stack — signals recency; year-level precision is enough, don't make it look like a formal dated entry.]

## EDUCATION

**B.S. Computer Science** – University of Colorado Boulder  |  August 2025

## RELEVANT TRAINING

• **[Cert name]** – [Issuer] – [one-line description of what was built/learned]

• **[Cert name]** – [Issuer] – [description]

• **[Cert name]** – [Issuer] – [description]

[Optional section — include only if present in master and genuinely relevant. Typically up to ~3, most relevant to JD. Each cert on its own blank-line-separated paragraph.]

## ACHIEVEMENTS

• **U.S. Armed Forces Veteran** – Chemical, Biological, Radiological & Nuclear (CBRN) Defense Specialist[, Security Clearance: Secret (Reinstatable) — append if defense/gov role]

• **Eagle Scout** – Highest rank in Boy Scouts; 31 merit badges and two Eagle Palms

[Include both for leadership/character-signal roles. Include just Veteran for defense/gov roles if space is tight. Bulleted + bold lead-in, same pattern as Technical Skills/Relevant Training — see the "Formatting system" note below Slot caps.]
```

**Revised 2026-07-02 — required sections and guidance, not hard caps:**
- Required, always present: Summary (2–3 sentences), Professional Experience, Projects, Education (1 line only — this one stays a hard format rule, not a content-length one)
- Optional, include only if genuinely strong: Technical Skills, Soft Skills, Relevant Training, Achievements
- Typical well-curated shape: ~3–4 skill categories × ~4 items, ~6 soft skills, ~4 roles (one synthesized overview each), however many projects are genuinely strong × ~2 bullets, ~3 certs — guidance for what good curation looks like, not limits to enforce
- 1-page fit is now the renderer's job (Phase 5 auto-shrink-to-fit), not content's job — see architecture.md ADR-006

**Formatting system, added 2026-07-02 — apply to every section, no exceptions.** Sections were drifting into random per-section formatting (some bulleted, some stacked plain lines; some bold labels, some not) with no underlying logic — direct user feedback. Every section is one of exactly two patterns:
- **Pattern A — itemized list** (2+ discrete parallel entries: skill categories, certs, achievements): `•` bullet, **bold the entry's label** (category name / cert name / achievement title), then the detail. Applies to: Technical Skills, Relevant Training, Achievements, and Projects' sub-bullets sit under this pattern too (though the project name bolding lives in the entry-title line above them, not the bullets themselves).
- **Pattern B — flowing content** (one paragraph or one line, not a list of parallel items): no bullet. Applies to: Summary, Experience's overview paragraph (though its company/title header line above it is its own bold+italic element — see Phase 4 Experience format), Soft Skills (a single comma/middot-separated descriptor line — intentionally not itemized, these are single terms not label:detail pairs), Education (bold the degree as the "label," rest of the line is detail, same convention as Pattern A's bold-then-detail shape, just without a bullet since it's one line).
When in doubt for a new/unusual master.md entry, ask: is this one of several parallel items, or one flowing statement? That answers which pattern it uses — don't invent a third shape.

### Phase 5: Render PDF

Run:
```
~/.agents/venv/bin/python ~/.agents/skills/resume-tailor/scripts/render_pdf.py \
  ~/.claude/resume/tailored/[slug]-[date]/resume.md \
  ~/.claude/resume/tailored/[slug]-[date]/resume.pdf
```

**Read stderr, not just stdout.** The renderer's auto-fit (architecture.md ADR-006) prints one of three things worth surfacing to Gil, not silently swallowing:
- `WARNING: content still spans N pages...` — content overflows even at the smallest legible size. Trim content.
- `NOTE: page is only ~X% full...` — content is sparse but plausibly fixable. Suggest the next-strongest Project from master.md or expanding an existing entry.
- `FLAG: page is only ~X% full even at the 13pt ceiling...` — **added 2026-07-02, per explicit user direction.** Content was never close to overflowing at any size in the ladder. Surface this directly and honestly: this isn't a formatting gap, it's a signal that master.md may not have enough genuinely JD-relevant content for this role. Say so plainly and let Gil make the call on whether to proceed — don't pad the resume to hide it, and don't silently proceed as if the sparse output were normal.

### Phase 6: Cover Letter (if requested)

**Mandatory pre-step, added 2026-07-02 — do this before drafting Paragraph 2.** Do NOT synthesize the "strongest match" paragraph from research.md/master.md facts alone — restating resume bullets in prose is what makes a letter read as generic/AI-written (confirmed both by direct user feedback on a real draft, and by Ask a Manager's core finding that a good letter tells the reader something the resume can't). Pick the single most relevant experience, then explicitly ask the user for the real story behind it: what was hard, surprising, or required a judgment call — not just the outcome. Draft Paragraph 2 from their actual words. Also ask, separately, why this company/role specifically — don't infer or invent motivation; if the honest reason is sensitive or shouldn't appear verbatim in the letter (e.g. career-strategy calculus), ask what can be said instead rather than fabricating generic enthusiasm.

Also, before finalizing, re-read the drafted opener against the Paragraph 1 rule below — "I'm writing to apply for..." and close variants (e.g. "I am excited to apply...") are easy to slip into even when the rule is known; this happened in the same 2026-07-02 session that produced the rule.

Write `~/.claude/resume/tailored/[slug]-[date]/cover-letter.md`. **Use blank-line separation between the name/contact/date lines** (same rule as the resume header, SKILL.md Phase 4) — without it, markdown collapses consecutive lines into one paragraph with soft line breaks, which most renderers (including this one) flatten into a single run-on line instead of three. This exact bug shipped once already (2026-07-02) before being caught.

```
# [Name]

[email]  |  [phone]

[Date]

[Hiring Manager Name], [Title] at [Company]   ← use researched name if found
[else: Dear [Company] Hiring Team,]

PARAGRAPH 1 — HOOK (3–4 sentences)
  Reference something specific and real about the company — a product, stated mission, recent
  announcement, or known challenge. Never open with "I am writing to apply for..."
  State the role and why this company specifically.

PARAGRAPH 2 — STRONGEST MATCH (4–5 sentences)
  The single most relevant experience mapped to the JD.
  Concrete: what was built, what was the impact, what was the scale.
  Use the JD's own language where it fits naturally.

PARAGRAPH 3 — UNIQUE ANGLE (3–4 sentences)
  What Gil brings that a typical candidate doesn't.
  Lead with aerospace + CS crossover if JD has any engineering/systems angle.
  Otherwise: MARVIN (self-directed agentic systems project), cross-disciplinary builder mentality,
  shipped public work.
  If clearance-relevant: mention Secret clearance reinstatable status.

PARAGRAPH 4 — CLOSE (2–3 sentences)
  Confident, not desperate. Genuine enthusiasm for the conversation.
  Clear call to action. Match company energy (direct for startups, measured for enterprise).
```

Then render:
```
~/.agents/venv/bin/python ~/.agents/skills/resume-tailor/scripts/render_pdf.py \
  ~/.claude/resume/tailored/[slug]-[date]/cover-letter.md \
  ~/.claude/resume/tailored/[slug]-[date]/cover-letter.pdf \
  --cover-letter
```

### Phase 6b: Mirror & Open

Gil works in a terminal-based Claude Code session — there is no GUI panel for SendUserFile to render into, and `~/.claude/resume/tailored/` is a hidden dot-folder he can't browse to in Finder. So every render in Phase 5/6 must also be mirrored somewhere visible and opened directly:

1. Copy the rendered PDF(s) to `~/Documents/Career/Resumes/Applications/[Company Name] - [Job Title] - [YYYY-MM-DD].pdf` (and `... cover-letter.pdf` if generated) — `mkdir -p` the folder if needed.
2. Run `open` on each copied PDF so it appears on screen without Gil having to navigate anywhere.

Do this after every resume/cover-letter render, not just on request.

### Phase 7: Package Summary

Print:
```
✓ Application package ready: ~/.claude/resume/tailored/[slug]-[date]/
  Files:
    resume.md / resume.pdf
    [cover-letter.md / cover-letter.pdf — if generated]
    research.md

  Key matches: [top 3 JD concepts → master resume entries that scored highest]
  Title used: [derived title]
  Contact found: [name + title if found, else "not found — used generic salutation"]
  Aerospace signal: [yes/no — surfaced or not]
  Clearance surfaced: [yes/no]
```

---

## [ADD-TO-MASTER] — Add New Entry

Input: natural language description of a new experience, project, skill, cert, or achievement.

1. Identify which section the entry belongs in (Experience / Projects / Skills / Relevant Training / Achievements)
2. Format it following the conventions in that section
3. For Experience: convert to action-verb bullets, extract implicit dates if mentioned
4. Insert at the correct chronological position (most recent first)
5. Write the updated master.md
6. Confirm: "Added to [Section]. Review: ~/.claude/resume/master.md"

---

## [MERGE-RESUME] — Merge Uploaded Resume

Input: file path to .pdf, .docx, .md, or .txt resume

1. Run `~/.agents/venv/bin/python ~/.agents/skills/resume-tailor/scripts/extract_text.py <path>`
2. Parse the extracted text into sections matching master.md structure
3. For each item:
   - **No match in master:** add to master in correct section and position
   - **Exact match:** skip (already present)
   - **Same entity, different content** (same company + overlapping dates, different title or bullets): keep ALL content, add `<!-- REVIEW: same company/dates, different content — verify and consolidate -->`
4. Write merged master.md
5. Print: "Merge complete: N added, N flagged for review (<!-- REVIEW --> comments), N unchanged"

**Rule:** NEVER silently remove any content from the master resume. Union only.

---

## [PARSE-TRANSCRIPT] — Extract Skills from Academic Transcript

Input: file path to transcript (.pdf or .txt)

1. Run `~/.agents/venv/bin/python ~/.agents/skills/resume-tailor/scripts/extract_text.py <path>`
2. Read ALL courses regardless of grade or pass/fail status
3. Map courses to skill categories:
   - CS/programming/data/AI courses → Technical Skills (add to appropriate subcategory)
   - Aerospace/engineering/physics/math courses → Engineering & Domain Knowledge
   - Leadership/military courses → Achievements section (if veteran-relevant) or Soft Skills
4. For courses implying hands-on work (labs, capstone, design courses, experimental courses):
   - Ask: "[Course name] — did this involve a project worth adding to your Projects section?"
5. Identify any new skills not already in master.md and add them
6. Print: "Parsed [N] courses. Added: [N] skills to Technical Skills, [N] to Engineering & Domain Knowledge, [N] projects flagged for review."

**Rule:** Grades do not gate skill extraction. ~80% of an aerospace engineering degree = genuine domain knowledge regardless of grade record.

---

## Quality Checks (apply to all outputs)

- No "Responsible for..." openers — use action verbs (Built, Designed, Led, Developed, Implemented)
- No years-of-experience filtering from JD applied to master resume entries
- No GPA, no honors, no coursework in Education section of tailored resume
- No photo placeholder
- No personal data in any file under `~/.agents/skills/resume-tailor/` (public repo)
- All scripts run via `~/.agents/venv/bin/python`