Resume Tailoring

/resume-tailoring — Job-Specific Resume Workflow

Build a resume around the whole person, not just the documents on file. Maximize fit to the target role while keeping every bullet factually defensible.

Core rule: Never fabricate experience. Reframe, emphasize, and translate terminology — do not invent.

Trigger prompt (canonical):

I want to apply for [Role] at [Company]. Here's the JD: [paste JD or URL]

Multi-job mode triggers automatically when the user pastes multiple JDs, multiple URLs, or says "these roles"/"batch"/"multiple jobs".

Quick Start

/resume-tailoring                              → run full workflow, single job
/resume-tailoring --library ~/resumes          → point at a specific library
/resume-tailoring --batch                      → force multi-job mode
/resume-tailoring --no-docx                    → skip DOCX/PDF, emit markdown only
/resume-tailoring --express                    → skip checkpoints (only use after first run)

Inputs the skill needs:

1. Job description (text or URL).
2. A resume library: markdown files under resumes/ in the current dir (or a path the user provides). Works with 1 resume, but 10+ is ideal.

Outputs written to resumes/tailored/{Company}_{Role}/:

• {Name}_{Company}_{Role}_Resume.md
• {Name}_{Company}_{Role}_Resume.docx (requires document-skills:docx)
• {Name}_{Company}_{Role}_Resume.pdf (optional)
• {Name}_{Company}_{Role}_Report.md — coverage %, matches, reframings, interview prep

Workflow

Phase 0 — Library build (always first)

1. Resolve library dir: use --library flag, else ./resumes/, else ask. Validate it exists.
2. Glob *.md in the library. Announce count: Building library... found {N} resumes.
3. For each resume, Read and parse: contact, roles (company/title/dates), bullets, skills, education.
4. Tag each bullet: themes (leadership, technical, analytics, strategy…), metrics (numbers, %, $), keywords (action verbs, domain terms), source_resume.
5. Detect user preferences: typical length (1 vs 2 page), section order, bullet style.
6. Hold the library in memory as the candidate pool for matching.

If library has <3 resumes, warn the user: coverage will be thinner; discovery (Phase 2.5) will be especially valuable.

Phase 1 — Research

Build a success profile — what would make any candidate successful for this role, not just what the JD says.

Use research-prompts.md templates. Three sub-steps:

1. Parse JD. Extract must-haves, nice-to-haves, technical keywords, implicit preferences, red flags, role archetype (IC / manager / cross-functional).
2. Company research. WebSearch: mission/values/culture, engineering blog, recent news, team structure.
3. Role benchmarking. WebSearch site:linkedin.com {title} {company} → WebFetch top 3-5 profiles → extract common backgrounds, skills, terminology.

If research tools are unavailable or sparse, fall back to JD-only analysis and tell the user what you couldn't get.

Checkpoint: Present the success profile (core reqs, valued capabilities, cultural signals, narrative themes, terminology map, risk factors). Wait for the user to confirm or adjust before continuing.

Phase 2 — Template

Build the resume structure optimized for this role. See matching-strategies.md for title reframing and consolidation rules.

1. Role consolidation. Combine same-company consecutive positions only when continuity serves the target role better than granularity. Never merge across companies.
2. Title reframing. Stay truthful; emphasize the aspect most relevant to the target. Company, dates, and core responsibilities must be exact.
3. Section order. Use the user's library preferences unless the target role dictates otherwise (e.g., Education top if fresh grad / required credential).
4. Bullet allocation. Weight by role relevance × recency. Front-load the most relevant role.

Checkpoint: Present template skeleton with consolidation decisions, title options (with rationale), and bullet counts per role. Wait for approval.

Phase 2.5 — Experience discovery (optional, offered when gaps found)

When any requirement has <60% confidence after a first-pass match, offer a 10–15 min branching interview. Use branching-questions.md patterns:

• Technical gap: "Have you worked with {skill} or {adjacent}?" → branch on YES / INDIRECT / ADJACENT / PERSONAL / NO.
• Soft skill gap: "Tell me about times you've {demonstrated_skill}" → branch on STRONG / VAGUE / PROJECT-SPECIFIC / VOLUNTEER.
• Recent-work probe: "What have you worked on in the last 6 months not on your resume yet?"

Capture each discovered experience as a draft bullet with context, scope, addressed gaps, and truthfulness note. Never pressure the user into a claim they can't defend. Time-box and move on after 2–3 dry attempts per gap.

Phase 3 — Assembly (match + score)

For each template slot, score every candidate bullet using matching-strategies.md:

Overall = 0.4·Direct + 0.3·Transferable + 0.2·Adjacent + 0.1·Impact

Bands: 90+ DIRECT · 75–89 TRANSFERABLE · 60–74 ADJACENT · 45–59 WEAK · <45 GAP.

For each slot, show top 3 candidates with scores and source resume, plus any reframing proposals (with a "why this is still truthful" line). For gaps: offer reframe, omit, cover-letter, or run another discovery loop.

Checkpoint: Present the full mapping plus coverage summary (% direct, % transferable, % gaps, JD coverage %). Wait for approval before generation.

Phase 4 — Generation

Write outputs to resumes/tailored/{Company}_{Role}/.

1. Markdown resume. Clean, consistent with user's library style.
2. DOCX. Use document-skills:docx sub-skill. Calibri 11pt body / 12pt headers, 0.5–1in margins, proper bullet numbering config (never unicode bullets), bold for company/title/dates.
3. PDF (if requested or DOCX succeeds). Via document-skills:pdf or DOCX→PDF conversion.
4. Generation report. Target summary, success profile, coverage metrics, every reframing with before/after + reason, source resume breakdown, remaining gaps with recommendations, interview prep notes (stories to prepare, likely questions, how to address gaps).

If DOCX/PDF generation fails, fall back to markdown-only and note the failure in the report.

Phase 5 — Save + learn (conditional)

Ask the user: (1) Save to library, (2) Revise, (3) Save locally only.

On save: move all artifacts into the library, re-parse to enrich the candidate pool, and persist a metadata file (.meta.json) with match scores, reframings, newly-discovered experiences, and jd_coverage. Future sessions start from a richer library.

Multi-Job Mode

Triggers when the user pastes multiple JDs/URLs or asks for batch. Full workflow in multi-job-workflow.md.

High-level phases:

1. Intake. Collect 3–5 JDs, priorities, notes. Initialize batch state at resumes/batches/batch-{date}-{slug}/.
2. Aggregate gap analysis. Match each JD's requirements against the library. Deduplicate across jobs. Classify: HIGH-leverage (3+ jobs), MEDIUM (2), LOW (1).
3. Shared discovery. One branching interview that addresses high/medium leverage gaps first. Tag each discovered experience with job relevance.
4. Per-job processing. For each job: research → template → matching → generation (sequential, reusing enriched library).
5. Batch finalization. Review all resumes together, approve/revise individually or as a batch, update library once.

Incremental adds. If a batch already exists, the user can add jobs later; only new gaps trigger discovery.

Typical time savings: 11% for 3 jobs, 27% for 5 jobs vs sequential single-job runs.

Checkpoints & User Control

Four hard stops (marked Checkpoint: in Phases 1–4 above) pause for approval before continuing. Any checkpoint can request: go back, adjust, or accept. --express skips checkpoints 1–3 (use only for repeat applications in a trusted batch).

Truthfulness Rules (non-negotiable)

• Never invent a role, company, date, metric, or skill.
• Never inflate seniority beyond what scope defends (e.g., "Lead" only if you led).
• Reframe, don't fabricate. Same facts, different emphasis and terminology — always.
• Disclose gaps honestly in the report; offer cover-letter recommendations rather than padding the resume.
• Show before/after for every reframing in the generation report, plus a one-line reason the reframing is still accurate.

Edge Cases

• Tiny library (<3 resumes): Warn, lean heavily on Phase 2.5 discovery.
• Critical gap (<60% on must-have): See matching-strategies.md gap-handling options — never force a match.
• Research fails: Fall back to JD-only; ask user for culture/team context; proceed best-effort.
• Vague JD: Extract what's possible, ask user for missing context, proceed.
• >2 page overflow: Rank bullets by match score, propose lowest-ranked for cut, let user decide.
• Career gap: Frame legitimately (startup, caregiving, study, etc.); surface skills gained during the gap via discovery.

References

• research-prompts.md — JD parsing, company research, role benchmarking, success profile synthesis templates.
• matching-strategies.md — scoring weights, reframing strategies, title reframing, role consolidation rules, gap handling.
• branching-questions.md — discovery interview branches for technical, soft-skill, and recent-work gaps.
• multi-job-workflow.md — full batch workflow with state schema, aggregate gap analysis, shared discovery.

Sub-Skills Used

• document-skills:docx — DOCX generation (required for .docx output).
• document-skills:pdf — PDF generation (optional).
• WebSearch / WebFetch — company and role research (falls back gracefully).