resume-tailoring

Resume Tailoring

Generate tailored, multi-format resumes optimized for specific job descriptions while maintaining factual integrity. Build resumes around the whole person by surfacing undocumented experiences through conversational discovery.

Core Principle: Truth-preserving optimization - maximize fit without fabricating experience. Reframe and emphasize relevant aspects; never invent work history.

References

• references/research-prompts.md - JD parsing templates, company research queries, role benchmarking strategy, success profile synthesis
• references/matching-strategies.md - Weighted scoring formula (Direct 40%, Transferable 30%, Adjacent 20%, Impact 10%), confidence bands, reframing strategies, gap handling options
• references/branching-questions.md - Technical/soft-skill/recent-work question trees, branching logic, multi-job context framing, capture structure
• references/multi-job-workflow.md - Batch initialization, aggregate gap analysis, shared discovery, per-job processing, incremental batch addition, pause/resume support

Requirements

From user:

1. Job description (text or URL)
2. Resume library location (defaults to <workspace>/library/ where workspace = ${JOB_HUNT_WORKSPACE:-$HOME/Documents/job_seeking})

Sub-skill dependencies:

• export-pdf - HTML → PDF export via ${PLUGIN_ROOT}/skills/export-pdf/scripts/html-to-pdf.sh

Workflow

Multi-Job Detection

Before starting, check if the user provides 2+ JDs, mentions "multiple jobs", "batch", "several positions", or lists multiple company/role pairs. If detected, confirm with user and follow references/multi-job-workflow.md. Otherwise proceed with single-job workflow below.

Phase 0: Library Initialization

1. Locate resume library directory (user-provided or ${JOB_HUNT_WORKSPACE:-$HOME/Documents/job_seeking}/library/)
2. Scan for HTML and markdown files using Glob
3. Parse each resume: extract roles, bullets, skills, education
4. Build in-memory experience database - tag each bullet with themes, metrics, keywords, and source resume
5. Announce library size to user

Phase 1: Research

Follow templates in references/research-prompts.md.

1. Parse JD: extract requirements (must-have vs nice-to-have), keywords, implicit preferences, red flags, role archetype
2. Research company via WebSearch: mission, values, culture, recent news, engineering blog
3. Benchmark role via WebSearch/WebFetch: LinkedIn profiles of similar role holders, common backgrounds, terminology
4. Synthesize into a success profile: core requirements, valued capabilities, cultural fit signals, narrative themes, terminology map, risk factors

Checkpoint: Present success profile summary to user. Wait for confirmation before proceeding.

Phase 2: Template Generation

1. Analyze library for role archetypes, experience clusters, career progression
2. Make role consolidation decisions (merge same-company roles when responsibilities overlap; keep separate when different companies or dramatically different scope)
3. Propose title reframing options - stay truthful, emphasize aspects most relevant to target role
4. Generate template skeleton with bullet allocation per role (more bullets to more relevant roles), section ordering, and guidance per slot

Checkpoint: Present template structure with consolidation decisions, title options, and bullet allocation. Wait for user approval.

Phase 3: Experience Discovery (Optional)

Trigger after template approval if gaps identified (confidence < 60% on critical requirements).

1. Present identified gaps with current confidence levels
2. Offer structured brainstorming session (typically 10-15 minutes)
3. Conduct branching interview per gap using patterns from references/branching-questions.md:
   • Start with open probe (technical, soft skill, or recent work)
   • Branch based on answer: YES/direct -> deep dive; INDIRECT -> explore transferability; ADJACENT -> explore related; PERSONAL -> assess recency; NO -> try broader category or move on
4. Capture each discovery immediately: context, scope, metrics, bullet draft, which gaps it addresses
5. Present integration options per experience: add to current resume, add to library only, refine further, or discard

Phase 4: Assembly

Follow scoring from references/matching-strategies.md.

1. For each template slot, score all candidate bullets:
   • Overall = (Direct x 0.4) + (Transferable x 0.3) + (Adjacent x 0.2) + (Impact x 0.1)
2. Rank by score and assign confidence bands: DIRECT (90-100%), TRANSFERABLE (75-89%), ADJACENT (60-74%), WEAK/GAP (<60%)
3. Present top 3 matches per slot with analysis and source resume
4. Apply reframing where needed (keyword alignment, emphasis shift, abstraction level, scale emphasis) - show before/after transparently
5. Handle gaps (<60%): offer reframe, flag for cover letter, omit slot, or use best-available with disclosure

Checkpoint: Present coverage summary (direct/transferable/adjacent/gap percentages), reframings applied, gap recommendations. Wait for approval.

Phase 5: Generation

Output directory: ${JOB_HUNT_WORKSPACE:-$HOME/Documents/job_seeking}/<company>/ Filename format: <First>_<Last>_<Role>_CV.<ext> — NO company name in the filename.

Edit-guard: If this skill needs to modify the master HTML at ${JOB_HUNT_WORKSPACE:-$HOME/Documents/job_seeking}/<First>_<Last>_<Role>_CV.html, call AskUserQuestion for explicit confirmation before doing so.

1. HTML: Compile mapped content into a tailored HTML file following the master CV's template. Save as <First>_<Last>_<Role>_CV.html inside the company subfolder.
2. PDF: Run the export-pdf skill:

   bash "${PLUGIN_ROOT}/skills/export-pdf/scripts/html-to-pdf.sh" \
     "<workspace>/<company>/<First>_<Last>_<Role>_CV.html" \
     "<workspace>/<company>/<First>_<Last>_<Role>_CV.pdf"
   

   Always use absolute paths for both arguments.
3. Report: Generate metadata file with coverage stats, reframings, source resumes, gap analysis before/after, interview prep recommendations. Save as <First>_<Last>_<Role>_CV_Report.md inside the company subfolder.

Present files and quality metrics to user.

Phase 6: Library Update

Present three options:

1. Save to library - move files, rebuild library database, preserve generation metadata
2. Need revisions - collect feedback, make changes, re-present
3. Save without library update - keep files in current location only

Error Handling

Scenario	Action
Insufficient library (1-2 resumes)	Warn about limited matching; recommend discovery phase; proceed with available content
No good matches (<60% on critical requirement)	Offer: run discovery, reframe adjacent, omit slot, note for cover letter
Research failures (WebSearch unavailable, sparse results)	Fall back to JD-only analysis; ask user for additional context
Vague JD	Flag missing areas; ask user for context; proceed with best-effort
Content exceeds page limit	Rank bullets by relevance; suggest pruning lowest-scored; let user decide
PDF export failure (html-to-pdf.sh non-zero exit)	HTML is still saved; report error with exit code; user can re-run export-pdf separately

Hard Rules

• NEVER fabricate experience or inflate seniority beyond defensible
• Company names and dates MUST be exact
• Show all reframings transparently with before/after and justification
• Every checkpoint requires user confirmation before proceeding
• Help articulate genuine experience; never create false experience