job-intake

name	job-intake
description	Parse job descriptions, score candidate fit, and set up application tracking folders. Use this skill whenever the user shares a job posting — pasted into chat, saved as a file, or referenced by URL — and wants to evaluate whether it's a good match. Also trigger when the user says "check this job," "is this a fit," "parse this posting," "new application," "evaluate this role," "score this job," or shares a block of text that looks like a job description (title, company, requirements, responsibilities). This skill feeds into resume-tailor and cover-letter-writer — run it first to create the application folder that downstream skills expect. Do NOT trigger for general career advice, salary research, or interview prep — those are different workflows.
recommended_model	{"default":"sonnet","reasoning":"This skill does structured extraction, pattern matching against an achievements inventory, file creation, and scoring. These are well-defined tasks that don't require deep creative reasoning or nuanced prose generation. Sonnet handles this reliably at ~5x lower cost than Opus. The learning loop (Step 6) involves simple append operations and rescoring — no model upgrade needed.\n","upgrade_to_opus_when":"The job description is unusually ambiguous or the candidate asks for strategic advice beyond scoring (e.g., \"should I pivot my career toward this kind of role?\"). Career-level reasoning benefits from Opus. For standard intake, Sonnet is sufficient.\n"}

Job Intake — Parse, Score, and Track Job Applications

Intent

1. Be honest about match quality — a "long shot" assessment that saves hours of wasted effort is more valuable than false encouragement
2. Surface red flags without burying them — unrealistic requirements, vague language, or signs of dysfunction deserve the same prominence as match strengths
3. Accumulate durable knowledge — every evaluation should make achievements.md richer, improving all future scoring
4. Prevent wasted applications — screen out poor-fit roles early so effort concentrates on strong and good matches
5. Create reliable structured data — downstream skills depend on accurate metadata; garbage in means garbage out across the entire pipeline
6. Never create folders from partial information — premature folder creation pollutes the workspace with incomplete data that misleads downstream skills
7. Process a full intake in under 10 minutes — speed matters, but never at the cost of honest scoring or incomplete data

Why this skill exists

Applying to jobs without a system means losing track of what you sent where, missing follow-up windows, and spending time on roles that aren't a good fit. This skill is the front door — every application starts here, and every downstream skill (resume tailoring, cover letter writing, tracking) depends on the structured output it produces.

Prerequisites

This skill expects a workspace at job-applications/ with:

• master/base-resume.docx — the candidate's canonical resume
• master/achievements.md — categorized, quantified accomplishments
• master/narrative.md — tone and positioning guide
• tracker.xlsx — master application spreadsheet

If any of these are missing, tell the user and offer to help set them up before proceeding.

Workflow

Step 0: Check for duplicate applications

Before doing any evaluation work, check for an existing application matching the company+role the user is asking about. This check must happen first — it's fast, and it prevents wasted effort and duplicate folders.

Primary:  Read job-applications/applications/index.json
          Match on: (company name, role title) — case-insensitive, fuzzy on company variants
Fallback: If index.json doesn't exist, scan job-applications/applications/*/metadata.json
          (slower with 300+ folders — rebuild index.json afterward)

After creating any new application folder (Step 4), append the new entry to index.json so future lookups stay fast.

If a match is found, tell the user immediately and show the existing application's status. The response depends on the state of the existing application:

Existing status	What to say
evaluating	"You started evaluating this role on [date] but didn't finish. Want me to pick up where you left off, or start fresh?"
ready_to_apply	"You already have a tailored resume and cover letter ready for this one. Want to review them or submit?"
applied	"You already applied to this on [applied_date]. Follow-up date is [date]. Want to check status instead?"
interviewing	"You're already in the interview process for this role. Want to prep or update status?"
rejected	"You were previously rejected for this role on [date]. Want to re-apply with a fresh application?"
withdrawn	"You previously withdrew from this role. Want to re-apply?"

For re-applications (rejected or withdrawn): create a new folder with today's date. The old folder stays as history.

For stalled evaluations (evaluating with no progress): offer to resume in the existing folder rather than creating a duplicate. Only create a new folder if the user explicitly wants to start over.

If no match is found, proceed to Step 1.

Step 1: Accept the job description

The user will provide the job description in one of three ways:

• Pasted into chat — the most common case. The text might be messy (copied from a web page with formatting artifacts). Clean it up mentally but preserve all content.
• File in the workspace — a .md, .txt, or .docx file in the job-applications folder or uploads. Read it with the appropriate tool.
• URL — fetch the page and extract the job posting content.

Handling partial or incomplete input:

Not every trigger comes with a full job description. The user might say "saw a VP of Eng role at Google, sounds like me right?" or "what do you think about this — Chief Architect at Cognizant, cloud migration, healthcare." When the input has a company and role but is missing a full requirements list:

1. Give a preliminary read — use what you DO have plus master/achievements.md to give quick directional feedback. Example: "Based on what you've shared, healthcare AI and cloud architecture are strong matches in your background. A Chief Architect role at a consultancy like Cognizant aligns well with your experience there."
2. Be explicit about what you CAN'T assess — "Without the full requirements list, I can't give you a proper match score or identify specific gaps."
3. Ask for the full posting — "Can you grab the full job description so I can do a proper evaluation? I'll set up the application folder and score it against your resume."
4. Do NOT create a folder or metadata.json from partial info — the application folder is only created in Step 4 after a full parse. Premature folder creation pollutes the workspace with incomplete data.

If the input is truly minimal (just a company name with no role, or a vague "any jobs in healthcare?"), this isn't a job-intake trigger. Respond conversationally and help them narrow down what they're looking for.

Step 2: Extract structured data

Pull out these fields from the job description. If a field isn't present, mark it as "Not specified" rather than guessing:

Field	What to extract
Company	Organization name
Role	Exact job title
Location	City/state, remote status, hybrid details
Compensation	Salary range, equity, benefits if listed
Requirements	Hard requirements (years of experience, specific skills, degrees)
Preferred	Nice-to-haves that aren't dealbreakers
Responsibilities	What the role actually does day-to-day
Red flags	Unrealistic combos (e.g., "10 years of Kubernetes"), vague language, signs of dysfunction
Keywords	Technical terms, tools, frameworks, and domain language that should appear in a tailored resume

Step 2b: Check for former-employer connection

Before scoring, scan master/achievements.md for the company name extracted in Step 2. Look for exact and fuzzy matches — "Cognizant" should match "Cognizant Softvision," and "Red Spot" should match "Red Spot Interactive."

If the company appears in the candidate's achievement history:

1. Flag it immediately — tell the user: "This is at [Company], where you previously worked as [Role]. That's a very different application strategy than a cold apply."
2. Note it in metadata.json — add "former_employer": true and "former_employer_role": "Previous role title" to the metadata.
3. Adjust the scoring context — when a candidate previously held a role at the same company (or a very similar one), most "requirements" are automatically strong matches because they literally did the job. The scoring in Step 3 should reflect this — don't manufacture gaps for skills the candidate demonstrably used at that exact company.
4. Surface strategic considerations — returning to a former employer means the user likely has internal contacts, knows the culture, and may be able to skip parts of the interview process. Mention this in the Step 5 summary.

If no match is found, proceed normally — no flag needed.

Step 3: Score the match

Read master/achievements.md and compare the extracted requirements against the candidate's background. For each requirement, classify it:

• Strong match — direct experience with quantified results
• Partial match — related experience that could be positioned
• Gap — no clear experience to point to

Produce an overall fit assessment:

Score	Meaning
Strong	80%+ requirements matched, no critical gaps
Good	60-80% matched, gaps are addressable in cover letter
Stretch	40-60% matched, significant positioning needed
Long shot	Below 40%, probably not worth tailoring materials

Be honest about gaps. The goal isn't to be encouraging — it's to help the candidate spend time on the right applications. A "long shot" assessment with clear reasoning saves hours of wasted effort.

For each gap identified, note whether it's addressable (can be positioned through adjacent experience) or hard (no realistic way to bridge it).

Step 4: Create the application folder

Create a dated folder in job-applications/applications/ using the naming convention:

YYYY-MM-DD_company-name_role-slug/

Company name and role slug should be lowercase, hyphenated, with special characters removed. Examples:

• 2026-02-24_healthfirst_vp-engineering
• 2026-03-01_amazon-web-services_senior-sde

Inside the folder, create:

job-description.md — the full job description, cleaned up with a header containing company, role, location, and the source URL if available.

metadata.json — structured data for tracking and downstream skills:

{
  "company": "Company Name",
  "role": "Job Title",
  "location": "City, State / Remote",
  "compensation": "Range if listed, otherwise null",
  "applied_date": null,
  "source": "LinkedIn / Indeed / Referral / etc.",
  "source_url": "URL if available",
  "status": "evaluating",
  "follow_up_date": null,
  "contact": "",
  "resume_version": null,
  "cover_letter": null,
  "former_employer": false,
  "former_employer_role": null,
  "notes": "",
  "match_score": {
    "overall": "strong | good | stretch | long_shot",
    "requirements_matched": ["list of matched requirements"],
    "requirements_partial": ["list of partial matches"],
    "gaps": ["list of gaps"],
    "addressable_gaps": ["gaps that could be positioned"],
    "hard_gaps": ["gaps with no realistic bridge"],
    "keywords": ["terms to use in tailored resume"]
  }
}

Step 5: Present the summary

Give the user a clear, conversational summary covering:

1. The role in one sentence — what this job actually is
2. Match score with reasoning — not just the label, but why
3. Top strengths — the 3-4 requirements where the candidate's experience is strongest
4. Gaps to address — what's missing and whether it's bridgeable
5. Keywords to hit — terms the resume and cover letter should include
6. Recommendation — should they apply? If stretch or long shot, say so clearly

End by asking if they want to proceed to resume tailoring (which would trigger the resume-tailor skill).

Step 6: Learn from corrections

After presenting the summary, the user may push back on gaps. They might say something like "actually, I did SOC2 at Red Spot — we went through the audit in 2019" or "I forgot to mention, I managed a $5M budget at MedQuest, not just $2M."

This is new information. Don't just use it for the current scoring — persist it so every future job evaluation benefits. When the user provides a correction or new accomplishment:

1. Update master/achievements.md — add the new item under the appropriate category. Include a [learned] tag and the date so the user can see what was added through conversation versus what was in the original resume. Format:

   - Led SOC2 Type II audit preparation and remediation (Red Spot, 2019) [learned: 2026-02-24]
   

2. Rescore the current job — rerun Step 3 with the updated achievements. The gap that triggered the correction should now show as a strong or partial match. Update the metadata.json in the application folder with the new score.

3. Confirm the update — tell the user what you added and where. They should know their achievements file is growing. Example: "Added your SOC2 audit experience to the compliance section of achievements.md. This bumped the HealthFirst match from 'good' to 'strong' — that gap is closed now."

This learning loop is one of the most valuable parts of the system. Over time, achievements.md becomes a comprehensive inventory of everything the candidate can claim — far richer than any single resume version. It captures experience the user might not think to put on a resume but that matters for specific roles.

When NOT to update achievements.md:

• Aspirational statements ("I could learn Terraform quickly") — these aren't accomplishments
• Vague claims without specifics ("I've done compliance stuff") — ask for details first
• Information that contradicts existing entries — flag the conflict and let the user resolve it

Skill composition

This skill is the entry point for the job application workflow:

Next step	Skill	When
Tailor resume	resume-tailor	User decides to apply
Write cover letter	cover-letter-writer	After resume is tailored
Update tracker	job-tracker	After application is submitted

The metadata.json created by this skill is the contract that downstream skills read.

Edge cases

• Multiple jobs at once — If the user pastes several job descriptions, process each one separately. Create individual folders. Don't try to batch the analysis.
• Incomplete postings — Some listings are just a title and a paragraph. Follow the partial-input handling in Step 1: give a preliminary read from what's available, be clear about what you can't assess, and ask for the full posting. Don't create folders.
• Internal referrals — If the user mentions a referral or connection, capture it in the metadata contact field and notes.
• Reprocessing — Handled by Step 0 duplicate detection. The user is warned before any work begins and can choose to resume, update, or start fresh.