Profile

Profile

Pre-flight

• Load active preferences (see memory skill):

  bashCopy code
  node scripts/db.js "SELECT category, key, value, confidence, source FROM preferences WHERE user_id = <user_id> AND status = 'active' ORDER BY category, key"
  

• Load existing profile if present:

  bashCopy code
  node scripts/db.js "SELECT data->'profile' AS profile, data->'job_preferences' AS job_preferences, data->'style_profile' AS style_profile FROM users WHERE id = <user_id>"
  

• If profile exists, validate changes before overwriting

Flow overview

Copy code
Step 1: CV analysis → extract experience, sector, profile, inferred seniority
Step 2: Gap questionnaire → only what the CV doesn't clarify (adaptive to inferred profile)
Step 3: Current situation + expectations → employment status, urgency, salary, work mode, availability
Step 4: Strategy → targets, sources, aggressiveness (informed by everything above)
Step 5: Polish suggestion → align CV and LinkedIn profile to the job target to maximize matches

Step 1: CV analysis

Ask user for CV (URL or PDF). Extract:

• Full name and title/profession
• Professional summary (elevator pitch)
• Work experience (company, role, period, achievements, team size, reporting line if applicable)
• Core competencies and tools (tech stack, software, methodologies, equipment, whatever is relevant to the field)
• Soft skills (leadership, communication, etc.)
• Certifications and courses
• Education (degrees, institutions)
• Languages and proficiency level
• Quantifiable achievements (metrics, impact)
• Notable projects or relevant work samples

Step 1b: Inferred profile

From the CV data, infer:

Save to users.data.profile as JSONB, including the inferred fields:

bashCopy code
node scripts/db.js "UPDATE users SET data = jsonb_set(data, '{profile}', '<json>'::jsonb) WHERE id = <user_id>" --write

The inferred profile is preliminary. Step 2 confirms or corrects it.

Step 2: Gap questionnaire

Not a fixed set of blocks. The agent generates questions based on what the CV already answered clearly. Only ask about gaps.

Each answer must have a weight: Must (non-negotiable), Strong (strong preference), Nice (would be a plus).

Core questions (always ask if the CV doesn't clarify)

Role and career direction

• Role type (IC, manager, mixed, specialist, leadership) — adapt options to career_stage and has_management
• Career stage confirmation (the agent proposes the inferred value, user confirms or corrects)
• Growth direction (stay in current path, wants to pivot, wants to move to leadership, unsure)
• Mentorship expectations (wants mentorship, wants autonomy, indifferent) — only if career_stage is junior/mid

Management and autonomy (only if has_management = true or user expressed leadership aspirations)

• Expected reporting line (CEO, CTO/COO, VP, other director) — adapt to sector
• Expected hiring/firing authority
• Budget authority (own budget decisions)
• Org scope (single team, multiple squads, department, company-wide)

Work mode and geography

• Work mode (remote, hybrid, on-site)
• Current location and willingness to relocate
• Accepted timezones (Americas, Europe, Asia, global)
• Contract type (employee, contractor, freelancer)

Company type and sector

• Company size (startup, scale-up, mid-size, corporate)
• Company stage (early-stage, growth, established, public) — only if relevant to the sector
• Preferred sectors or industries (if industry_history shows concentration, confirm rather than ask open-ended)
• Sectors to avoid (with nuance: absolute or accepts partial exposure?)
• Product/service type (own product, internal platform, consulting, services) — adapt to field

Sector-specific focus

• Area of specialization within the field (the agent proposes based on CV, user confirms or refines)
• Tools/methodologies the user wants to keep using vs open to learn
• Type of impact desired in first 6 months (adapt to role type)

Deal-breakers

• Graduated deal-breakers (the agent presents an empty list, the user adds their own). Never pre-load deal-breakers like "junior" or any role level. Each user defines their own.

How the agent decides what to ask

1. For each topic above, check if the CV already provides a clear answer
2. If clear → skip the question, save the inferred value with confidence note
3. If unclear or missing → ask the question
4. If the topic only applies to certain career stages (management questions) → skip if not applicable
5. Present questions in blocks of 4, multi-select where applicable

Save to users.data.job_preferences as JSONB with weights:

bashCopy code
node scripts/db.js "UPDATE users SET data = jsonb_set(data, '{job_preferences}', '<json>'::jsonb) WHERE id = <user_id>" --write

Step 3: Current situation and expectations

Ask the user directly. These are not inferable from a CV.

Employment status and urgency

• Current situation (employed active change, employed passive, unemployed, about to be unemployed, first job)
• How urgent is the search? (no urgency, in the coming months, soon, desperate)
• Availability to start (immediate, 2 weeks, 1 month, 3 months)

Compensation

• Salary range (min, expected, currency)
• Equity/participation expectations (if applicable to sector and career stage)
• Important benefits (health, education budget, equipment, etc. — adapt to sector)
• Flexibility on must-haves (e.g: 100% remote absolute or accepts 1 quarterly trip)

Availability for interviews

• Preferred time slot for interviews (e.g: "13:00 to 16:00 AR")
• Fixed blocked days/times (e.g: "Tuesday 14:00 to 15:00, English class")
• Timezone (default: )

Save to users.data:

bashCopy code
node scripts/db.js "UPDATE users SET data = jsonb_set(data, '{availability}', '{\"preferred_hours\":\"<start>-<end> AR\",\"timezone\":\"<tz>\",\"blocked\":{\"<day>\":\"<start>-<end> (<reason>)\"}}') WHERE id = <user_id>" --write

The news flow uses this to filter available slots from scheduling links without asking the user each time.

Step 4: Strategy

Now that the agent has the full profile (CV + gaps + situation), it can propose a strategy informed by everything above.

See the strategy skill for the full flow. Summary:

1. Show the 4 levels (passive, selective, active, aggressive)
2. Ask situation questions adapted to the user's career stage:
   • If career_stage is junior/mid: "Are you open to roles above your current level, or only same-level matches?" instead of "IC or Manager?"
   • If career_stage is senior+: "Would you accept IC roles or only Manager?" (only if has_management = true)
3. Propose a level based on answers
4. Allow customization of individual parameters
5. Save to DB (preferences.workflow.strategy_level + users.data.strategy)

The strategy's relax_must_haves resolves dynamically from the user's Must-weighted preferences (see AGENTS.md "Strategy levels"). No hardcoded keys.

Step 5: Polish suggestion

After the strategy is defined, the agent suggests aligning the CV and LinkedIn profile to the job target to maximize match chances.

What to do

1. Compare the user's current CV and LinkedIn profile (from users.data.profile and users.data.linkedin_profile) against the defined job target (role, seniority, sector, must-haves)
2. Identify gaps:
   • Keywords missing that recruiters search for
   • Titles or descriptions that don't align with the target role
   • Skills underrepresented relative to what the target market demands
   • LinkedIn headline/summary that doesn't position the user for the target
3. Present the analysis to the user
4. If the user agrees, trigger the polish skill which:
   • Audits the CV and LinkedIn profile section by section
   • Drafts improvements aligned with the target
   • Applies changes with per-section approval
   • Exports a polished CV to PDF

How to present it

One short message, conversational (Gold Rule 7 applies):

"Ahora que sabemos que buscas en , tu CV y perfil de LinkedIn tienen algunas cosas que se pueden alinear mejor para que te encuentren más fácil. Querés que haga una revisión y te proponga cambios?"

If the user says yes → run polish flow. If the user says no or later → skip, but remind them once at the end of onboarding.

Phase: Voice and style

This runs as part of Step 1 or as a separate phase after Step 5. It's independent of the career stage changes.

Automatic inference

• Open headless browser with persistent profile
• Read last 20-50 sent messages on LinkedIn (filter "You:")
• Read relevant sent emails in Gmail (to recruiters, HR, companies)
• Infer: tone, default language, writing characteristics, average length
• Extract 3-5 representative samples (1-2 recruiter, 2-3 personal)

Confirmation with options

Ask the user:

• Tone (formal, casual-professional, casual, direct/no-nonsense)
• Default language (Spanish, English, depends on context)
• Preferred length (short 1-3 lines, medium 4-6, long 7+)
• Preferred greeting (Hi [name], Dear, no greeting, other)
• Preferred closing (Regards, Cheers, no closing, other)
• Use bullet lists in messages? (yes, no)
• Emojis in professional messages? (yes, no, only in personal)

Validation

• Save inference + preferences to users.data.style_profile as JSONB
• Show 3 messages drafted with the style to the user for validation
• If user corrects → update style_profile

Phase: Platforms (output, not input)

• Consult PLATFORMS.md
• Cross-reference user profile vs role types/industries/seniority of each platform
• Assign Tier 1/2/3 to platforms based on fit
• Save to users.data.platforms as JSONB
• Don't ask the user. This is the output of analysis

Rules

• CV is source of truth. Questionnaire covers only what the CV doesn't say
• CV-first inference drives the questionnaire. The agent analyzes the CV, infers a preliminary profile, and generates only the gap questions that matter for that specific profile
• Career stage determines which questions appear. Management questions only for senior+ or users with management experience. Mentorship questions only for junior/mid. The repo never assumes a career stage
• No hardcoded deal-breakers. The deal-breakers list starts empty. Each user adds their own. Never pre-load role levels (junior, senior, etc.) as system deal-breakers
• Sector-agnostic. The profile flow works for any field (software, design, marketing, finance, operations, etc.). Never assume dev-specific concepts (tech stack, RAG, agents, IC) as universal. Adapt terminology to the user's field
• All DB access via scripts/db.js (see db skill). Read-only by default, --write for saves
• Persist everything to users.data as JSONB via jsonb_set:

  bashCopy code
  node scripts/db.js "UPDATE users SET data = jsonb_set(data, '{profile}', '<json>'::jsonb) WHERE id = <user_id>" --write
  node scripts/db.js "SELECT data->'profile' AS profile FROM users WHERE id = <user_id>"
  

• If user already has a profile in DB, validate changes before overwriting
• Profile is updated when user changes CV or answers new questions
• Questions in blocks of 4, multi-select where applicable
• Each preference with weight: Must / Strong / Nice
• Platforms = output of analysis, never user input
• Single user (repo owner)