/salary-calculator

Runs the full Salary Calculator pipeline. Generates ready-to-copy research prompts for local-market salary data, ingests your research results, applies international purchasing-power adjustments, and produces a final salary table with shown calculations. Runs standalone or scoped to a single country handed off from Country Finder. Resumes from the last completed step if interrupted.

Flow

flowchart TD
  Start([User runs /salary-calculator]) --> Handoff{Single country<br/>handed off?}
  Handoff -->|yes| RecordCountry[Record target_country —<br/>skip country list in Step 1]
  Handoff -->|no| Profile
  RecordCountry --> Profile

  Profile{profile.md<br/>exists?} -->|no| Extract[Resume extraction —<br/>wait for upload and confirmation]
  Profile -->|yes| State
  Extract --> State

  State{.salary-calculator-state.json<br/>exists?} -->|yes| Resume[Inform user: resuming from step N]
  State -->|no| S1
  Resume --> S1

  S1[Step 1: Research prompt generator<br/>local-market salary prompts per country] --> S2
  S2[Step 2: Data ingestion<br/>paste research results] --> S3
  S3[Step 3: International adjustment<br/>deep-reasoner agent] --> S4
  S4[Step 4: Final table calculation<br/>calculator agent] --> RCQ

  RCQ{User wants<br/>reality check?} -->|no| Done([Results delivered])
  RCQ -->|yes| S5[Step 5: Reality check<br/>deep-reasoner agent]
  S5 --> Done

Steps

Profile check

Checks for profile.md in the workspace. If absent, reads prompts/shared/resume-extraction-prompt.md, waits for the user to upload their resume, and waits for explicit confirmation of the extracted profile. The profile is reused on all subsequent runs without re-extraction.

Handoff check

If invoked with a single country (handed off from Country Finder), records it as target_country in the state file and skips the country list question in Step 1. If invoked standalone, asks for the target country list in Step 1.

State check

Checks for .salary-calculator-state.json. If found, reads last_completed_step and informs the user which step will resume. If absent, creates the file with last_completed_step: 0 and target_country: null, and starts from Step 1. Updates the file after each step completes.

Step 1 — Research prompt generator

Generates ready-to-copy research prompts for each target country, scoped to the candidate’s role and profile from profile.md. Each prompt instructs the researcher to find realistic local-market annual base salary ranges — excluding expat, FAANG-only, US-skewed, contractor, and equity-heavy data. Prompts request two company tiers (mid-size local-market and premium/international), city-level breakdowns where relevant, and sourced, dated evidence.

Step 2 — Data ingestion

Accepts pasted salary research results one country at a time. Validates each message: one country per message, required fields present, no silent overwrite if a country was already stored. Data is preserved verbatim — no analysis or adjustments during ingestion.

Step 3 — International adjustment

Claude asks whether to use the deep-reasoner subagent (Opus, high effort) for higher reasoning accuracy — if declined, the step runs with your current model. Estimates the international candidate adjustment for each country — the realistic hiring discount an overseas applicant may face compared to a local candidate, based on employer risk perception, visa complexity, remote interview logistics, and local talent availability. Shows reasoning for each adjustment.

Step 4 — Final table calculation

Claude asks whether to use the calculator subagent (Opus, max effort) for higher arithmetic precision — if declined, the step runs with your current model. Reads all ingested salary data and adjustment figures, shows full arithmetic for every country before producing the final table, and double-checks each calculation before finalising. Precision takes priority over speed.

Step 5 — Reality check (optional)

Claude asks before running. If you confirm, Claude then asks whether to use the deep-reasoner subagent (Opus, high effort) — if declined, the step runs with your current model. Audits the final table output for inconsistencies, outliers, or weak evidence. Skipped if the user declines.

Stop conditions

Profile not yet uploaded. Claude waits — it does not proceed or fill in placeholder data.
Any step instructs Claude to wait. Claude stops and waits. No guessing, no assumptions.
Data ingestion receives multiple countries, missing fields, or a duplicate. Claude stops and explains the issue before storing anything.