salary-research

name	salary-research
description	White-hat salary benchmarking for a (geography, role) pair. Cached at cache/salary-<geo>-<role>.md so repeat calls don't re-fetch. Public sources only — postings, Levels.fyi, Glassdoor, Payscale, BLS/Eurostat, SEC, Blind/Fishbowl as signal. Confidence-tagged per data point.
disable-model-invocation	false
allowed-tools	Read, Write, Bash(mkdir *), Bash(test *), Bash(date *)

Salary Research

Compensation ranges for a target role in a given market, using only publicly available data. Cached so you don't waste a search budget on the same query.

Inputs

$ARGUMENTS:

• Required: --role=<title> and --geo=<market> (e.g. --role="staff-pm" --geo="tel-aviv", or --geo=remote-emea).
• Optional: --seniority=<junior|mid|senior|staff|principal> — disambiguates if role doesn't already imply.
• Optional: --currency=<CCY> — default = market default (USD for US/remote, EUR for EU, ILS for Israel).
• Optional: --company=<slug> — narrows to a specific employer.
• Optional: --refresh — ignore cache and re-fetch.

Examples

/career:salary --role="staff-pm" --geo="tel-aviv"
/career:salary --role="ai-researcher" --geo=remote-emea --refresh
/career:salary --role="solutions-architect" --geo="dublin" --company=stripe

White-hat rule

Use only:

• Public job postings (employer sites, LinkedIn, Indeed).
• Salary platforms (Levels.fyi, Glassdoor, Payscale, Salary.com, LinkedIn Salary Insights).
• Government wage data (BLS, Eurostat, national stats).
• SEC / regulatory filings (exec comp).
• Anonymous community sources (Blind, Fishbowl) — signal, not fact.
• Published comp surveys.

Never impersonate recruiters, submit fake applications, scrape behind auth, or pressure contacts to leak bands.

Procedure

1. Resolve cache path

${WORKING_FOLDER}/cache/salary-<geo-slug>-<role-slug>.md. If exists and --refresh not set and mtime < 90 days, read it and short-circuit to step 5 (with a "cached" note).

2. (Optional) Routing decision

If config RESEARCH_MODEL != claude, call research-router --task=salary-research and dispatch the bulk fetch to the routed model.

3. Research

For the (role, geo) pair:

• 3+ salary platforms where possible.
• Capture base / bonus / equity separately when distinguishable.
• Per-company level equivalencies where relevant.
• Geographic adjustment if source data is in a different market.
• Confidence per data point: Confirmed (employer/regulator) | Reported (platform aggregate) | Inferred (triangulated).

If --company set, narrow to that employer plus 3-5 peers as comparison.

4. Write findings

Cache file shape:

# {{role}} — {{geo}} salary benchmark

**Generated:** YYYY-MM-DD
**Currency:** {{CCY}}
**Sources used:** {{list}}

## Summary

p25 / p50 / p75 / p90 total comp: {{...}}

## By company

### {{company}}
- source: {{url}}
- confidence: Confirmed | Reported | Inferred
- | level | title | base | bonus | equity (annual) | total |
- notes: signing / refreshers / vest schedule.

## Caveats

- {{any market-specific notes, e.g. ILS volatility, EU equity rarity, etc.}}

5. Print summary

cache: cache/salary-<geo>-<role>.md  ({{cached|fresh}})
range: {{p25}}–{{p75}} ({{currency}})
sample: {{N companies, M data points}}

Guardrails

• White-hat only. Don't roleplay as a recruiter or applicant for data.
• Tag confidence honestly. "Inferred" beats a fake "Confirmed."
• Cache hits don't get a refresh — explicitly require --refresh. The user is paying for the search budget.
• If geo is "remote", state which timezone bands the data covers; remote-PT and remote-EU pay differently.

Failure modes

• Insufficient data (< 3 sources) → still write the cache, but mark confidence: low at the top and recommend /career:research-company for specific targets.
• Currency conversion ambiguous → use one clear CCY and note the exchange-rate date.