CrimeStats

• The numbers are for the city the user meant: the named city verified by its state (and county and ZIP for the configured city), or the configured default city when none was named.
• Every reported figure names its source and data year, and the answer flags the caveat that shapes it (source baselines that disagree, FBI lag, felony-only portal counts, partial incident feeds, small-city rate swings).
• Rate answers carry, when available, the total, property, violent, burglary, robbery and vehicle theft rates per 100,000, the YoY change as a percentage, and the percentage above or below the national average, with absolute numbers alongside for small cities.
• Each pull is saved as dated JSON under the Data Storage path, so a later run can compare against it.

• Never reports or stores the names of people arrested, accused or involved in an incident: figures, offenses, dates and block-level locations only.

All city-specific values (default city, county, state, ZIP, neighboring cities, the local police portal URL, and the concrete per-city source URLs) live in the principal's own config, not in this skill. First-time setup and a proof command: Setup.md.

<configRoot>/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/CrimeStats/Config.md configRoot is $CLAUDE_CONFIG_DIR when set, else ~/.claude.

Default city: when no city is specified, use default_city (with its state, ZIP, and population) from Config.md. If Config.md is absent or default_city is unset, ask which city and state to use, and point at Setup.md so the answer can become the default; never guess a city. Load Config.md for the local police portal, neighboring cities, and city-resolved source URLs whenever an intent needs them. Skill files stay generic and describe the procedure against "the configured city."

Neighboring cities for comparison: the neighboring_cities list in Config.md (or, for any other target city, geographically adjacent cities of similar size).

Historical data persists at: ~/.claude/LIFEOS/MEMORY/DATA/CrimeStats/. The directory does not exist on a fresh install, so every intent runs mkdir -p ~/.claude/LIFEOS/MEMORY/DATA/CrimeStats before its first write.

File naming: YYYY-MM-DD_{city}_{state}_{intent}.json

Example: 2026-03-22_springfield_xx_quickstats.json

• Police-portal counts are not FBI counts. Each portal publishes whatever offenses the department chooses (CitizenRIMS, for example, lists felonies only), so read a dataset's offense list before comparing its totals with FBI UCR figures, which include all reported offenses.
• Portal date parameters are vendor-specific. CitizenRIMS wants JS toDateString() form with the correct weekday ("Tue Mar 24 2026") and returns 500 on a wrong one; generate the string, never type it (DataSources.md Tier 0). Socrata and ArcGIS take ISO timestamps.
• FBI annual data lags the calendar year by five to eight months. The national estimates arrive as a First Look around May and the full Reported Crimes in the Nation report around August of the following year; agency-level monthly data has been on the Crime Data Explorer since August 2025. Check the CDE for the latest year before saying a year of FBI data does not exist, and name the data year with every figure.
• AreaVibes vs NeighborhoodScout vs CrimeGrade use different baselines. Rates can differ 10-40% across sources. Always cite the source.
• SpotCrime incident data is not comprehensive. It only shows reported incidents from participating agencies. Absence of data != absence of crime.
• Same-name-different-state confusion is common. Many US cities share a name across states. Always verify you're pulling the configured city in its state (county + ZIP in Config.md), not a same-named city elsewhere.
• Crime mapping sites are often paywalled or bot-protected. NeighborhoodScout limits free access. Fall back to AreaVibes or CrimeGrade.
• State DOJ open-data portals have agency-level data where the state publishes it, but the explorers are client-rendered apps — use the downloadable CSV datasets, not web scraping.
• Per-capita rates swing wildly for small cities. Below roughly 100,000 residents (check population in Config.md or the census figure for a named city) a handful of incidents moves the rate noticeably. Always show absolute numbers alongside per-100K rates.

Example 1: Quick city snapshot

User: "What are the crime stats for my city?"  (or a named city + state)
-> QuickStats intent (default city from Config.md if none named)
-> Rates per 100K, % vs national and YoY change for each category, a safety grade and a one-line takeaway, sources cited

Example 2: Trend analysis

User: "Is violent crime going up here?"
-> TrendAnalysis intent
-> YoY table against neighboring cities, the state and the nation, plus a multi-year table when the data exists and 3-5 key findings

Example 3: Recent incidents

User: "What crimes were reported in my city this week?"
-> IncidentReport intent
-> Incident list with dates, locations, types, a count summary and any patterns

Example 4: Regional comparison

User: "How does my city compare to the neighboring one for crime?"
-> CompareRegion intent (neighbors from Config.md)
-> Side-by-side table of every metric with the national baseline, YoY trends, victimization risk, and the safest city named