by k7cfo
Delete yourself from data brokers for $0. An agent skill: paste the repo URL into Grok Bot, Claude, Codex, Cursor or any coding agent.
From $49 · 30 days · terms & refunds
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM lists this project for discovery and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
Automated security scan — not a certification.
# Add to your Claude Code skills
git clone https://github.com/k7cfo/remove-your-dataGuides for using ai agents skills like remove-your-data.
Free, first-party takedown of people-search listings and data-broker records. Log every step so a regulator, letter service, or attorney can use it later.
This is an operations playbook, not legal advice.
Canonical repo: https://github.com/k7cfo/remove-your-data (author: Nick Newsom (@k7cfo), Apache-2.0) — improvements welcome as PRs; forks are welcome if NOTICE is kept. Boot: AGENTS.md.
python3 scripts/ryd.py recap --workspace "$WORKSPACE". Read it fully. The recap and verb output are the only source of truth: if it is not in the ledger, say "not recorded"; never say a listing is gone or filed unless the log shows it.python3 scripts/ryd.py report --workspace "$WORKSPACE" --image writes a PNG status card (path printed); attach it in chat together with report (markdown) output. If the host cannot show images, send report --text in a code block. --json is for other agents.The localhost dashboard (serve) is an optional read-only view for people who want it. Do not make it the way you report.
Data brokers collect, scrape, and sell names, phones, addresses, relatives, and inferred profiles. People-search sites publish that dossier to anyone who types a name.
Paid "delete my data" services (Incogni, DeleteMe, Optery, and the rest) mostly submit the same free opt-out forms this skill files, then charge a subscription because brokers re-scrape and re-list. Recurrence is the product.
The conflict is structural:
This skill does not pay those services. It exercises the person's own rights (CA DROP, CCPA/CPRA, GDPR Art. 17, and the site's official opt-out), keeps a legal log, and repeats on a clock because listings come back.
If they already pay a removal service: do not buy more. Do not blindly re-file every private broker it claims to cover. Still file the public indexed people-search sites yourself — those are what Google shows, and paid plans often leave them up.
consent_basis is unconfirmed.self (primary), parent_of_minor, or authorized_agent (they asked you to). Filing a family member's request in their own name from their own inbox (household inbox, §1) is the default for family; file with the person's knowledge.blocked, move on.keep_host / keep_url (LinkedIn, news, personal site, interviews). Default: skip articles, LinkedIn, and other non-broker coverage even without roster rows.references/dead-urls.md. Load STATUS.md with the queue. defunct = skip.This playbook does not require Python, a browser tool, a mailbox API, or a cron. Those help. Missing one is not a stop.
Detect first. Before intake, list what this host already has: python3, a writable folder, a browser you can drive, search, a mailbox you can read, a scheduler. Do not install software they did not ask for. If you can list tools yourself, skip intake Q8 and record the list. If you cannot tell, Q8 is that one question.
For each gap, one offer (1:1), then degrade:
ryd.py, a browser MCP / Playwright, AgentMail, the host’s scheduler.| Need | If you have it | If you do not |
|---|---|---|
python3 + scripts/ryd.py |
SQLite workspace, /roster, dashboard |
Paper log: copy templates/paper-log.md outside the clone. Same facts, no dashboard. |
| Writable folder | $WORKSPACE with evidence |
Coach lines. They paste a notes file back next session. |
| Browser you can drive | You file the official form | Coach: official URL, fields already on the listing, what success looks like. They paste the confirm back. One Turnstile attempt still — they must not loop either. |
| Search | Query pack in references/search.md |
They paste listing URLs / SERP they already see. Do not invent URLs. |
| Mailbox you can read | You complete verify/OTP | They click the link. You never ask for their password. |
| Scheduler | Host cron / heartbeat / systemd | Exact re-invoke phrase + next due date. No daemon. |
Do not fake a filing from a homepage curl. Do not refuse the job because the harness is “too small.” File pass / verify pass / allowlist / dead-urls still apply.
Aliases, addresses, phones, and family members live in the workspace, not in chat memory. SQLite + /roster when Python works. Otherwise templates/paper-log.md. Every later pass reads the roster and scans each consented person separately.
python3 scripts/ryd.py init --workspace "$WORKSPACE"
python3 scripts/ryd.py serve --workspace "$WORKSPACE"
# http://127.0.0.1:8765/roster
python3 scripts/ryd.py add-ident --workspace "$WORKSPACE" --person-id 1 --kind alias --value "Janie Public"
python3 scripts/ryd.py add-person --workspace "$WORKSPACE" --name "Alex Public" --relationship child --consent parent_of_minor
python3 scripts/ryd.py pack --workspace "$WORKSPACE"
On first invocation, if the roster is empty or the primary has intake_complete=0, run intake 1:1. Ask exactly one question, then stop and wait. Do not paste this list. Do not bundle a second question, a menu, or “anything else?” in the same turn. If the answer is unusable, rephrase that question once — still nothing else.
Resume the next unanswered item on the next turn. If they volunteer later answers early, store them and skip those numbers; still ask the next gap, one at a time.
Prefer /roster only after they have a primary person or they ask to paste long lists (many aliases, old addresses). Chat 1:1 for the required set.
Primary person, required, this order:
dedicated / personal / maxkeep_host / keep_url): LinkedIn, news, personal site, interviews — or “defaults”Then, still 1:1: aliases, prior addresses, extra phones, family. Optional: DOB / MAID / VIN only if DROP or a form needs them. They may say “skip” or “I’ll use /roster”.
After Q2 you know residency; still finish Q3–Q9 before searching or filing. Write person + identifier as soon as you have name + residency (so a crashed turn does not lose it). intake_complete=1 only after Q9. Confirm they are not asking you to impersonate.
Family: add as another person row, not extra identifiers on the primary. Set relationship and consent_basis:
| consent_basis | Meaning | Scan? |
|---|---|---|
self |
Primary subject | yes |
parent_of_minor |
User is the parent; child is a minor | yes |
authorized_agent |
Adult family member (or another adult) the user files for; family defaults to household inbox | yes |
unconfirmed |
Listed but not authorized | no |
Family members get a short 1:1 intake: name and their own email, one question per turn. Set relationship and consent_basis from what the user already said (spouse/partner/adult family → authorized_agent, their kid → parent_of_minor); do not stop to collect a consent date, method, or inbox confirmation.
Most households file for a spouse/partner, an adult family member, or their kids. The recommended default is household inbox: the helper files the family member's own request, in that person's name, from that person's own email inbox. Broker opt-outs and privacy requests send verify links and OTP codes to the address on the request, so they land in an inbox the helper can already open and the flow completes without forwarding. It is the person's own request, so it is not authorized-agent filing and state authorized-agent limits do not apply to it.
Nothing to set up: law-plan / law-request file a family member's request this way by default (filing: household inbox), using the email on their roster row as the reply address.
email_used. Do not reuse it for another person.ryd.py record-filer --mode authorized-agent, or --mode guardian for a minor's guardian letter). It keeps each state's authorized-agent limits and unverified warnings, and the letter says an agent is sending it. Roster adults with relationship other default to it. record-filer is optional; use it only to switch modes or leave a note.On every later run:
SELECT * FROM person WHERE active=1 AND consent_basis != 'unconfirmed'ryd.py pack --person-id (or identifier where scan=1) → search packperson_id. Own DROP. Own verify email. Own clocks.Cap: 12 people, 40 identifiers each. Pause an identifier with scan=0 instead of deleting history.
Pick the primary legal track from residency. Always still do public people-search opt-outs — registries do not remove Google-indexed profile pages.
| Residency | First move | Clock |
|---|---|---|
| California | CA DROP then public opt-outs | Brokers process DROP at least every 45 days; status may take 90 days. CCPA deletion otherwise 45 days (+45). |
| Other US state or DC | law-plan → law-request (state pack if one is in effect and residency is attested; else the universal voluntary request) + public opt-outs. TX / OR / VT also have data-broker registries (not a one-shot DROP). |
From the pack, counted from the controller's receipt date. No pack → no legal deadline. |
| Vermont | Registry lookup + individual broker requests | Per broker / VT law |
| EU / EEA | GDPR Art. 17 erasure to each controller | 1 month, extendable to 3 months with notice |
| UK | UK GDPR erasure | 1 month, extendable |
| Canada | PIPEDA access/deletion to each institution | ~30 days |
| Australia | APP access/correction/deletion | ~30 days |
| Anywhere else | Official opt-out forms + any local statute they name | Site-stated window; if none, 30 days |
Load references/jurisdictions.md when present.
California extras (do these first, in order):
DROP does not delete first-party accounts the person created themselves, exempted data, or some publicly available records. People-search pages can still rank. File them.
Other US states and DC (not legal advice; the ledger records what was assumed):
python3 scripts/ryd.py attest-residency --workspace "$WORKSPACE" --person-id N --region CO --confirm. Ask the person; never infer it from a listing.ryd.py law-plan --workspace "$WORKSPACE" --person-id N [--rights delete,opt_out]. It names the pack (laws/packs/us-xx.json, 20 states as of 2026-10-29) or us-universal and why (no pack, not yet in effect, stale, superseded, no attestation, narrow law).ryd.py law-request --workspace "$WORKSPACE" --person-id N --controller "Broker" [--out exports/req.txt]. The exact text, its sha256, the pack version and the assumptions are stored. Send it yourself; it does not send.ryd.py law-event --workspace "$WORKSPACE" --request-id R --event received --date YYYY-MM-DD. Only then do statutory clocks open. Also --event extended (their extension notice), --event appeal (render an appeal where the law has one), --event closed.Rules the verbs enforce: wording says "To the extent your organization is subject to …" and never claims coverage; Florida's law reaches few companies, so it uses universal wording unless you pass --controller-covered; Iowa and Utah deletion covers only data the person provided, so the letter also asks voluntarily for the rest; authorized agents are blocked for rights the pack says they cannot use (most states: opt-out only); no appeal deadline is ever set for the consumer ("reasonable period" stays text). California residents: DROP first, then law-request for businesses DROP does not reach.
Create a workspace outside this repo (PII never belongs in git).
Python available:
$WORKSPACE/
takedown.db # SQLite source of truth
evidence/YYYY-MM-DD/ # screenshots, raw confirm pages, .eml
exports/ # CSV / JSON dumps for a lawyer
python3 scripts/ryd.py init --workspace "$WORKSPACE"
(scripts/schema.sql is applied for you. chmod 700 workspace, 600 db.)
No Python: copy templates/paper-log.md to $WORKSPACE/roster.md and keep action-log.csv / clocks / leftovers there. chmod 700 if you can. CSV is the system of record on that host. If they later get Python, ryd.py init and copy the rows in — do not keep two sources of truth.
SQLite beats CSV when you have it: unique listing URLs, due-date queries, request IDs, joins across person / broker / clock / email. CSV is an export from SQLite, or the whole log when SQLite is impossible.
On every real action insert action_log (SQL or paper) with:
occurred_at_utc and occurred_at_local (person's timezone from intake)agent, user, family)After each filing, insert or refresh a clock row (verify email, site drop window, legal response, DROP 45/90). When a listing becomes gone, insert relist_90d (~90 days). Closed is not forever.
Export when asked, or before escalation:
sqlite3 -header -csv "$WORKSPACE/takedown.db" \
"SELECT * FROM v_evidence_chronology ORDER BY occurred_at_utc" \
> "$WORKSPACE/exports/evidence-log.csv"
Keep leftovers in table leftover (or leftovers.md on paper): URL, what PII is still shown, how you found it, where it is stuck, next step.
After each pass, refresh the running report if Python works:
python3 scripts/ryd.py serve --workspace "$WORKSPACE"
# http://127.0.0.1:8765/ — loopback only
python3 scripts/ryd.py export --workspace "$WORKSPACE"
Report in chat first (see "Session start and reporting"): run python3 scripts/ryd.py report --workspace "$WORKSPACE" --image and attach the card. Mention the optional read-only dashboard URL only when serve is running. Do not bind it off localhost. Paper log: the markdown/CSV is the report. On DROP submit, set drop_filed in SQL config or on the paper roster (store the DROP ID in action_log.request_id, not in git).
Brokers send confirmation links and OTP codes. Prefer a mailbox the agent can read. If none: coach — they click the link and tell you it worked. Never ask for their password.
Prefer this order:
+ aliases. Keep one real Gmail/Yahoo/Outlook for those sites only.Store which mailbox was used per listing (email_used). Never reuse a one-per-person address across two people.
Family members: file from their own inbox (household inbox, §1). Verify links and OTP codes then arrive where the helper can read them, and the request is in the person's own name. Otherwise use a dedicated inbox for that person, or coach them to click.
When a verify email arrives: fetch it, extract the link or code, complete the flow, log email_event, recheck the public URL the same day.
If the code went to a personal inbox you cannot read: tell them the exact subject / sender to look for (including spam). Do not ask for the account password.
Goal: the takedown does not create a new dossier.
Default dedicated:
chmod 700, db chmod 600, never committed.personal: allowed to use their real email and a logged-in browser when a site demands it (Whitepages phone-verify, Login.gov for DROP). Still keep the SQLite workspace private.
max (harder, some filings will fail):
SERPAPI_API_KEY is not set, prompt them: (1) sign up at https://serpapi.com/users/sign_up (email + phone verify, free 250 searches/mo); (2) copy the key from https://serpapi.com/manage-api-key; (3) provide it as SERPAPI_API_KEY (env/secret store; never print it). Option: SerpApi's official MCP (github.com/serpapi/serpapi-mcp, https://mcp.serpapi.com/mcp with Authorization: Bearer <key>). Declined → skip. SearchAPI.io is a different service; its keys will not work here. Details: references/search.md.max, expect some brokers to reject AgentMail — log blocked: mailbox rejected instead of falling back unless they authorize it.DROP residency verify is inherently identifying (California Identity Gateway or Login.gov). For CA residents, that is the point. You cannot complete Login.gov 2FA for them — hand them the session at the sign-in page. They cannot switch verify methods after they pick one. A regular myDMV login is not Identity Gateway.
VPN: optional. If captchas explode, try without VPN or a residential path. Do not hammer.
Detect what this host already has. Do not install software they did not ask for.
Prefer a real browser over curl. People-search opt-outs are JS, email-verify, and bot-walled.
Pick the first that exists:
OMP-style pattern (adapt to whatever tool names this host exposes): open the URL, wait for DOM, take an accessibility snapshot, act on the form, screenshot the confirmation, save that file under evidence/. Do not fake a successful opt-out from a curl of the homepage.
If the form is Turnstile-blocked from this IP: one attempt, log blocked, leftover it. Do not loop.
Run the same query pack across several engines. Catalog and query pack: references/search.md. ryd.py pack prints roster-backed queries when Python exists; otherwise type the pack from the roster by hand.
Preference (use what exists, skip the rest):
anonymity_mode=max: no third-party search API unless they said yes.
When a URL matches this person (same phone, street, or city): insert listing before filing.
After opt-outs, leftovers that still rank on Google go to §7a only after the source URL is down. That is hide-from-index, not delete-at-source.
Two named passes. Do not collapse them.
File = official opt-out + verify email/call. Snapshot the public page first. Save request ID. Complete verify. Insert clock for the site's stated window (else 7 days, then the legal clock). One clean filing plus verify per listing. Recheck later instead of spamming the form.
Verify = re-fetch the exact public listing URL after the stated window. Confirm-page text is not success. Origin TLS/timeout/500 is not a drop if search snippets still show the identifier. Curl Cloudflare and a browser fetch can disagree — do not mark drop from one transport. Hourly CF after a recent live reading is opacity, not a drop.
Closed only when the exact URL is gone or the identifier is gone from it. Then start relist_90d.
Official form only. Prefer email-verify flows you can complete. Solve captchas if they authorized that.
Load references/brokers.md as a queue, STATUS.md as status, references/dead-urls.md as skip-without-loop. Opt-out URLs rot — verify the live privacy /opt-out page before submitting. defunct / dead-url rows: skip. Send URL fixes upstream.
Priority order:
references/brokers.md whose STATUS is not defunct that matched.Per listing (file pass, then later verify pass):
clock for the site's stated window (Spokeo often 24–48h; otherwise 7 days, then the legal clock).Site notes that keep biting:
/optout. Keep optout_request_id. Uncommon domains (including some AgentMail) are rejected; use the consumer fallback inbox. Verify links can land on an empty /optout; do not loop. Email privacy@spokeo.com and watch 24–72h. A Zendesk agent may remove it; do not reply in that chat.+ aliases collapse; custom domains often rejected. Quoted-printable mail can corrupt token= (=11 → control char); reconstruct from the raw .eml if the link 404s. Confirm page can lie; recheck the public URL. optout@ can be null-MX dead; prefer /contact subject Other. Vehicle/VIN cards can survive; after the stated window, letter (they may publish no street address, only a WHOIS privacy proxy). Do not refile during the window. If the origin parks (registrar marketplace, people and privacy URLs 404), cancel the letter; leftover SERP snippets wait until other live people-search listings are down. Confirm parked with a browser plus RDAP/WHOIS, not curl timeout alone.property.whitepages.com) is a separate product — people-profile suppression does not take property pages down. Whitepages address pages (whitepages.com/address/…) are another separate leftover from Property Intel — SERP-hiding or suppressing one does not take the other down; recheck each exact origin URL. Automated /suppression-requests rejects property URLs; email privacyrequest@whitepages.com with the exact property URL. Zendesk auto-replies can list many state privacy laws while omitting California; reply once on the same thread asserting CCPA / CPRA / California Delete Act. A same-thread follow-up can mint a new Zendesk request number; treat it as the same request, watch both. Do not reply only to acknowledge an auto-ack (empty ack spam). Do send one substantive same-thread follow-up when all of: the claimed reply window (~2 business days) has passed; the exact property URL is still live with owner/resident identifiers; and there has been no human resolution — cite the still-live URL + delete request under CCPA/CPRA. Escalation ladder: (1) file privacyrequest@ with the exact property URL → (2) recheck origin after ~2 business days → (3) if still live + only auto-acks, one substantive same-thread follow-up → (4) if still live after ~14–30 calendar days of vendor silence, use letter / templates/regulator-complaint.md / attorney packet (do not auto-send). Queue ID upload only when they demand ID; do not send government ID alone from the agent. Do not open a brand-new ticket for the same URL unless they close the thread and instruct a new request. A Zendesk human may claim they could not locate any listings / the listing may already be removed and ask for a screenshot, while Property Intel (or another sibling product) is still origin 200 with matching owner/address identifiers — agents often search people profiles only. Do not mark the leftover closed. Reply on the same thread with the exact still-live property URL, what identifiers remain, and a dated screenshot of that page. A partial scrub (residents or a landline gone, FAQ says no current residents) while owner-on-record / trust name / street address remain is not a full drop; keep the leftover open and the escalation clock running. Stale SERP can still show scrubbed residents..com and .net are different hosts. .net opt-out can be a Google Form that 401s unless the browser is signed into a consumer Google account. Do not invent another email./opt-out: name + email + captcha, then emailed form link (~3 days). Captcha is human; do not loop. A relative/spouse page can reprint the subject's phones even when the subject is not named — sweep by phone, not only by name. A processed USPhoneBook opt-out can also clear the same person's FastBackgroundCheck people page (and drop them from FBC address/relative pages) within ~3 days; recheck FBC after a USPhoneBook processed email before starting FBC's captcha flow./optout is email-verify plus Turnstile; datacenter checkbox often never ticks (one attempt, do not loop). Fallback: support@contactout.com. After SENT / support@ fallback, if a verify link arrives, complete it from a residential / consumer browser session the person controls (datacenter agents often cannot finish Turnstile). Profile FAQs can reprint full phone and work email. Origin CF/timeout is not a drop. Hourly CF vs 200 with the same identifiers is opacity, not a drop. If the public profile or FAQ still reprints phone/email after the legal clock, escalate via letter / templates/regulator-complaint.md — do not loop Turnstile./do-not-sell-info. A support@ CCPA email can confirm expunge while the public URL stays Cloudflare-blocked for up to ~48h. Do not refile or loop Turnstile after a written confirm. After that window, a Google name-index hit (/name/First-Last/in-ST) can still show PII in the snippet while origin is Cloudflare. Different URL than the UUID profile. Do not refile. SERP hide is last-step after remaining live listings are down./people/…. Official /optout.html is Turnstile; datacenter checkbox often never ticks (one attempt, do not loop). Footer “Do Not Sell My Info” is the same page. Fallback: support@nationalpublicdata.com with the live unique profile URL from a consumer inbox. SENT-only until origin 404s or they reply. A support@ confirm can cover unique and name-index; still watch 24–72h for reappear. Do not file SERP while this origin or other leftover people-search origins are still live./removal is Turnstile from a datacenter (one attempt, do not loop). If the subject has no detail/PID with matching identifiers but a household relative does, do not invent a subject opt-out; household only if easy.?page=N) for the name and confirming the person's card is gone from all of them.If a site has no form: send templates/erasure-request.md to the privacy contact, start a legal clock, log the message-id. After the legal clock with the leftover still live: templates/regulator-complaint.md. Do not auto-send.
Refresh references/brokers.md from the live California data-broker registry CSV with python3 scripts/registry_pull.py (Vermont/Oregon/Texas when a bulk file exists). Dedup by hostname against this queue. DROP still covers registered CA brokers for CA residents. Public people-search pages are still filed one-by-one. Do not POST a generic form to every registry row.
Broker opt-out first. Do not file Google/Bing/Yahoo/DDG removals until the source listing is down (404/410/gone). A Google Results-about-you or SERP ⋮ Remove approval for one URL is not an origin drop for that URL or sibling broker pages; recheck the live page. A Cloudflare/JS 403 wall after approval is still not a clean 404/410. Stale snippets after a parked/404 origin are expected. If the public URL is still live or a 404 source still ranks on Google, hide the SERP snippet only after leftovers at the source are gone. This does not delete the source. Full steps: references/search.md (Index failsafe).
On each search pass, diff broker/people-search hostnames against $WORKSPACE/exports/serp-hosts.txt (create on first pass). Alert when the person's name+city/phone/street shows up on a new people-search host. Update the snapshot. Do not file against keep_host / keep_url.
Do not use the agent's Google account for someone else's Results about you — Google may deny it. Hand them the session.
Log action=google_serp_remove, channel=google, request ID. Clock: site_window / basis Google Results about you.
Gov, school, and newspaper URLs often have no Remove option. Leave as leftover.
After Google: Bing content removal (feeds Yahoo/Ecosia/DDG). Then Yandex if it still ranks there.
Only if they are a California resident (or filing for one with a permitted authorized-agent path).
Official: https://privacy.ca.gov/drop/ App: https://consumer.drop.privacy.ca.gov/ How it works: https://privacy.ca.gov/drop/how-drop-works/ Registry: https://cppa.ca.gov/data_broker_registry/
Rules:
DROP is ongoing, not one-shot. Keep a clock for 45-day recheck and 90-day status.
This is not a one-sitting job. Listings reappear. Laws give brokers weeks.
Household routine lives in config when Python exists (dashboard gear → /settings, or CLI). Paper log: cadence on the roster table. Read it every pass:
python3 scripts/ryd.py settings --workspace "$WORKSPACE" --show
python3 scripts/ryd.py settings --workspace "$WORKSPACE" --cadence 7d --paused 0 --apply-people
If paused=1, do not search or file. Still process due legal clocks if the user asked. Unpause from Settings or --paused 0. Paper: a paused note on the roster.
Cadence default 168h (weekly). Do not catalog every product. Detect what this host already uses to run a job later, then recommend one specific setup. Same ladder as §0: use it, offer to get one, or coach.
Detect (quietly — do not dump a menu):
systemctl --user, launchctl, schtasks, crontab.Then say, in one short block: the exact re-invoke phrase (the same words they used today), cadence_hours, and the next due clock (UTC + local). Do not install a daemon they did not ask for. Do not invent a second loop beside the host's.
If several exist, pick the one already wired to this agent.
Each invocation:
if tools unknown → §0 detect or Q8 (one question, stop)
if roster empty / primary intake incomplete → section 1 (one question, stop)
read settings (paused, cadence_hours) # or paper roster
if paused → report paused, stop search/file
for each consented person:
process due clocks (overdue first)
check mailbox for verify / OTP, or ask them if you cannot read mail
recheck listing URLs whose window elapsed (**verify pass**) — coach a browser if you have none
if scan is due → pack (ryd.py or hand) → search pack (section 6); diff serp-hosts.txt
file new matches (**file pass**, section 7); skip keep_* and dead-urls; coach the form if no browser
leftovers still in Google SERP → §7a only after source URL is down
CA: if DROP status due → check DROP
skip unconfirmed family
export evidence if anything changed (ryd.py or paper-log)
report per person in chat (ryd.py report --image): gone / pending / leftover / overdue
refresh the optional dashboard only if serve is already running
Clock kinds:
| kind | when to set | when it fires |
|---|---|---|
verify_email |
after filing | every 4–12h until clicked or 7d |
site_window |
after verify | site-stated (else 7d) |
legal_response |
GDPR/CCPA/letter; law-event received for state packs |
30d or 45d from send; state packs: pack days from receipt (business days where the pack says so); warn at 7d before |
drop_45d |
after DROP submit | every 45d |
drop_90d |
after DROP submit | once at 90d |
rescan |
always | cadence_hours |
relist_90d |
listing status gone |
~90 days; brokers re-scrape |
A listing becomes gone only when the exact URL no longer shows the identifier (verify pass). Confirm page is not success. Verified filing + still visible after the window → leftover. Origin TLS/timeout/500 is not a drop if search snippets still show the identifier. Curl Cloudflare and a browser fetch can disagree — do not mark drop from one transport. Hourly CF after a recent live reading is opacity, not a drop. Count people / property / reverse-phone URLs. A bare directory landing is not a listing by itself, but track it when its SERP snippet exposes a matching phone, street, email, or relative even if the origin is Cloudflare-gated. Bing/Yahoo can keep snippeting the original subject on a reverse-phone URL after the live origin has flipped to a different person or a household relative at the same street and number — recheck the live owner name; if the subject is no longer named, clear the subject-primary leftover and do not cite that URL in a suppression follow-up. A household leftover may remain if that relative is in scope and the opt-out is easy; do not restart a PeopleConnect-family four-brand suppression from the reattribution alone. If an unsent human press-Send draft is superseded by an already-sent residual on the same thread, close the pending-send row; do not recreate or nag Send; the person can trash the orphan draft. Keep a separate human-needed tracker for steps only the person can do (signed-in Google Forms, phone robot codes, ID upload when demanded, Results about you). Mark search-engine account removals deferred / last-step — not active do-now — while any people-search origin leftover is still live. After a written vendor confirm, close or demote matching Turnstile / form human rows; do not leave them as pending redo. Keep any ops UI human section in sync with that tracker so stale cards do not look actionable. Closed is not forever; many brokers re-list inside ~90 days — that is relist_90d, not a reason to refile during the window.
Escalation, unless they say otherwise:
templates/erasure-request.md) with the leftovers filetemplates/regulator-complaint.md) using exports/evidence-log.csv. Do not auto-send.unconfirmed family members: show on /roster, do not search or file.keep_host / keep_url: never search-to-file, never SERP-hide.A pass is done when:
ryd.py export / serve is current if Python works; otherwise the paper log is current.Then wait for the next cadence. The work is the loop.
Last scanned: 11/20/2026
{
"issues": [
{
"file": "SKILL.md",
"line": 278,
"type": "secret-exfiltration",
"message": "Instruction appears to send credentials/secrets to an external endpoint",
"severity": "medium"
}
],
"status": "PASSED",
"scannedAt": "2026-11-20T12:32:38.684Z",
"npmAuditRan": true,
"pipAuditRan": true,
"promptInjectionRan": true
}See how remove-your-data compares with popular alternatives.
Delete yourself from data brokers for $0. Just point your AI at it.
Original repo: https://github.com/k7cfo/remove-your-data by Nick Newsom (@k7cfo). If you are reading a copy or a fork, the original is the place for fixes and updates.
An agent skill that removes a person's listings from people-search sites and data brokers without paying a deletion service. It files first-party opt-outs, routes California residents through DROP, keeps a local SQLite log of every request, and reports progress in chat. This is not legal advice.
My actual results (as of Nov 2026). The story behind this: thread on X.
Grok Bot (easiest): Get the Remove My Data bot. One click and it's ready. It walks you through everything.
Any other AI agent: paste this link into Claude Code, Codex, Cursor, Hermes, OpenClaw, or any agent that can read a repo, then say remove my data from people-search sites.
https://github.com/k7cfo/remove-your-data
The agent asks where you live first (California residents start with DROP), files the opt-outs, and keeps the log.
How remove-your-data compares with paid removal services. Prices checked Nov 14, 2026 for one US adult. Paid services are real products with real strengths: they are hands-off, they have human support, and they run large automated broker lists. This project trades that for cost, control, and transparency.
Legend: ✅ yes, ◐ partial, ❌ no, ? not stated by the vendor.
| remove-your-data | DeleteMe | Incogni | Optery | Kanary | Aura¹ | Privacy Bee | EasyOptOuts | |
|---|---|---|---|---|---|---|---|---|
| Cheapest annual plan | $0 | $129 | $95.88 | $39 | $250 | $144 | about $8/mo, billed yearly² | $19.99 |
| Monthly plan | n/a | no³ | $15.98 | $3.99 | $25 | $15 | not shown | none listed |
| Open source, you can read every request it sends | ✅ | ❌⁴ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| Your data stays on your own machine (local SQLite log) | ✅⁵ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| Files from your own email and handles the back-and-forth | ✅ | ❌⁶ | ❌⁶ | ❌⁶ | ?⁶ | ❌⁶ | ❌⁶ | ◐⁶ |
| State privacy-law requests with tracked legal deadlines | ✅ | ? | ◐⁷ | ◐⁷ | ◐⁷ | ? | ◐⁷ | ? |
| California DROP | ✅⁸ | ❌ | ? | ❌ | ? | ? | ? | ? |
| Household or family members | ✅ free | ◐⁹ | ◐⁹ | ◐⁹ | ◐⁹ | ◐⁹ | ◐⁹ | ❌ |
| Follow-up on stale requests | ✅ | ✅ | ✅ | ◐ | ✅ | ✅ | ✅ | ? |
| Phone-call escalation | ✅ on hosts that can call¹⁰ | ? | ? | ? | ? | ? | ? | ? |
| Custom removal for any site you name | ✅ | ◐¹¹ | ◐¹¹ | ◐¹¹ | ◐¹¹ | ? | ◐¹¹ | ❌ |
| Multi-engine leftover scan | ◐¹² | ? | ? | ✅ | ✅ | ◐ | ? | ◐ |
| Recurring re-scans | ◐¹³ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Works inside the AI you already use (Grok Bot, Claude, Codex, Cursor) | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| Brokers covered (advertised) | any site the agent can reach¹⁴ | not stated | 420+ | 375+ to 620+ | highest-traffic sites | 425+ | 597 (people search) | 200+ |
Where paid services win: you do nothing after signup, a person answers when something breaks, and the broker lists are large and maintained for you. Where this project differs: it is free, you can read the code and every request, the log stays on your machine, and it can work through the replies using your own inbox.
Notes
references/brokers.md. Advertised counts from paid services are not directly comparable to that.Price sources, all checked Nov 14, 2026: DeleteMe, Incogni, Optery, Kanary, Aura, Privacy Bee, EasyOptOuts. Prices and plans change; check the vendor before you buy. For independent results on paid services, see Consumer Reports' 2024 study.
If this saved you a subscription, a star helps others find it.
After the opt-outs go through, the agent can check whether listings still show up in search results. That is the optional SerpApi step.
What it is. SerpApi is a search-results API. Its free plan gives 250 searches a month (checked on serpapi.com/pricing, Nov 14, 2026).
What it does. ryd.py serp-check searches for each person's name, name plus city or street, and phone numbers. It uses Google and Bing by default, and can also use Yahoo, DuckDuckGo, Yandex, Baidu, and Naver (--engines google,bing,yahoo,duckduckgo). Each hit prints as one JSON line with the engine, query, URL, title, and snippet. Hits on hosts you told it to keep are dropped.
Optional and watch-only. It is off unless a key is set and you pass --serpapi. It never files anything; removing a stale search result stays the last step, after the original page is gone. Name
remove-your-data is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by k7cfo. Delete yourself from data brokers for $0. An agent skill: paste the repo URL into Grok Bot, Claude, Codex, Cursor or any coding agent. It has 188 GitHub stars.
Yes. remove-your-data passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/k7cfo/remove-your-data" and add it to your Claude Code skills directory (see the Installation section above). remove-your-data ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
remove-your-data is primarily written in Python. It is open-source under k7cfo on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh remove-your-data against similar tools.
No comments yet. Be the first to share your thoughts!