Unbroker

unbroker

Find where a person's personal information (name, addresses, phone, email, relatives) is exposed on data brokers and people-search sites, then remove it - automatically where possible, with guided human steps only where a site demands a CAPTCHA, government ID, phone call, or fax. Manages multiple people independently. It does not defeat anti-bot systems, does not act on anyone without recorded consent, and does not remove public records (voter/property/court) or accounts the person controls.

The Python CLI (scripts/pdd.py) owns the deterministic state - config, dossiers + consent, the broker database, tier planning, the ledger, drafts, reports, email sending (SMTP), verification-link polling (IMAP), and the autonomous action queue (next). You (the agent) do the scanning and form-driving with native tools: web_extract and browser_navigate for searching and web forms, and cronjob for recurring re-scans.

Autonomy contract

This skill is designed to run hands-off. After intake (+ recorded consent) there are exactly TWO legitimate human touchpoints: (1) the intake conversation itself, and (2) ONE consolidated human-task digest at the end of the run ($PDD tasks). Between those:

• Never ask the operator to choose configuration. $PDD setup --auto detects capabilities and picks the most autonomous valid config itself.
• Never pause before individual submissions when autonomy=full (the default): the consent recorded at intake is standing authorization for T0-T2 opt-outs. (autonomy=assisted restores per-submission confirmation for cautious operators - honor confirm_first flags in next output.)
• Never interrupt the run for human-only work. Record it (record ... human_task_queued --reason "...") and keep going; it all surfaces once in the final digest.
• Drive the whole run as a loop over $PDD next <subject> - it returns the exact ordered actions to take right now (scan, poll verification, re-check, opt out parents-first, requeue blocked), plus the human digest. Execute every action, record outcomes, re-run next, repeat until done_for_now. Then present the digest, report, and schedule the cron.

The hard limits that autonomy never overrides: no acting without recorded consent, no disclosure beyond disclosure_fields, no CAPTCHA/anti-bot bypass, and confirmed_removed only after a verifying re-scan.

When to Use

• "Remove my (or my family member's) data from data brokers / people-search sites."
• "Opt me out", "delete me from Spokeo/Whitepages/etc.", "clean up after a doxxing."
• "Set up recurring privacy monitoring" (brokers re-list people).
• Checking which brokers still expose someone and why.

Prerequisites

• python3 (stdlib only; no extra packages needed for the core engine).
• Optional upgrades (the skill works zero-config without these; setup --auto turns on every one it detects, reading credentials from the shell env and from $HERMES_HOME/.env so keys Hermes already loads for its own tools are picked up without re-exporting - each one converts a class of human tasks into agent actions):
  • Cloud browser (recommended default): BROWSERBASE_API_KEY. setup --auto selects it whenever the key is present, and it is the intended baseline: a real residential-IP cloud browser clears soft/managed CAPTCHAs (Cloudflare Turnstile, hCaptcha/reCAPTCHA checkbox) as normal operation, so those brokers stay automated (T1) instead of becoming human tasks. This is not CAPTCHA "solving" - no solver service, no fingerprint spoofing; only interactive/behavioral ("hard") challenges the browser genuinely cannot pass fall back to a human task. Without the key, the plain agent browser is used and soft-CAPTCHA brokers drop to T2 (human).
  • Email automation, two credential-free-or-not options:
    • Browser mode (no password): setup --email-mode browser. The agent sends opt-out/CCPA emails and opens verification links through the operator's logged-in webmail using browser_* tools. Nothing is stored. This requires Hermes to be pointed at the operator's own logged-in browser, NOT a cloud browser: a headless cloud browser (Browserbase) holds no webmail session and is itself Cloudflare/DataDome-gated on webmail and on session-bound broker gates (e.g. PeopleConnect guided-mode). Drive the operator's real Chrome over CDP - launch chrome --remote-debugging-port=9222 --user-data-dir="$HOME/.hermes/chrome-debug" (a dedicated debug profile signed into the webmail once, not the Default profile) and connect the browser tools to 127.0.0.1:9222. $PDD cdp launches this for you (finds Chrome/Chromium/Brave/Edge, starts it detached on the dedicated profile, prints the CDP endpoint; --check to test, --print for the command). See references/methods.md -> "Browser backends: scan vs execute". Falls back to drafts for an email if the inbox isn't reachable.
    • SMTP/IMAP (stored creds): EMAIL_ADDRESS + EMAIL_PASSWORD (+ EMAIL_SMTP_HOST / EMAIL_IMAP_HOST for non-mainstream providers; gmail/outlook/yahoo/icloud/fastmail inferred). The CLI sends via send-email and reads verify links via poll-verification. The agentmail skill (per-broker aliases) also counts.
  • Google Sheets tracker: the google-workspace skill.
  • The scrapling skill for stealth/Cloudflare-protected pages.

How to Run

Run everything through the terminal tool. From this skill's directory:

PDD="python3 scripts/pdd.py"

The engine stores data under $PDD_DATA_DIR (default $HERMES_HOME/unbroker), written 0600. Run via terminal, not execute_code (that sandbox scrubs env and redacts output, which breaks reading the dossier).

Quick Reference

Batch operation (two-phase: crawl-all, then delete)

For anything past a couple of brokers, run this as map → reduce → act, not broker-by-broker:

• Phase 1 - DISCOVER (read-only, parallel, idempotent). Crawl every broker first and record a verdict for each (found / not_found / indirect_exposure / blocked). Scanning has no side effects, so it is safe to parallelize and retry. Getting the full exposure map before acting is what unlocks cluster dedup and prioritization below. Default: the parent drives web_extract probes directly - most people-search sites render name/phone/address results as static HTML that web_extract reads in seconds. Escalate to browser_* only for the few JS-only sites, and to delegate_task subagents only for genuinely reasoning-heavy work (large-scale namesake/relative disambiguation). Do NOT hand a browser-toolset subagent a big list of brokers to crawl - in the field this timed out repeatedly (600s, ~5-6 brokers each, no summary) because browser navigation is heavy; the ledger writes that survived came at 10x the cost of parent web_extract. A blocked (DataDome/Cloudflare/antibot) site is not a subagent job either: record blocked and requeue it for a stealth/cloud browser (Browserbase) pass. Subagent reports are self-reports - the parent re-fetches key URLs to confirm a found before trusting it (this cuts both ways: it caught a real listing the parent had wrongly assumed was a false positive).
• REDUCE - $PDD plan <subject> --batch. Collapses the crawl into a phase-oriented plan: groups by next action, collapses ownership clusters (a parent removal that clears children is ONE action, not N - e.g. one Intelius/PeopleConnect suppression covers Truthfinder/Instant Checkmate/US Search/…), and prints next_actions. phase is discover while anything is unscanned, else delete.
• Phase 2 - DELETE (sequential, irreversible). Work the reduced groups parents first: plan --batch orders the found group cluster-parents-first (most children first) and emits a parent_playbook with tailored, ordered steps per parent - follow that order and those steps (full recipes in references/methods.md → "Ownership clusters - DO PARENTS FIRST"). Do the cluster parents (skipping the covered children), re-scan each parent's children after it confirms (they usually drop out), then the standalone listings; send the indirect_exposure cases as CCPA/GDPR delete-my-PII emails (send-email --kind ccpa_indirect), and defer blocked to the stealth-browser pass. Opt-outs hit CAPTCHAs, email-verification loops, and session binding - work them one at a time, carefully (this is the opposite of fan-out), but do NOT stop to ask permission per submission in autonomy=full; in assisted, confirm each one. Usually prefer deletion over suppression where a broker offers both (Spokeo/BeenVerified) - but follow the record's deletion.prefer: PeopleConnect is the exception (prefer: false), where deleting your user data removes your suppressions and does not stop public-records re-listing, so you suppress-and-maintain instead.
• Blind opt-out is the DEFAULT, not a fallback. Submit an opt-out/deletion on every site with an accessible removal channel, even when a listing was not first confirmed - it discloses only the subject's own identifiers to the broker's own official channel, so it does not violate least-disclosure. Two corollaries: (1) a guided flow that matches email+DOB+name and says "no results" is a stronger not_found than any scrape - the opt-out flow doubles as the search; (2) when a form is automation-hostile (hard CAPTCHA, Cloudflare/DataDome, slide-to-verify slider), default to the broker's cited rights-request email (name+state+contact-email only) rather than recording blocked. CAPTCHA policy: never defeat behavioral/token/slider challenges; OK to read a static distorted-text or plain-arithmetic CAPTCHA on the subject's own opt-out, but stop if the site rejects the whole submission after a correct answer (it is fingerprinting the automation). Third-party/indirect records are the exception - still confirm those before acting. Per-site game plans + the meta-search no-op skip-list are in references/site-playbooks.md; the full policy is in references/methods.md.
• PeopleConnect delete-wipes-suppression (permanent rule). A PeopleConnect deletion wipes the suppression and the subject re-lists across the whole affiliate cluster. If a "Your deletion request for PeopleConnect.us is Complete" email ever appears, the suppression is gone -> re-run suppression and re-verify the Control step reads "suppressed". Never leave this cluster on a completed deletion (see references/brokers/intelius.json).

Subagent reports are self-reports: the parent re-verifies key claims (listing URLs, match basis) before recording found and before any deletion.

Procedure (the autonomous loop)

1. Setup (once, no questions). Run $PDD setup --auto - it detects capabilities and configures the most autonomous valid combination itself (programmatic email when EMAIL_* creds exist, Browserbase when its key exists, age encryption when the binary exists, autonomy=full). Then $PDD doctor and show the operator the readiness output for information, not as a question - proceed immediately. Mention what would unlock more automation (e.g. email creds) but do not wait.

2. Intake + consent (the ONE human conversation). $PDD intake ... with --consent (and --consent-method). Without consent the engine refuses to plan or act. Collect everything in one pass - names/aliases, current + prior cities, emails, phones - so you never have to come back with questions. For California subjects, also read references/legal/drop.md: next will surface a drop_submit one-shot that deletes from every registered broker (~545) at once, which is the single highest-leverage action. File it, then drop <subject> --filed. For non-CA subjects the registry is covered by targeted CCPA/GDPR emails (registry --search, then send-email); the people-search sites are worked directly in either case.

3. Drain the queue. Loop:

   while true:
     q = $PDD next <subject>
     if q.actions is empty: break
     execute EVERY action in order; record each outcome via $PDD record
   

   next emits, in order: refresh_brokers (stale cache), fanout_scan/scan_inline (Phase 1 crawl - see step 4), poll_verification (in-flight email confirmations), verify_removal (due re-checks), optout_web_form/optout_email_send (Phase 2, parents-first with playbook steps), indirect_email_send, and stealth_rescan. Human-only work never appears as an action - it accumulates in q.human_digest. In autonomy=full, execute actions without pausing; honor confirm_first in assisted mode.

4. Scanning (when next says so). For fanout_scan: run $PDD fanout <subject> and spawn one delegate_task subagent per batch, in parallel, passing that batch's ready-made brief - do not scan all brokers yourself sequentially. For scan_inline: scan the few brokers yourself. Either way, each broker gets every search_vectors entry via the references/methods.md ladder (web_extract → site: probe → browser_navigate → scrapling), a 404 is INCONCLUSIVE (not not_found), blocked is recorded when antibot is set and no stealth browser is available, and subject vs namesake/relative is confirmed before recording: $PDD record <subject> <broker> <found|not_found|indirect_exposure|blocked> --found <bool> --evidence '{"listing_urls":[...]}'. The parent re-verifies key found claims from subagents before trusting them.

5. Opt-outs (when next says so). Actions come pre-ordered parents-first with steps from each broker record's own optout.playbook (field-verified; cluster parents like PeopleConnect, Whitepages, BeenVerified, Spokeo have exact, live-checked recipes). Deletion usually beats suppression: when an action carries prefer_deletion, complete the record's DELETION lane, not just the hide-my-listing flow. When it carries prefer_suppression instead (PeopleConnect - deleting removes your suppressions and does not stop re-listing), do the suppression flow and keep it maintained; use their Delete button only for a deliberate data-purge. Per method:

   • web_form → drive optout_url with browser_navigate/browser_type/browser_click, submit only disclosure_fields, screenshot the confirmation, then the action's after record command. Playbooks may end with a right-to-delete send-email follow-up - do it (full erasure, not just listing suppression).
   • email → $PDD send-email <subject> <broker> --kind <ccpa|gdpr|generic> --to <addr> --listing <url> records + discloses in one step (recipient locked to addresses the broker record declares; next picks the kind from residency - never claim CCPA/GDPR for someone who can't). In browser mode it returns a recipient-locked compose payload: compose a new message to compose.to with compose.subject/compose.body exactly in the operator's webmail via browser_* and send (no password); in programmatic mode it SMTP-sends. next also routes human-gated forms (phone-callback/gov-ID) through a broker's deletion email when one exists - the rescue lane (verified Whitepages pattern). Draft-only falls back to render-email + a digest entry.
   • captcha → soft/managed challenges clear automatically on the default cloud browser (proceed as normal); only a hard interactive/behavioral challenge it can't pass is recorded blocked (requeued for the stealth/operator-browser pass). Never a solver service.
   • phone_callback / account / gov_id / fax / mail / voice (T3) without a deletion email → never an agent action; next already routed these to the digest. Record them: $PDD record <subject> <broker> human_task_queued --reason "...".

6. Verification (when next says so). In programmatic mode $PDD poll-verification <subject> finds arrived confirmation links via IMAP (anti-phishing scored, auto-advances state). In browser mode, open the broker's confirmation email in the operator's webmail and run $PDD verify-link <subject> <broker> --text '<body>' to score the link. Either way open the link in the same browser (several brokers bind the verification session to the browser that opens it), finish the flow, then record awaiting_processing. confirmed_removed ONLY after a verifying re-scan shows the listing gone - never off the submission flow's own confirmation page.

7. Wrap up (once per run). When next returns no actions: present $PDD tasks <subject> (the consolidated human digest) if non-empty, then $PDD status <subject>; if the Sheets tracker is on, append $PDD report <subject> --sheets rows via the google-workspace skill.

8. Schedule the next wake-up. next returns next_wake_at (earliest due re-check). Create ONE cronjob that re-runs this skill's loop for the subject (a prompt like: "run the unbroker loop for : $PDD next and execute all actions"). Processing windows, verification polls, and reappearance sweeps all flow through the same queue, so the case keeps advancing with zero human attention.

Pitfalls

• Never disclose more than the broker already shows. Submit only disclosure_fields. The engine never volunteers SSN/ID numbers; you must not either.
• No consent, no action. The engine enforces this; do not work around it to "research" a third party.
• send-email is idempotent + rate-limited. It refuses to re-send a case already submitted or beyond (use --force only if a genuine re-send is needed), and SMTP sends are paced by email_min_interval_seconds (default 20s) with retry/backoff. Do not loop it to "make sure" - a successful SMTP handoff is not proof of delivery; the due-queue re-scan is the real confirmation.
• Ledger writes are locked. Concurrent runs (cron + manual) serialize safely; if you ever see a lock timeout, another run is mid-write - let it finish, don't delete the .lock by hand.
• Autonomy ≠ improvisation. Full autonomy means not asking between steps; it does not loosen any gate. If a broker demands MORE than the planned disclosure_fields mid-flow, stop that case and queue it (human_task_queued --reason) rather than deciding alone to disclose extra PII.
• Don't interrupt the run with questions. Config choices are setup --auto's job; human-only work goes to the digest. The only mid-run question that's ever warranted is a missing-identity fact that blocks scanning (e.g. no city at all) - and that should have been collected at intake.
• Use terminal, not execute_code for pdd.py (secret scrubbing + output redaction break it).
• Dossiers are plaintext by default (JSON, 0600 under HERMES_HOME). For at-rest encryption run $PDD setup --encryption age - it generates a local age key and encrypts dossiers + ledgers (the audit log holds field names only and stays plaintext). It guards casual/backup/commit exposure, not a full-HERMES_HOME read; set PDD_AGE_IDENTITY to a separate volume for real key separation. $PDD doctor shows whether encryption is actually engaged (not just whether age is installed).
• "Hidden from free search" ≠ deleted. Only mark confirmed_removed after verifying the record is actually gone; note paid-tier retention in the report.
• Soft CAPTCHAs clear by default; don't fight the hard ones. The default cloud browser passes managed/soft challenges as normal operation (those brokers stay T1). For a hard interactive one it genuinely can't pass, record blocked and let the stealth/operator-browser pass take it - never a third-party solver service or fingerprint spoofing.
• Broker pages change. If a flow breaks, $PDD record ... blocked and flag the broker file in references/brokers/ for re-verification instead of guessing.
• Verify non-field-verified records before submitting. confidence: auto records came from parsing BADBOOL (read optout.notes/optout.links, confirm the real opt-out URL). confidence: documented records (several people-search sites) carry the correct published opt-out URL but have not been field-verified (they 403 datacenter IPs), so confirm the live flow via the operator's residential browser on first use, then set last_verified. Field-verified curated records (no confidence, e.g. the cluster parents) have checked mechanics and take precedence.

Verification

• scripts/run_tests.sh tests/skills/test_unbroker_skill.py (hermetic; no network), or the dependency-free runner python3 tests/skills/test_unbroker_skill.py.
• Dry run: $PDD setup --auto && $PDD doctor && SID=$($PDD intake --full-name "Test Person" --email t@example.com --consent | python3 -c 'import sys,json;print(json.load(sys.stdin)["subject_id"])') && $PDD next "$SID" and confirm a readiness summary plus an ordered action queue.

Source: NousResearch/hermes-agent → optional-skills/security/unbroker/SKILL.md