archive-evidence-packs
ChatGPTArchives active evidence packs, moving them out of the active list (unarchive-evidence-packs restores them). For specific packs pass evidencePackIds (from list-evidence-packs). For "archive all / every matching" requests pass filters (same as list-evidence-packs) instead; dryRun: true only reports how many active packs match.
create-evidence-pack-from-url
ChatGPTCreates an evidence pack by scraping a public social post or profile URL, or a public website or news article URL. Only use this tool if the user has explicitly given you a URL to create an evidence pack for. Before calling this tool, warn the user that it may take up to 3 minutes; then invoke it and wait for the result. Pass organizationId from context when acting on an organization scope. Pack URL after creation: https://guardian.umanitek.ai/evidence-pack/{id}.
create-evidence-packs
ChatGPTCreates evidence packs; threats that already have one are skipped. For specific threats pass threats (identifiers from the fetch tools). For "packs for all / every matching" requests pass filters (same as the fetch tools) instead: it creates up to 100 per call and stops when quota and credits run out, and dryRun: true only reports how many match and how many the quota covers. If isComplete is false, report remaining and stoppedReason; call again with the same filters only when stoppedReason is cap, otherwise stop. Each pack is visible on https://guardian.umanitek.ai/evidence-pack/{id}.
create-takedown-requests
ChatGPTSubmits automated takedown requests. For specific evidence packs pass evidencePackIds (use list-evidence-packs to find them); notes apply to every request in the call. For "take down all / every matching" requests pass filters (same as the fetch tools) instead: it reuses each threat's evidence pack or creates one, skips threats that already have a takedown request, handles up to 50 per call and stops when quota and credits run out or the takedown terms are not accepted; dryRun: true only reports how many match, quota left and whether the terms are accepted. If isComplete is false, report remaining and stoppedReason; call again with the same filters only when stoppedReason is cap, otherwise stop. Requester details and threat URLs come from the user's Automated Takedown ToS acceptance and the evidence pack; never ask the user for them. Not available for llm_risk (AI answer) or dark_web packs. If ToS has not been accepted, the user must submit one takedown manually in the app first (one-time ToS form).
delete-evidence-packs
ChatGPTPermanently deletes evidence packs. This is irreversible: always confirm with the user before calling. Packs with a takedown request are always kept and come back in notDeletedEvidencePackIds. For specific packs pass evidencePackIds (from list-evidence-packs). For "delete all / every matching" requests pass filters (same as list-evidence-packs) instead; dryRun: true only reports matched, wouldDelete and wouldKeep.
fetch-account-context
ChatGPTReturns who the signed-in Guardian user is and which organizations and personas they can act on, with the IDs to pass as organizationId and personaId. Use it when the user asks about their account, organizations or personas, or before acting on an organization.
fetch-custom-topics
ChatGPTLists the custom monitoring topics the user has defined. Custom topics are extra names, phrases, narratives, or queries Guardian runs targeted lookups for on every monitoring cycle, on top of the standing impersonation and harmful-content monitoring. Also returns the plan limit and how many topic slots are left. Use this whenever the user asks what Guardian is watching for, and before changing topics with update-custom-topics.
fetch-detection-settings
ChatGPTReturns Detection Settings for the active scope: industry (category), role, region, optional public profile URLs (Instagram, TikTok, X, YouTube, LinkedIn, Facebook), and custom detection context. If the user has organization memberships in context, pass organizationId from Active Scope. Use before update-detection-settings.
fetch-evidence-pack
ChatGPTFetches an evidence pack by UUID including all threat data. Returns the complete evidence pack information needed to generate takedown notices
fetch-evidence-pack-takedownability
ChatGPTGets how strong a takedown case looks for one evidence pack. Explain the result in everyday language. Use list-evidence-packs first if you do not have the pack ID.
fetch-historical-risk-scores
ChatGPTGets historical risk scores (postRiskNormalized and impersonationRisk) for the authenticated user, sorted by generatedAt timestamp in descending order. If user has memberships, pass organizationId from context. limit/offset pages both series together, newest first: postRiskEntryCount and impersonationRiskEntryCount are the totals per series, and hasMore/nextOffset say whether older entries exist; pass nextOffset back as offset for the next page.
fetch-impersonators
ChatGPTLists impersonators (similar/fake accounts) as slim rows: identifier (canonical profile URL), platform, username, riskScore. If user has memberships, pass organizationId from context. Supports text search, date range, platforms, followers, risk score and sortBy/sortOrder (default highest riskScore first). Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page. To act on every matching account, pass the same filters to the action tool instead of paging; to act on specific accounts, pass their identifier.
fetch-llm-risks
ChatGPTFetches what large language models (ChatGPT, Claude, Grok, Gemini, Perplexity) say about the user, plus the combined LLM risk score, as slim rows: identifier, model, resultType, severity, riskScore, and excerpts of the prompt and response. Covers fabricated claims, verified threats retrieved from the live web, and results for the user custom prompts. Filter by riskLevel or textSearch, and set showOlder for history. Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page. Call this before create-evidence-packs for LLM rows.
fetch-pipeline-runs
ChatGPTAdmin only. Lists the monitoring pipeline runs of one persona, newest first: pipelineType, status, timing, itemsFound, itemsCollected (items the run collected), errorMessage, LLM inputTokens/outputTokens, and spend in USD (tokenCostUsd for LLM calls, scraperCostUsd for scraping, totalCostUsd for both). pipelineType is null and usage is 0 for runs that no longer match the persona's configured pipelines. Filter by pipelineType, status and startedFrom/startedTo to answer a question instead of paging through every run; fetch further pages only when the user needs more rows. Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page.
fetch-posts
ChatGPTLists harmful posts as slim rows: identifier (canonical post URL), platform, username (author), riskScore. If user has memberships, pass organizationId from context. Supports text search, date range, platforms, threatCategories (exact graph values), followers, riskScoreMin/Max, viralityScoreMin/Max and sortBy/sortOrder (default highest riskScore first; sortBy viralityScore sorts by spread). Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page. To act on every matching post, pass the same filters to the action tool instead of paging; to act on specific posts, pass their identifier.
fetch-protection-context
ChatGPTLists the user Protection Context: natural-language automated-takedown criteria and labeled whitelist URLs that must never be taken down. Use before changing Protection Context with update-protection-context.
fetch-risk-score
ChatGPTReturns the final combined risk score (0-100) using weighted formula: postRisk 0.5 + impersonationRisk 0.3 + llmRisk * 0.2. If user has memberships, pass organizationId from context. Set includeHistorical to true for historical data.
fetch-subscription-quotas
ChatGPTFetch the current subscription tier, effective quota limits, and remaining quotas for the active scope (personal plan, or organization persona plan when managing an org). Use when the user asks about their plan, tier, limits, remaining takedowns, evidence packs, deepfake scans, seats, aliases, monitoring runs, or whether they can perform a quota-limited action.
fetch-takedown-guidance
ChatGPTReturns step-by-step takedown guidance for a given platform (e.g. X/Twitter, TikTok, YouTube, Instagram, Facebook, Reddit, OpenAI, Anthropic, Gemini, Perplexity). Use the platform name from the threat data (e.g. "x", "tiktok", "facebook"). Omit platform to get guidance for all platforms. Includes direct reporting categories, form URLs, and notes for content removal requests.
fetch-takedown-request-status
ChatGPTReturns the full status timeline and current status of the takedown request for a given evidence pack. Includes all status history entries with timestamps and descriptions, plus any attached documents. Use this to answer "what is happening with my takedown for X?" questions.
fetch-threats
ChatGPTFetch threats (impersonators and posts) as slim rows: identifier (canonical URL), platform, username, riskScore. If user has memberships, pass organizationId from context. Supports platforms, date range, followers, riskScoreMin/Max, viralityScoreMin/Max (posts), textSearch and sortBy/sortOrder (default highest riskScore first). Each category returns total, returned, hasMore and nextOffset; pass them back as impersonatorsOffset / postsOffset (same filters) for the next page of up to 100 per category. To act on every matching threat, pass the same filters to the action tool; to act on specific threats, pass their identifier.
fetch-true-seal-scan
ChatGPTReads TrueSeal (deepfake) scan results. Pass scanBatchId to check whether a scan started earlier by run-true-seal-scan has finished. Omit scanBatchId to list scans newest first, optionally filtered to one evidence pack. Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page. Use this when the user asks about a previous authenticity check or whether a background scan is done — never use it to poll immediately after starting a scan.
list-evidence-packs
ChatGPTLists evidence packs for the authenticated user, newest first, including their takedown request status. Returns pack IDs, threat identifiers, threat types, creation dates, and current takedown status. Supports search, status, dateFrom/dateTo, archived (default false lists active packs), and limit/offset paging (hasMore and nextOffset say whether another page exists; pass nextOffset back as offset). Narrow with the filters instead of paging through everything: the response reports total (packs matching the filters), statusCounts (packs per takedown status) and viewCounts (active vs archived), so counts can be answered without fetching every page. Use this to get pack IDs before calling create-takedown-requests, delete-evidence-packs, archive-evidence-packs, unarchive-evidence-packs or fetch-takedown-request-status.
list-safe-threats
ChatGPTReturns a page of the threats (impersonator accounts and harmful posts) that the user has marked as safe. Returns total, returned, hasMore and nextOffset; pass nextOffset back as offset (same filters) for the next page. Use this to see which accounts or posts have been dismissed. Each entry includes the threat type, identifier (URL), and full threat data.
mark-threats-safe
ChatGPTMarks threats safe. For specific threats pass threats (identifiers from the fetch tools; the server loads each threat). For "mark all / every matching" requests pass filters (same as the fetch tools) instead: it marks up to 1000 per call, and dryRun: true only reports how many would change. If isComplete is false, report remaining and stoppedReason; call again with the same filters only when stoppedReason is cap, otherwise stop.
run-true-seal-scan
ChatGPTStarts a TrueSeal authenticity check on a social post or profile URL. Under the hood this is a deepfake scan: the media is downloaded and run through AI-generation detection models. The scan runs in the background and usually takes a few minutes — do not wait for it. Tell the user it has started, that they will get a notification when it finishes, and that they can check the TrueSeal section of the dashboard or ask you later to look it up. It counts against the monthly deepfake scan quota, though repeat scans of the same URL return the cached result without consuming quota. Pass evidencePackId to attach the scan to an existing evidence pack. Use fetch-true-seal-scan later with the returned scanBatchId to check whether it has finished.
unarchive-evidence-packs
ChatGPTRestores archived evidence packs to the active list. For specific packs pass evidencePackIds (from list-evidence-packs with archived: true). For "unarchive all / every matching" requests pass filters (same as list-evidence-packs) instead; dryRun: true only reports how many archived packs match.
unmark-threats-safe
ChatGPTRemoves one or more threats from the safe list, returning them to the main threats view. Pass the threat identifiers (URLs) to unmark. Persona scope is the authenticated session (organization vs personal is already determined by the chat request). You can get these identifiers from list-safe-threats. This is idempotent — already-unmarked threats are skipped.
update-custom-topics
ChatGPTCreates or edits the custom monitoring topics Guardian runs targeted lookups for, such as a specific narrative, event, product name, or emerging term. Use add and remove for incremental changes, which preserve topics the user did not mention; only use replace when they want to start the list over. Do not combine replace with add or remove. Call fetch-custom-topics first when you need to show or reason about the current list. Topics are capped by the subscription tier, and changes take effect on the next monitoring run.
update-detection-settings
ChatGPTUpdates Detection Settings: industry (category), role, region, public profile URLs, and custom detection context. If the user has organization memberships in context, pass organizationId from Active Scope. Call fetch-detection-settings first when you need current values. Changes apply from the next monitoring run.
update-protection-context
ChatGPTCreates or edits Protection Context criteria and whitelist URLs. Built-in default criteria: 1 = NCIIOrDeepfake, 2 = Impersonation, 3 = Doxxing. Prefer add/remove for incremental changes; use replace only when the user wants a full rewrite. Provide at least one of defaultCriteria, criteria or whitelistUrls. Call fetch-protection-context first when you need the current list.
archive-evidence-packs
Claudecreate-evidence-pack-from-url
Claudecreate-evidence-packs
Claudecreate-takedown-requests
Claudedelete-evidence-packs
Claudefetch-account-context
Claudefetch-custom-topics
Claudefetch-detection-settings
Claudefetch-evidence-pack
Claudefetch-evidence-pack-takedownability
Claudefetch-historical-risk-scores
Claudefetch-impersonators
Claudefetch-posts
Claudefetch-protection-context
Claudefetch-risk-score
Claudefetch-subscription-quotas
Claudefetch-takedown-guidance
Claudefetch-takedown-request-status
Claudefetch-threats
Claudefetch-true-seal-scan
Claudelist-evidence-packs
Claudelist-safe-threats
Claudemark-threats-safe
Clauderun-true-seal-scan
Claudeunarchive-evidence-packs
Claudeunmark-threats-safe
Claudeupdate-custom-topics
Claudeupdate-detection-settings
Claudeupdate-protection-context
Claude