Moderator Actions

Comprehensive moderation toolkit for Civitai. Covers user actions, strike management, image moderation, report handling, generation moderation, content/training moderation, NCMEC/CSAM reporting, and Creator Shop purchase refunds.

All scripts use the Civitai tRPC API with Bearer token authentication, ensuring actions go through the proper service layer with full side effects.

Setup

1. Copy .env-example to .env in this skill directory
2. Add your Civitai API key (must belong to a moderator account)
3. Optionally set the API URL (defaults to production)

cp .claude/skills/mod-actions/.env.example .claude/skills/mod-actions/.env
# Edit .env and add your API key

Important: The API key must belong to a user with moderator privileges.

Variable	Required	Default	Description
CIVITAI_API_KEY	Yes	-	Your Civitai API key
CIVITAI_API_URL	No	https://civitai.com	API base URL

---

Skill Files

File	Domain	Commands
query.mjs	User moderation	user, ban, mute, leaderboard, remove-content, dm
strikes.mjs	Strike system	get-user, standings, list, create, void
images.mjs	Image moderation	review-queue, review-counts, moderate, tos-violation, rescan, report-csam, poi-tags, user-images, rating-requests, ingestion-errors, resolve-ingestion, downleveled, pending-ingestion, toggle-flag
reports.mjs	Report handling	list, set-status, bulk-status, update, appeals, appeal-details, resolve-appeal
generation.mjs	Generation moderation	flagged-consumers, flagged-reasons, consumer-strikes, review-strikes, user-generations, restrictions, resolve-restriction, allowlist-add, debug-audit, todays-counts, suspicious-matches
content.mjs	Content moderation	models, flagged-models, resolve-flagged, model-versions, rescan-model, restore-model, toggle-cannot-promote, toggle-cannot-publish, articles, mod-rule
model3ds.mjs	3D Models moderation	list, get, files, unpublish, delete, restore, set-nsfw-level, toggle-tos, toggle-poi, toggle-minor, toggle-nsfw, toggle-unlisted
csam.mjs	NCMEC/CSAM reporting	reports, stats, image-resources, create-report
shop.mjs	Creator Shop	refund

---

query.mjs — User Moderation

node .claude/skills/mod-actions/query.mjs <command> [options]

Command	Description
user <id|username>	Look up user info
ban <id|username>	Toggle ban status
mute <id|username>	Toggle mute status
leaderboard <id|username> <true|false>	Set leaderboard eligibility
remove-content <id|username>	Remove all user content (DESTRUCTIVE)
dm <id|username>	Send a DM (requires --message)
creator-program-ban <id|username> <set|clear>	Ban from / unban from the Creator Program (requires --internal <reason>; optional --gate <ref>; needs the user.creatorProgram.ban permission). Use this, never a raw onboarding UPDATE.

Options: --json, --dry-run, --reason <code>, --message <text>, --internal <text>, --gate <ref>

Ban Reason Codes: SexualMinor, SexualMinorGenerator, SexualMinorTraining, SexualPOI, Bestiality, Scat, Nudify, Harassment, SpamBot, LeaderboardCheating, BuzzCheating, RRDViolation, Other

node .claude/skills/mod-actions/query.mjs user 3879899
node .claude/skills/mod-actions/query.mjs ban 3879899 --reason Other --message "ToS violation" --dry-run
node .claude/skills/mod-actions/query.mjs dm 3879899 --message "Please review our guidelines"
node .claude/skills/mod-actions/query.mjs creator-program-ban 12345 set --internal "Counterparty farming" --gate hub:request:abc

---

strikes.mjs — Strike System

node .claude/skills/mod-actions/strikes.mjs <command> [options]

Command	R/W	Description
get-user <userId>	READ	Strike history for a user
standings	READ	User standings (filter: active/muted/flagged)
list	READ	Paginated strike list with filters
create <userId>	WRITE	Issue a strike (requires --reason, --description)
void <strikeId>	WRITE	Void a strike (requires --reason)

Strike Reasons: BlockedContent, RealisticMinorContent, CSAMContent, TOSViolation, HarassmentContent, ProhibitedContent, ManualModAction

Options: --reason, --description, --points <1-3>, --entity-type, --entity-id, --report-id, --expires-days, --status <Active,Expired,Voided>, --sort <points|score|lastStrike|created>, --sort-order <asc|desc>, --flagged-for-review, --has-active-strikes, --page, --limit, --json, --dry-run

node .claude/skills/mod-actions/strikes.mjs get-user 3879899
node .claude/skills/mod-actions/strikes.mjs standings --has-active-strikes --limit 10
node .claude/skills/mod-actions/strikes.mjs create 3879899 --reason BlockedContent --description "Uploaded prohibited content" --points 1 --dry-run
node .claude/skills/mod-actions/strikes.mjs void 42 --reason "False positive from scanner"

---

images.mjs — Image Moderation

node .claude/skills/mod-actions/images.mjs <command> [options]

Command	R/W	Description
review-queue	READ	Image review queue
review-counts	READ	Queue tab counts
moderate <ids> <action>	WRITE	Block/unblock images
tos-violation <imageId>	WRITE	Flag as TOS violation
rescan <imageId>	WRITE	Re-ingest/rescan image
report-csam <imageIds>	WRITE	Report images as CSAM
poi-tags	READ	Person-of-interest tags
user-images <userId>	READ	All images for a user
rating-requests	READ	NSFW rating review requests
ingestion-errors	READ	Images with ingestion errors
resolve-ingestion <id>	WRITE	Resolve ingestion error
downleveled	READ	Downleveled images
pending-ingestion	READ	Pending ingestion images
toggle-flag <imageId>	WRITE	Toggle image flag

Options: --ids, --action, --review-type, --cursor, --page, --limit, --json, --dry-run

node .claude/skills/mod-actions/images.mjs review-counts
node .claude/skills/mod-actions/images.mjs review-queue --limit 20
node .claude/skills/mod-actions/images.mjs moderate 12345,67890 block --dry-run
node .claude/skills/mod-actions/images.mjs user-images 3879899 --json
node .claude/skills/mod-actions/images.mjs rescan 12345

---

reports.mjs — Report Handling

node .claude/skills/mod-actions/reports.mjs <command> [options]

Command	R/W	Description
list	READ	Reports by entity type (requires --type)
set-status <reportId> <status>	WRITE	Set report status
bulk-status <status>	WRITE	Bulk update statuses (requires --ids)
update <reportId>	WRITE	Update report with status/notes
appeals	READ	Recent appeals
appeal-details <id>	READ	Appeal details
resolve-appeal	WRITE	Resolve appeal (requires --ids, --entity-type, --status)

Report Entity Types: model, comment, commentV2, image, resourceReview, article, post, reportedUser, collection, bounty, bountyEntry, chat, comicProject

Report Statuses: Pending, Processing, Actioned, Unactioned

Appeal Statuses: Pending, Approved, Rejected

Options: --type, --status, --ids, --entity-type, --internal, --message, --user-id, --start-date, --page, --limit, --query, --json, --dry-run

node .claude/skills/mod-actions/reports.mjs list --type image --limit 20
node .claude/skills/mod-actions/reports.mjs set-status 456 Actioned --dry-run
node .claude/skills/mod-actions/reports.mjs bulk-status Actioned --ids 1,2,3
node .claude/skills/mod-actions/reports.mjs appeals   # the signed-in account's own recent appeals
node .claude/skills/mod-actions/reports.mjs resolve-appeal --ids 10,11 --entity-type Image --status Approved

---

generation.mjs — Generation Moderation

node .claude/skills/mod-actions/generation.mjs <command> [options]

Command	R/W	Description
flagged-consumers	READ	Flagged generation consumers
flagged-reasons	READ	Flagging reasons
consumer-strikes <userId>	READ	Consumer's generation strikes
review-strikes <userId>	WRITE	Mark strikes as reviewed
user-generations <userId>	READ	User's generated images
restrictions	READ	All generation restrictions
resolve-restriction <id>	WRITE	Resolve restriction (Upheld/Overturned)
allowlist-add	WRITE	Add prompt trigger to allowlist
debug-audit <prompt>	READ	Test prompt auditing
todays-counts	READ	Today's prohibited request counts
suspicious-matches	READ	Saved suspicious audit matches

Options: --status <Pending|Upheld|Overturned>, --reason, --start-date, --message, --trigger, --category, --negative-prompt, --user-id, --username, --restriction-id, --page, --limit, --json, --dry-run

node .claude/skills/mod-actions/generation.mjs flagged-consumers
node .claude/skills/mod-actions/generation.mjs consumer-strikes 3879899
node .claude/skills/mod-actions/generation.mjs restrictions --status Pending
node .claude/skills/mod-actions/generation.mjs resolve-restriction 42 --status Upheld --message "Confirmed violation" --dry-run
node .claude/skills/mod-actions/generation.mjs debug-audit "some prompt text"
node .claude/skills/mod-actions/generation.mjs allowlist-add --trigger "girl" --category "age" --reason "False positive"

---

content.mjs — Content Moderation

node .claude/skills/mod-actions/content.mjs <command> [options]

Command	R/W	Description
models	READ	Models for mod review
flagged-models	READ	Flagged models
resolve-flagged	WRITE	Resolve flagged models (requires --ids)
model-versions	READ	Model versions for mod
rescan-model <id>	WRITE	Re-ingest a model
restore-model <id>	WRITE	Restore deleted model
toggle-cannot-promote <id>	WRITE	Toggle promotion eligibility
toggle-cannot-publish <id>	WRITE	Toggle publish ability
articles	READ	Articles for mod review
mod-rule <id>	READ	Get moderation rule

Options: --ids, --page, --limit, --json, --dry-run

Training moderation (list / approve / deny training data) moved to the moderator spoke (/audit/training-models, /audit/training-data) in the moderator-app migration and is no longer reachable via tRPC — see docs/moderator-app/page-migration-checklist.md.

node .claude/skills/mod-actions/content.mjs flagged-models --limit 10
node .claude/skills/mod-actions/content.mjs resolve-flagged --ids 1,2,3 --dry-run
node .claude/skills/mod-actions/content.mjs rescan-model 789

---

model3ds.mjs — 3D Models Moderation

node .claude/skills/mod-actions/model3ds.mjs <command> [options]

Command	R/W	Description
list	R	List Model3Ds (--status, --username, --limit, --cursor)
get <id>	R	Get a Model3D by id
files <id>	R	List files attached to a Model3D
unpublish <id>	W	Set status -> Unpublished
delete <id>	W	Soft-delete (status -> Deleted, sets deletedAt/deletedBy)
restore <id>	W	Restore (Deleted -> Unpublished, Unpublished -> Published)
set-nsfw-level <id>	W	Override nsfwLevel (--level <n> [--lock])
toggle-tos <id>	W	Toggle tosViolation (locks the field)
toggle-poi <id>	W	Toggle poi (locks the field)
toggle-minor <id>	W	Toggle minor (locks the field)
toggle-nsfw <id>	W	Toggle nsfw (locks the field)
toggle-unlisted <id>	W	Toggle unlisted (locks the field)

Strikes against a Model3D use the existing strikes.mjs (the strike system is entity-type-agnostic):

node .claude/skills/mod-actions/strikes.mjs create <userId> \
  --entity-type Model3D --entity-id <model3dId> --reason TOSViolation --points 1

Common flags:

• --json — raw JSON output
• --dry-run — preview without making changes (write commands only)

Examples:

node .claude/skills/mod-actions/model3ds.mjs list --status Draft --limit 100
node .claude/skills/mod-actions/model3ds.mjs get 42
node .claude/skills/mod-actions/model3ds.mjs unpublish 42
node .claude/skills/mod-actions/model3ds.mjs set-nsfw-level 42 --level 16 --lock
node .claude/skills/mod-actions/model3ds.mjs toggle-tos 42 --dry-run
node .claude/skills/mod-actions/model3ds.mjs restore 42

---

csam.mjs — NCMEC/CSAM Reporting

node .claude/skills/mod-actions/csam.mjs <command> [options]

Command	R/W	Description
reports	READ	Paginated CSAM reports
stats	READ	CSAM report statistics
image-resources <imageIds>	READ	Resources used to generate flagged images
create-report <userId>	WRITE	Create NCMEC CyberTipline report

CSAM Report Types: Image, TrainingData, GeneratedImage

Options: --type <CsamReportType>, --image-ids <id1,id2,...>, --minor-depiction <real|non-real>, --page, --limit, --json, --dry-run

node .claude/skills/mod-actions/csam.mjs stats
node .claude/skills/mod-actions/csam.mjs reports --limit 10
node .claude/skills/mod-actions/csam.mjs image-resources 12345,67890
node .claude/skills/mod-actions/csam.mjs create-report 3879899 --type GeneratedImage --image-ids 111,222 --dry-run

---

shop.mjs — Creator Shop

node .claude/skills/mod-actions/shop.mjs refund <buzzTransactionId...> --reason <text> (--clawback | --no-clawback) [--writable]

Command	R/W	Description
refund <ids...>	WRITE	Refund purchases: Buzz back to the buyer, payouts reversed with --clawback, cosmetic revoked, purchase marked refunded (creatorShop.refundPurchases)

Dry run unless --writable (the server's dry run lists each purchase's ledger legs). Re-running is safe. --no-clawback means Civitai pays the refund, --clawback the creator. Choose per case: an agent's --writable run needs Justin's approval card, and the card must name the clawback value. Refusals and behaviour: docs/features/creator-shop.md → "Refunding a purchase by hand".

---

Global Options

All scripts except shop.mjs (dry run by default; --writable writes) support:

Flag	Description
--json	Output raw JSON (for agent/scripting consumption)
--dry-run	Preview WRITE actions without executing
--page <n>	Page number for paginated results (default: 1)
--limit <n>	Page size for paginated results (default: 20)

Safety Notes

• Always use --dry-run first for destructive/write actions (shop.mjs: omit --writable first)
• All actions are logged in the ModActivity table
• Ban actions invalidate user sessions immediately
• CSAM reports are submitted to NCMEC — use with extreme care
• Document reasons with --reason and --message flags
• API key must have moderator privileges
• Use --json for programmatic consumption by agent daemons