Skip to main content

Garden catalog

First-party filters. Categories display in this order: Privacy → Injection → Safety → Advice → Quality.

Unless noted, every card is deterministic (zero extra model latency) and the default action is block. Modes: PRE (prompt), POST (completion), BOTH.

Privacy

CardDescription
PII patternsSSNs, payment cards, emails, phone numbers
Secret / API key detectorLeaked provider keys, GitHub tokens, Slack tokens, AWS keys

Injection

CardModeDescription
Jailbreak / system promptPREIgnore-instructions, DAN, system-prompt extraction
SQL injectionPRESQL injection attempts in prompts
Data exfiltrationPREDump secrets, env vars, or training data
Malicious code fencesBOTHExecutable markdown / script payloads

Safety

CardDescription
Harmful violenceViolence and criminal planning keywords
Harmful self-harmSelf-harm and suicide content
Harmful child safetyContent that endangers minors
Illegal weaponsWeapons manufacturing or acquisition
Insults & personal attacksInsults directed at people
Toxic languageSlurs and abusive language
Competitor namesConfigurable competitor name list — edit it on the guardrail
Keyword blockingCustom word list — configure keywords when you add it

Advice

CardDescription
Denied financial advicePersonalized financial-advice phrases
Denied legal adviceUnauthorized legal advice
Denied medical adviceMedical diagnosis or treatment advice

These are phrase packs, not licensed classifiers.

Quality

CardModeDescription
Word count limitBOTHMaximum word count on prompt or response
URL allow / denyBOTHSuspicious or denylisted URLs
JSON schema (post)POSTValidate model output against a JSON schema
Block executable markdownBOTHFenced code blocks that look executable

Adding a card from the Garden creates a BUILTIN guardrail instance. It still does nothing until a policy step references it and that policy is attached.