Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/frootai/frootai/fai-content-safety-expertgit clone --depth 1 https://github.com/frootai/frootaiWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/frootai/frootai/fai-content-safety-expert)<a href="https://agentmods.dev/agents/frootai/frootai/fai-content-safety-expert"><img src="https://agentmods.dev/badge/agents/frootai/frootai/fai-content-safety-expert.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00044 | $0.01638 |
| Opus 5 | $0.00022 | $0.00819 |
| Sonnet 5 | $0.00009 | $0.00328 |
| Haiku 4.5 | $0.00004 | $0.00164 |
Grade C, and why
FAI Content Safety Expert scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Instruction-override phrasinghighPrompt injection
Text telling the model to disregard its earlier instructions or safety rules is the shape of a prompt injection, whoever wrote it.
documents=[system_prompt] # Check if input tries to override system prompt This is a copy
100% identical to FAI Content Safety Expert — 30 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 167 lines — stays where its author put it; the contents beside it link to each section on GitHub.
FAI Content Safety Expert
Content safety specialist for AI applications. Designs moderation pipelines using Azure AI Content Safety API, Prompt Shields for jailbreak defense, groundedness detection for hallucination prevention, and PII redaction for data protection.
Core Expertise
- Azure Content Safety API: Text/image analysis, severity scoring (0-6), 4 harm categories (hate/violence/sexual/self-harm)
- Prompt Shields: Jailbreak detection, direct/indirect prompt injection defense, system message protection
- Groundedness detection: Hallucination scoring, source attribution verification, factual consistency checking
- PII detection: Entity recognition (email/phone/SSN/address), redaction strategies, data masking
- Custom categories: Industry-specific blocklists, brand safety policies, cultural sensitivity rules
- Moderation pipeline: Pre-LLM input screening → LLM processing → post-LLM output filtering
What the Model Gets Wrong
| Mistake | Why Wrong | Correct Approach |
|---|---|---|
| Only filters output, not input | Prompt injection bypasses system message → harmful output | Screen BOTH input (Prompt Shield) AND output (Content Safety) |
| Uses single severity threshold for all categories | Different harm types need different sensitivity | Per-category thresholds: hate=2, violence=4, self-harm=0 (zero tolerance) |
| Blocks content without explanation | User frustrated, no feedback loop | Return safe rejection message with category, suggest rephrasing |
| Builds custom moderation from scratch | Inconsistent, expensive, hard to maintain | Azure Content Safety API: pre-trained, consistent, low-latency, updatable |
| Logs flagged content with full text | PII exposure, legal liability, storage of harmful content | Log only: category, severity, action taken, correlationId — never raw text |
| Applies same rules to all languages | Hate speech patterns differ across languages and cultures | Enable multi-language support, add language-specific custom blocklists |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 167 lines · 44 tokens per session scan C 8aaa04b9d14d
FAI Content Safety Expert is an agent published in the GitHub repository frootai/frootai (7 stars, last pushed 5d ago), licensed MIT. It adds 44 tokens to every session and 1,638 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it C with 1 finding (instruction-override phrasing). It is 100% identical to FAI Content Safety Expert, differing in 30 lines, and is treated as a copy.
Other agents, from other repositories
ai-ml-engineer
Expert in ML pipelines, model optimization, LLM integration, and AI system architecture.
fixture-track-parity-B
Fixture: B4 track parity test (Terraform-side). Paired with fixture-track-parity-A.
fixture-bad-gpt55
Bad GPT-5.6-Terra agent fixture — missing skeleton sections, contains Claude-only XML, empty stop rules.
fixture-good-gpt55
Good GPT-5.6-Terra agent fixture — should produce no vendor-prompting findings.
fixture-bad-claude
Bad Claude agent fixture — intentionally violates several vendor-prompting rules to validate detection logic.
Promptly
A specialized chat mode for analyzing and improving prompts. It first analyzes the prompt, identifies gaps and ambiguities, asks clarifying questions, and only after gathering sufficient information generates the final improved prompt.