detect_prompt_injectionClassify a prompt before it reaches your LLM. Brainiall Prompt Shield engine.
Returns category (jailbreak | prompt_injection | data_exfiltration | impersonation | none), severity, reason, confidence.
| Name | Required | Description | Default |
|---|---|---|---|
| prompt | Yes | The prompt text to classify (NOT executed) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, openWorldHint=true, and destructiveHint=false, so the safe, side-effect-free profile is covered. The description adds value beyond annotations by disclosing the output contract (category, severity, reason, confidence) and the open-world nature of the input. It does not contradict the annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two tight sentences. The purpose is front-loaded in the first sentence, and the return structure is listed compactly in the second. There is no fluff or redundant restatement of the title — appropriate brevity for a single-purpose classifier.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a one-parameter classifier with 100% schema coverage, comprehensive safety annotations, and no output schema, the description adequately discloses the input, the output contract, and the intended timing of use. The only minor gap is not explicitly routing the agent away from overlapping detection siblings, which is mitigated by the clear category list.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% — the single `prompt` parameter is fully documented in the schema, including the important 'NOT executed' caveat. Per the baseline rule for high coverage, the description doesn't need to add parameter detail, and the tool objectively adds nothing beyond the schema here.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb and resource — 'Classify a prompt before it reaches your LLM' — and enumerates the exact output categories (jailbreak, prompt_injection, data_exfiltration, impersonation, none), which clearly differentiates it from sibling detectors like detect_pii, detect_protected_material, and analyze_toxicity. The title 'Prompt Shield (Jailbreak / Injection Detection)' reinforces the specific purpose.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The phrase 'before it reaches your LLM' provides clear context for when to invoke this tool — it is a pre-flight safety gate. However, the description never names alternatives or gives explicit when-not-to-use guidance, even though several sibling tools (detect_pii, detect_protected_material, analyze_toxicity) occupy adjacent detection space.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Add one secure layer between your agents and this server.
Most tools map cleanly to distinct capabilities, and the descriptions make the intended use clear. A few adjacent pairs—extract_entities vs link_entities_to_wikidata and detect_pii vs detect_conversational_pii—require careful selection, but they are distinguishable by their stated outputs.
Names are uniformly lowercase snake_case and mostly follow a verb_object pattern, such as analyze_*, detect_*, extract_*, summarize_text, and translate_text. A few outliers like aspect_sentiment, fraud_feedback, and knowledge_ingest break the verb-first feel, but the overall pattern remains predictable.
At 22 tools, this is on the heavy side of the borderline range. Each tool has a distinct job, but the mix of core NLP, safety, fraud, health-checking, and knowledge-base management makes the surface feel sprawling rather than tightly scoped.
The core NLP coverage is broad: sentiment, toxicity, PII, entities, QA, summarization, translation, and groundedness are all present. However, the knowledge-base tools support ingest/list/query but no delete or update, creating a dead end when documents need correction or removal.