prompt-guard

Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, <1% FPR. Fast (<2ms GPU). Multilingual (8 languages). Deploy with HuggingFace or batch processing for RAG security.

Quality

65%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./skills/safety-alignment/prompt-guard/SKILL.md

Quality

Content

68%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is highly actionable with comprehensive executable examples and good section organization, but it is somewhat verbose and the batch workflow lacks explicit validation checkpoints. Splitting long reference material into bundle files would further improve structure.

Suggestions

Add explicit validation/feedback steps to the batch workflow (e.g., log blocked docs, review threshold tuning, retry on tokenization errors) to lift the workflow_clarity cap.

Remove the fabricated per-pattern score comments and repeated model-loading boilerplate to tighten conciseness.

Consider moving the threshold-recommendations table, hardware requirements, and pattern galleries into a reference bundle file and linking to it from the overview.

Dimension	Reasoning	Score
Conciseness	Mostly efficient executable code, but includes padded illustrative sections ('Common jailbreak patterns detected' with fabricated scores, repeated model-loading boilerplate) and restates basics Claude already knows.	3 / 5
Actionability	Fully executable, copy-paste-ready code covering the common cases: user-input filtering, third-party/injection filtering, batch processing, long-text sliding window, and context-aware thresholds.	5 / 5
Workflow Clarity	Workflows are clearly sequenced, but the batch-processing workflow lacks explicit validation/feedback checkpoints, and the rubric caps batch-operation workflows without validation loops at 3.	3 / 5
Progressive Disclosure	Well-organized into clear sections (quick start, workflows, issues, thresholds, hardware) with no nested references; minor gaps since all content is inline with no separate reference files.	4 / 5
	Total	15 / 20 Passed

Description

61%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific and uses third-person voice with good trigger keywords, but it omits any explicit 'Use when...' guidance, which caps completeness. It reads more as a product blurb with performance stats than a skill invocation description.

Suggestions

Add an explicit trigger clause such as 'Use when filtering user prompts or third-party data for prompt injection or jailbreak attempts in LLM/RAG applications.'

Trim the performance-marketing stats (TPR/FPR/latency) from the description; they pad tokens without aiding skill selection.

Expand concrete actions to be more comprehensive (e.g., detect indirect injections, batch-score documents, sliding-window long text).

Dimension	Reasoning	Score
Specificity	Names the domain and a couple of concrete actions ('Filters malicious prompts and third-party data', 'Deploy with HuggingFace or batch processing') but the action set is not comprehensive enough for a 4.	3 / 5
Completeness	Has a clear 'what' (prompt injection/jailbreak detector that filters prompts and third-party data) but no explicit 'Use when...' trigger clause, which caps completeness at 3 per the rubric.	3 / 5
Trigger Term Quality	Strong natural keyword coverage ('prompt injection', 'jailbreak', 'LLM apps', 'RAG security') that users would actually say, with only a few synonyms missing.	4 / 5
Distinctiveness Conflict Risk	The prompt-injection/jailbreak-detection niche is distinct and specific with minimal overlap risk versus content-moderation or policy skills.	4 / 5
	Total	14 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: OpenLAIR/dr-claw
Path: skills/safety-alignment/prompt-guard/SKILL.md
Commit: 4b35756

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.