The Agent Skills Directory

[PROMPT_INJECTION]: No instructions attempting to override agent behavior or bypass safety filters were detected. The SKILL.md file contains robust guidelines that emphasize safety, human review gates, and adherence to source-backed facts.
[DATA_EXPOSURE_AND_EXFILTRATION]: No hardcoded credentials, sensitive file paths, or data exfiltration patterns were found. The skill explicitly instructs agents to protect customer data and reviewer PII, and to avoid storing secrets in public files.
[OBFUSCATION]: Analysis of the skill body and referenced files revealed no Base64-encoded commands, zero-width characters, homoglyphs, or other obfuscation techniques designed to hide malicious intent.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote scripts from unknown sources. Referenced URLs in sources.yml and country packs target reputable international organizations and official government portals.
[COMMAND_EXECUTION]: The included utility script scripts/list-vendors.mjs is a benign script used to list country identifiers and does not perform dangerous system operations or handle untrusted input.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests market data, it implements a structured evidence model and claims ledger system.
Ingestion points: Reads country-specific context from references/countries/*.md and source metadata from sources.yml.
Boundary markers: Explicitly uses Confidence Model, Evidence Model, and Claims Ledger to scope and validate information.
Capability inventory: No file-write or external network-send capabilities are present in the provided scripts.
Sanitization: Emphasizes native and expert review gates for any cultural, linguistic, or high-risk business claims.

arab-market-context