The Agent Skills Directory

[PROMPT_INJECTION]: No patterns of system prompt extraction, safety filter bypass, or instruction override were detected. The persona-based instructions ('Diretor de Comunicacao Digital') are within the expected operational scope for a social media assistant.
[DATA_EXFILTRATION]: No evidence of hardcoded credentials, sensitive file path access (e.g., .ssh, .env), or unauthorized network exfiltration. The skill references legitimate integrations for Instagram, Telegram, and WhatsApp APIs.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where it ingests untrusted data for processing.
Ingestion points: User-provided content strings and media through the /Publish_All and /Campaign commands in SKILL.md.
Boundary markers: Absent; there are no specific delimiters or instructions to ignore embedded commands in the source content.
Capability inventory: Uses the instagram, telegram, and whatsapp-cloud-api tools to perform network operations (publishing).
Sanitization: No explicit sanitization or validation of the input content is described before adaptation and cross-platform distribution.
[OBFUSCATION]: No hidden content, Base64 encoding of commands, zero-width characters, or homoglyph attacks were identified in the document.
[REMOTE_CODE_EXECUTION]: No remote script downloads or dynamic code execution patterns (like eval or subprocess calls) were found in the provided instruction file.
[METADATA_POISONING]: The metadata fields in the YAML frontmatter are consistent with the skill's stated purpose. The 'risk: critical' field appears to be a user-defined classification for business criticality rather than a security warning.

social-orchestrator