The Agent Skills Directory

[SAFE]: Comprehensive analysis of the skill files confirms the absence of malicious patterns.
No evidence of prompt injection, data exfiltration, or persistence mechanisms was found across the instructions, templates, and metadata.
The skill is purely informational, serving as a framework for the agent to provide negotiation advice using provided text references.
[INDIRECT_PROMPT_INJECTION]: The skill operates on user-supplied negotiation scenarios, establishing a theoretical attack surface for indirect injection.
Ingestion points: User input regarding negotiation parties, interests, and history is processed in SKILL.md and references/method.md.
Boundary markers: The skill lacks explicit delimiters or "ignore embedded instructions" warnings for processing external scenario data.
Capability inventory: The skill has no access to sensitive tools; it lacks the ability to execute shell commands, perform network operations, or write to the file system.
Sanitization: No automated sanitization of user-provided strings is implemented.
Result: While the ingestion surface exists, the lack of exploitable capabilities prevents any significant security impact.
[NO_CODE]: The skill consists entirely of Markdown and JSON configuration.
No Python, JavaScript, or shell scripts are included or referenced for execution.
All external URLs are bibliographic references to academic and news sites (CEMS, Nova SBE, Observador), which are handled as static data rather than executable resources.

costa-negotiation-dynamics