The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It instructs the agent to fetch and analyze session transcripts from transcript_url (untrusted external data) and use that analysis to perform 'surgical edits' on AI scenarios.
Ingestion points: Transcripts and failure evidence are fetched from external URLs provided in session data, as specified in SKILL.md (Step 1).
Capability inventory: The agent has the capability to modify scenario logic, instructions, and guardrails via the ttai:update_scenario tool.
Boundary markers: The skill does not define clear boundary markers or instructions for the agent to ignore potentially malicious directions embedded within the transcripts.
Sanitization: No sanitization or validation of the transcript content is performed before the agent uses it to plan and apply changes to the scenario instructions.
[EXTERNAL_DOWNLOADS]: The skill connects to an external MCP server at https://api.toughtongueai.com/api/public/mcp. This connection is necessary for the skill's primary function and targets the official domain of the technology provider.

scenario-refiner