The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external Voice of Customer (VOC) data from platforms such as Reddit, Amazon, and YouTube (specified in voc-mining-playbook.md). This represents a potential surface for indirect prompt injection.
Ingestion points: The skill reads local files (products.md, audience.md, competitors.md) and processes external textual data from social media and e-commerce platforms.
Boundary markers: The skill includes explicit instructions in the Context Matrix (SKILL.md) to never use "unverified fictional data" and references a system-level interaction protocol for safety.
Capability inventory: The skill is capable of appending findings to a local memory file (learnings.jsonl). It does not have shell execution or automated network exfiltration capabilities.
Sanitization: The workflow includes a "cleaning" step in the VOC mining SOP to remove invalid, robotic, or overly vague comments, reducing the risk of processing malicious or irrelevant data.
[DATA_EXPOSURE]: The skill references standard project configuration files such as .env (by instructing users to store secrets there, which is a safe practice) and local markdown files. No sensitive system paths or hardcoded credentials were detected.
[REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution. The mentions of "AI-powered exploration" are described as methodological frameworks for the agent to follow, not as scripts to be executed.

afa-explore