The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent learning mechanism where it reads from and appends to references/learnings.md. This architecture creates a vulnerability surface for indirect prompt injection, as malicious instructions provided in a user prompt could be stored and subsequently influence the agent's behavior in future sessions.\n
Ingestion points: Data is ingested from the user's prompt during the 'Actionable guidance' phase and read from references/learnings.md at the start of each invocation.\n
Boundary markers: There are no explicit delimiters or instructions defined to isolate the content of the learning file or to instruct the agent to ignore executable commands within that file.\n
Capability inventory: The skill possesses the capability to write to the local file system (references/learnings.md). It does not currently trigger high-risk operations (such as network or shell execution) based directly on the contents of this file.\n
Sanitization: No sanitization or validation of the 'gotchas or tips' is performed before they are appended to the persistent storage file.\n
Remediation: To mitigate this risk, wrap the content of the learnings file in clear delimiters (e.g., XML tags) with an explicit instruction to the agent to treat the content as data only and ignore any embedded instructions. Implement sanitization for user-provided tips before storage.\n- [EXTERNAL_DOWNLOADS]: The skill documentation includes an installation command (npx skills add sales-skills/sales) to fetch related tools. This resource is hosted under the vendor's own namespace and is consistent with the skill's stated purpose.

sales-litextension