The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a self-improving workflow where it discovers new insights or 'gotchas' during user interaction and appends them to the references/learnings.md file. This file is read at the beginning of each session to influence the agent's behavior.
Ingestion points: Untrusted data enters the context via user queries and descriptions of their acquisition deals, metrics, and process results.
Boundary markers: Absent. The instructions do not define delimiters or provide warnings to disregard embedded instructions when the agent reads from or writes to the learnings.md file.
Capability inventory: The skill instructions direct the agent to perform file-write (append) operations to a local reference file based on runtime information.
Sanitization: Absent. There is no requirement or logic for sanitizing or validating user-provided content before it is persisted in the knowledge base, allowing for potential persistent instruction poisoning.

sales-acquire