The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads from references/learnings.md and references/platform-guide.md to provide context-aware help. It also instructs the agent to append new findings to references/learnings.md to maintain persistent knowledge. This is a standard mechanism for self-improving skills within a local environment.
[EXTERNAL_DOWNLOADS]: The skill mentions installation via well-known package managers (brew, winget) and the official GitHub repository for the Handy project. These are standard procedures for software distribution and do not involve suspicious or unverified sources.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it reads a persistent file (references/learnings.md) that it also modifies. While this creates a theoretical attack surface for indirect prompt injection, the skill lacks dangerous capabilities (such as network operations or arbitrary code execution) that could be exploited via this channel.
Ingestion points: references/learnings.md is read in Step 1 of the SKILL.md file.
Boundary markers: None are explicitly used when reading the learnings file.
Capability inventory: The skill provides informational guidance and lacks file-system write (except to its own reference), network, or subprocess execution capabilities.
Sanitization: Not present, as the data is treated as plain text documentation.

sales-handy