The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions define a workflow that ingests untrusted data from external sources including behavior specifications, source code, and test suites (identified in SKILL.md). This represents an attack surface for indirect prompt injection if the external content contains hidden instructions meant to manipulate the agent's behavior. The skill directs the agent to 'Run relevant tests and quality checks', which is a capability that executes code potentially influenced by this untrusted data. There are no explicit boundary markers, sanitization rules, or escaping mechanisms defined in the instructions to protect against instructions embedded within the processed data.

tdd-verify