The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ground marketing copy in 'product truth' by reading external files such as codebase, docs, changelogs, and specs. This creates a surface for indirect prompt injection, as malicious instructions embedded in these files could override agent behavior during the copy-generation process.
Ingestion points: SKILL.md references codebase, documentation, changelogs, specs, and product notes as sources of truth.
Boundary markers: Absent. There are no instructions to delimit external content or ignore embedded commands.
Capability inventory: File reading is required for discovery and verification tasks.
Sanitization: Absent. The skill does not provide methods for validating or filtering ingested technical documentation.
[DATA_EXFILTRATION]: In the 'Product Discovery' section, the skill suggests that the agent check likely locations such as 'auth/security configuration' to verify product capabilities. Directing an agent to access security-sensitive configurations increases the risk of sensitive data exposure if the agent includes details from these files in its verification notes or summary output.

marketing-expert