The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the use of an ExecutionProxy Interface (default prefix rtk) for all terminal actions. It specifically instructs the agent to execute shell-level commands such as rtk npm test to fulfill its 'Iron Law of Execution' (TDD) protocols.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its core functionality of ingesting and analyzing untrusted external data.
Ingestion points: The agent is instructed to process data from various external sources including keyword exports (Ahrefs, SEMrush, GSC), customer call transcripts, survey responses, and web search results from community platforms like Reddit and Quora (specifically via site:reddit.com queries).
Boundary markers: No specific delimiters or 'ignore embedded instruction' warnings are provided to isolate the research data from the agent's primary instruction set, though the skill does mention redacting secrets/PII before sharing context with subagents.
Capability inventory: The agent has access to a command execution proxy (rtk prefix) and file system access to the project's docs/ directory and package.json file.
Sanitization: The instructions do not define any sanitization, validation, or filtering protocols for the external content before it is processed by the agent.
[DATA_EXFILTRATION]: The skill directs the agent to access internal project documentation and architectural metadata, specifically docs/departments/Growth/product-marketing-context.md and docs/graph.json. While these files are internal to the project, this represents an automated access pattern to business strategy and project structure information.

content-strategy