The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates data from external files into the agent's operational context, creating a surface for indirect prompt injection.
Ingestion points: The skill reads data from data/engineering/tech_stack.json, data/product/roadmap.json, and data/cfo/latest_forecast.json in SKILL.md.
Boundary markers: Absent. There are no delimiters or instructions provided to the agent to treat the content of these files as untrusted or to ignore embedded instructions.
Capability inventory: The skill has the capability to write files to the local file system (data/engineering/adrs/).
Sanitization: No sanitization, validation, or escaping of the input data is performed before it is used to generate the final ADR output.
[DATA_EXFILTRATION]: The skill accesses sensitive business data by reading financial forecasts from data/cfo/latest_forecast.json. While no explicit network exfiltration patterns were detected, the access to financial data represents a high-value target for potential exposure if the agent's context is leaked.
[COMMAND_EXECUTION]: The skill performs automated file system operations, specifically writing generated markdown content to the data/engineering/adrs/ directory. This write capability, combined with user-influenced arguments ($ARGUMENTS), could be misused to overwrite existing records if not carefully handled by the agent.

architecture-decision