The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script scripts/resolve-feature.sh using a --feature flag with a variable <key>. If this key is populated from unsanitized user input or external files, it could allow for command injection within the bash tool's environment.
[PROMPT_INJECTION]: The skill processes external design records and authority documents without sufficient isolation. 1. Ingestion points: design_records array and current_authority document. 2. Boundary markers: The skill lacks explicit separators or guardrails to prevent instructions within documentation from overriding the agent's behavior. 3. Capability inventory: The skill has access to Bash, Write, and Agent (subagent dispatch). 4. Sanitization: There is no indication that ingested content is sanitized or validated before processing.

feasibility-study