The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external sources (Step 3: Capture raw sources via ingest). This data is then processed and synthesized into local research articles, which could lead to an indirect prompt injection if the external source contains malicious instructions designed to influence the agent's behavior during the analysis or writing phases.
Ingestion points: Step 3 uses tools like WebFetch or WebSearch to pull external content into the Knowledge Base.
Boundary markers: The instructions emphasize citations and verbatim capture but do not explicitly enforce isolation markers to prevent the agent from following instructions embedded within the fetched sources.
Capability inventory: The skill uses exec for shell commands (grep, ls, cat) and performs file-write operations to the local file system.
Sanitization: No specific sanitization or filtering of the external content is described within this skill's instructions.
[COMMAND_EXECUTION]: The skill utilizes the exec function to perform shell operations for searching and reading files within the local content directory (e.g., exec("grep -rln ..."), exec("ls -A ..."), exec("cat <path>")). While these are used for legitimate project management tasks, the execution of shell commands based on user-provided or externally derived keywords represents a capability that must be monitored.

research-with-sources