The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection as it processes business questions alongside data retrieved from external research tools and local project files. Maliciously crafted content in external research sources (e.g., a website summarized by a research tool) could theoretically attempt to manipulate the agent's behavior. This ingestion is, however, fundamental to the skill's primary function of providing grounded, research-backed advice.
Ingestion points: Processes user-provided questions together with local files like project_*.md and portfolio.local.md, as well as transcripts and search results from tools like deep-research and watch-video.
Boundary markers: The instructions include strict grounding rules and simulation labels to maintain persona integrity, though they do not specify technical delimiters or explicit instructions to ignore embedded commands for the ingested external data.
Capability inventory: The skill has the capability to write session archives to the local filesystem (under ~/.config/makerskills/) and to invoke other agent skills such as decide or unstuck.
Sanitization: No specific evidence of sanitization, filtering, or escaping for the data interpolated from research tools or local files is present in the instructions.

maker-council