The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates a tasks.json file containing a nextActionPrompt field which is used to guide future agent actions. This prompt directly interpolates untrusted user input.\n
Ingestion points: Workflow Step 2 in SKILL.md collects project context (topic, goal, hypothesis) directly from user conversation and existing project files like research_brief.json and instance.json.\n
Boundary markers: The nextActionPrompt template defined in references/generation-rules.md lacks explicit delimiters or instructions to ignore embedded commands within the <relevant extracted values> placeholder.\n
Capability inventory: The skill possesses the write_file tool, which it uses to persist these potentially malicious prompts into the project workspace for execution by other agents.\n
Sanitization: There is no evidence of sanitization, filtering, or validation of user-provided content before it is embedded into the generated prompts.

inno-pipeline-planner