The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the execution of a curl POST request to http://localhost:8888/notify at the start of every session. While targeting localhost, this establishes an automated shell execution pattern.
[COMMAND_EXECUTION]: The CreateSalesPackage.md and CreateVisual.md workflows execute a local script (Generate.ts) via the bun runtime, using command-line arguments derived from processed user input.
[PROMPT_INJECTION]: The skill instructions in SKILL.md direct the agent to load and apply configuration files from a specific user directory (~/.opencode/skills/PAI/USER/SKILLCUSTOMIZATIONS/Sales/) to override default behavior, creating a vector for persistent behavioral modification.
[PROMPT_INJECTION]: The use of mandatory execution instructions ("🚨 MANDATORY", "This is not optional") is a technique often used to bypass agent safety filters or reasoning by forcing immediate action.
[PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by processing untrusted documentation into prompts for an image generation tool.
Ingestion points: Product documentation and specifications input in CreateNarrative.md and CreateSalesPackage.md.
Boundary markers: None; the content is transformed into visual descriptions and talking points without delimiters.
Capability inventory: Subprocess execution via bun run and shell execution via curl.
Sanitization: No input validation or escaping is performed on the documentation before it influences the generated output.

Sales