The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied business context, competitor names, and strategic objectives to generate analysis.
Ingestion points: The competitive_scenario tool in server.js and the navigation guides in SKILL.md accept natural language inputs.
Boundary markers: The skill uses a rigorous INPUT_CONTRACT.md and OUTPUT_SCHEMA.md to define formatting and safety constraints, explicitly warning against fabricating data.
Capability inventory: The server reads local markdown files from the scenarios/ and chatgpt/ directories to provide templated context.
Sanitization: Inputs are parsed for keyword extraction and classification; the skill does not execute user-provided text as shell commands or code.
[EXTERNAL_DOWNLOADS]: The skill specifies standard Node.js dependencies for its server component.
package.json includes well-known libraries such as the official @modelcontextprotocol/sdk, express, cors, and zod sourced from the public npm registry.
[DATA_EXPOSURE]: The server script (server.js) reads files from the local project directories to serve the agent's knowledge requests.
This access is restricted to the skill's own scenario cards and configuration files and does not attempt to access sensitive system paths or user credentials.

competitive-intelligence-market-research