The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses the WebFetch tool to ingest untrusted content from public web sources (job boards, LinkedIn, podcasts) to extract industry-specific vocabulary and signals. This creates a surface for indirect prompt injection where malicious instructions embedded in the external text could attempt to manipulate the AI agent's behavior.
Ingestion points: External content fetched in modes/construct.md (Step 6) and modes/setup.md (Step 3).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when processing the retrieved web content.
Capability inventory: The skill has permissions to write to local project files (brand-config.json, SOUL.md) and execute a local Python script.
Sanitization: The instructions do not define sanitization or validation steps for the fetched strings before they are incorporated into prompts or saved to the filesystem.
[COMMAND_EXECUTION]: The skill executes a local Python script (score_psp.py) using the Bash tool to score and validate Pain Signal Profiles. The tool is configured with a restricted command pattern that only allows execution of this specific script within the plugin's installation directory.
[EXTERNAL_DOWNLOADS]: The skill retrieves business data and public signals from arbitrary URLs using the WebFetch tool for market research purposes.

psp