The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file references/research-sources.md contains instructions for the agent to recommend and install the agent-browser package globally using the command npm install -g agent-browser. This dependency is not from a verified or trusted vendor.
[COMMAND_EXECUTION]: The skill executes local commands, including the validation script python3 scripts/validate_launch_pack.py mentioned in SKILL.md and the shell-based interaction commands for the agent-browser tool in references/research-sources.md.
[PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection (Category 8).
Ingestion points: The skill researches and ingests data from external, untrusted sources including Reddit, X/Twitter, LinkedIn, and various competitor websites as defined in references/research-sources.md.
Boundary markers: The personality prompts for subagents in references/subagent-orchestration.md lack explicit delimiters or instructions to ignore malicious directives potentially embedded in the researched data.
Capability inventory: The agent has the capability to write files to the local file system and execute shell commands (python3, agent-browser).
Sanitization: No sanitization or validation of the retrieved web content is specified before the data is incorporated into the agent's context or used to generate copy.

viral-launch-pipeline