The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection during its research and onboarding phases.
Ingestion points: In references/brand-update.md, the agent is instructed to use the agent-browser tool to visit external websites to extract brand identity elements such as colors, typography, logos, and brand voice. Additionally, references/brief-analysis.md involves processing user-provided reference materials and external links.
Boundary markers: The instructions do not specify the use of delimiters or boundary markers to isolate or sanitize untrusted content extracted from web pages or external briefs before it is processed by the agent.
Capability inventory: The skill has capabilities to write to the local file system (via scripts/init-memory.py and scripts/render-template.py), invoke other agents for production tasks, and perform network operations through the agent-browser tool.
Sanitization: No sanitization or validation logic is defined for content retrieved from external sources.
[DATA_EXFILTRATION]: The skill facilitates the access and management of sensitive data and credentials.
Sensitive Data Exposure: As documented in references/tool-discovery.md, the skill provides instructions for capturing and reusing browser authentication states (including cookies and session data) via the agent-browser tool (e.g., agent-browser --auto-connect state save ./social-auth.json). While intended for authenticated research on social media platforms, these files contain highly sensitive credentials.
Credential Scanning: The scripts/tool-discovery.py script programmatically reads local configuration files (.pawbytes/config/config.yaml and .pawbytes/config/config.user.yaml) to verify the presence of API keys for integrated services like fal.ai and ElevenLabs.
[COMMAND_EXECUTION]: Multiple utility and test scripts included with the skill execute shell commands.
Tool Verification: scripts/tool-discovery.py uses subprocess.run to execute version checks for local binaries such as ffmpeg and agent-browser to confirm their installation and version status.
Test Execution: The test suite (scripts/tests/test-init-memory.py and scripts/tests/test-tool-discovery.py) utilizes subprocess.run to execute the primary Python scripts for integration and validation testing.

paw-cra-agent-creative-director