The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and installs the agentic-stack tool and the brain CLI from a personal GitHub repository (github.com/codejunkie99) via brew tap and git clone. This source is not associated with a verified organization or the skill author.
[REMOTE_CODE_EXECUTION]: The project migration workflow involves generating and executing a curl command to transfer memory. This pattern can be exploited to download and run arbitrary code from a remote server.
[DATA_EXFILTRATION]: The transfer feature provides a mechanism for exporting memory context, which may contain sensitive project details or user preferences, to external endpoints.
[DYNAMIC_EXECUTION]: Python tool scripts (e.g., brain_bridge.py) use sys.path.insert(0, ...) to dynamically import modules from the project's .agent/ folder, enabling the execution of code from potentially untrusted local directories.
[COMMAND_EXECUTION]: The skill relies on executing various CLI tools and shell scripts (install.sh, install.ps1, agentic-stack) to configure adapters and manage project state.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple memory layers and external queries to influence the agent's behavior.
Ingestion points: Data from files in .agent/memory/, episodic history, and brain ask results.
Boundary markers: Personal preferences use YAML frontmatter, but semantic and episodic layers lack explicit boundary markers to delimit content.
Capability inventory: File system access, execution of local Python scripts, and network operations via integrated CLI tools.
Sanitization: The skill does not document any sanitization, filtering, or validation of data retrieved from the memory layers before it is injected into the agent's context.

agentic-stack-portable-agent-memory