The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several local shell and Python scripts (bootstrap.sh, sync.sh, index-memory.py) to automate setup, full-text search indexing, and synchronization tasks.
These scripts perform legitimate file system management and database operations within the designated ~/.agent-brain/ directory.
[DATA_EXFILTRATION]: The skill transmits memory files (summaries, knowledge, and experience layers) to pCloud (api.pcloud.com) to provide cross-session persistence.
This behavior is the primary intended function of the skill. The implementation specifically excludes the .env file (containing access tokens) and the STATE.md file from synchronization to maintain local security and privacy.
[EXTERNAL_DOWNLOADS]: The skill pulls stored memory and configuration data from pCloud to synchronize the local state with the cloud backup.
The data downloaded consists of markdown files and a derived SQLite database, which are processed locally rather than executed as scripts.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests and processes summaries of past sessions and project files which could contain untrusted content.
Ingestion points: The agent reads all markdown files within the ~/.agent-brain/ directory tree during the session initialization phase.
Boundary markers: No specific delimiters or safety instructions are currently implemented to isolate the ingested memory content from the agent's active instruction set.
Capability inventory: The skill context includes the ability to execute shell scripts, run Python code, and perform network requests to pCloud.
Sanitization: The skill does not perform validation or sanitization of the markdown files before they are read into the agent's context.

agent-brain