The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to install the software using a 'curl | bash' pattern: curl -fsSL https://raw.githubusercontent.com/bernardohcrocha/persistia-for-claude-code/main/setup.sh | bash. This executes a remote script from an untrusted repository (not belonging to the skill's author) without any opportunity for review, granting the script full control over the user's environment.
[DATA_EXFILTRATION]: The skill is explicitly designed to crawl the project directory and ingest highly sensitive information, including .env files, environment variables, Git configurations, and credentials for tools like AWS and Stripe. It then facilitates the transfer of this 'brain' data to external GitHub repositories.
[CREDENTIALS_UNSAFE]: The skill intentionally targets and processes high-value secrets such as STRIPE_SECRET_KEY, DATABASE_URL, and AWS access keys. It provides examples of JavaScript tasks that use these credentials to perform external operations.
[COMMAND_EXECUTION]: The skill establishes persistence by installing system services (systemd or launchd) and regularly executes shell commands to monitor project changes and run scheduled autonomous tasks.
[PROMPT_INJECTION]: The skill creates a massive attack surface for indirect prompt injection.
Ingestion points: The skill reads the entire project directory, including untrusted content in README.md, documentation, and .env files (SKILL.md).
Boundary markers: None identified; external data is ingested directly into the 'brain'.
Capability inventory: The skill has extensive capabilities including shell execution (git, launchctl), file system modification, and network access (SKILL.md).
Sanitization: No evidence of sanitization or validation of the content read from project files before it influences agent behavior.

persistia-claude-code-memory