The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses highly imperative language in the <EXTREMELY-IMPORTANT> section to override standard agent behavior, stating the framework is a 'mandatory workflow' that cannot be skipped, and should be used 'before any response or action'.
[EXTERNAL_DOWNLOADS]: The instructions direct users to install the livepowers package via npm (npx livepowers install or npm i -g livepowers). This is a standard installation procedure for the framework's tools and originates from the vendor's expected package name.
[COMMAND_EXECUTION]: The framework relies on executing several Python scripts (lp.py, agent_switch.py, scan_sql.py) to perform project management tasks, such as scanning database schemas, auditing agent communications, and managing status machines.
[DYNAMIC_EXECUTION]: The 'crystallization' process (crystallize-to-system1) involves the generation of code or CLI tools based on successful agent explorations. While this involves creating executable content at runtime, the skill includes safety gates such as writing to a 'pending' list for manual adoption.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it is designed to ingest data from untrusted external environments (databases and system schemas) via the env-scan-ontology and scan_sql.py tools.
Ingestion points: Environment and database scanning tools (scan_sql.py, scan_sqlite.py) read external schema data into the agent's context.
Boundary markers: The skill does not explicitly define delimiters or warnings to ignore instructions embedded in the scanned data.
Capability inventory: The skill has the ability to write to the file system (.livepowers/), execute shell commands (lp.py), and query external databases.
Sanitization: No specific sanitization or escaping mechanisms are mentioned for the data retrieved during environment scans.

using-livepowers