The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The troubleshooting documentation recommends an installation pattern: curl -fsSL https://<PCKLE_HOST>/install.sh | sh. This practice of piping remote scripts directly into a shell is highly dangerous as it can lead to arbitrary code execution. Because the host is specified as a variable template, an AI agent could be manipulated into fetching and executing a malicious script from an untrusted domain.
[COMMAND_EXECUTION]: The skill leverages the Bash tool to execute pckle CLI commands. Several commands (e.g., agent create, workflow create) support reading local files using an @ prefix (e.g., --setup @setup.sh). This functionality could be exploited to read and transmit sensitive local system files, such as SSH keys or configuration files, to the remote API if the agent is provided with malicious file paths.
[EXTERNAL_DOWNLOADS]: The skill requires network access to the PCKLE API (defaulting to https://alpha.pckle.io) and suggests downloading binaries from remote servers.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core functionality of ingesting and retrieving external knowledge:
Ingestion points: Untrusted data enters the context through pckle workflow create --input and via the results of knowledge retrieval commands like pckle workflow get or pckle workflow create --wait.
Boundary markers: The instructions lack boundary markers or specific directives to ignore instructions embedded within the retrieved knowledge base content.
Capability inventory: The skill has access to the Bash tool, which can perform network operations and read/write files via the CLI's specialized flags.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from the remote knowledge index before it is processed by the agent.

pckle