The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to execute shell commands via the ae-cli tool. It implements a strict security workflow for 'high-risk-write' capabilities (such as deletions), requiring the agent to perform a dry-run to inspect risks and then obtain explicit authorization from the user in chat before proceeding with execution.
[DYNAMIC_EXECUTION]: A template is provided that uses node -e to generate a local telemetry file (.ae/cli-agent-context.json). This involves interpolating user-controlled data, such as user_intent and session_goal, into a script string. While intended for context maintenance, this pattern requires the agent to ensure proper escaping to prevent script injection.
[DATA_EXFILTRATION]: The skill collects operational metadata, including user intent, session goals, and model identifiers, and writes them to a local JSON file to be used as telemetry by the ae-cli gateway. The instructions explicitly describe this as a telemetry feature for the tool's ecosystem and provide guidance on privacy options, such as using hashes for user intent or opting out if the user requests it.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to update its core tooling using npx @thinkingai/ae-cli@latest install. This command targets a package associated with the vendor's ecosystem for tool installation and maintenance.
[INDIRECT_PROMPT_INJECTION]: The skill performs discovery and inspection of external capability schemas and metadata. To mitigate the risk of malicious instructions embedded in these external sources, the collaboration.md reference instructs the agent to treat all retrieved content as evidence for evaluation rather than as new authoritative instructions.

ae-capability