The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The run.mjs script utilizes npx to download and execute the rightstack package from the public npm registry. The risk associated with external dependencies is mitigated by pinning the tool to version 0.3.2 and explicitly disabling npm lifecycle scripts via npm_config_ignore_scripts: "true" in the child environment.
[COMMAND_EXECUTION]: Shell commands are executed via node:child_process's spawnSync function. The implementation correctly passes arguments as an array and avoids shell interpolation, effectively preventing command injection from user-provided input variables. The skill also invokes a local ./notify script to send output.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests data from local memory logs and the output of the external rightstack tool.
Ingestion points: memory/MEMORY.md, memory/logs/, and the JSON output of the rightstack CLI.
Boundary markers: The instructions mandate schema validation (schema_version: "1.0") and JSON parsing before processing content.
Capability inventory: The skill has the capability to execute the npx command and a separate ./notify script.
Sanitization: The agent is explicitly instructed to treat the tool output as advisory evidence rather than authoritative commands, and to verify all package versions and architectural claims before including them in the final brief.

rightstack