The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator uses a TypeScript driver (scripts/loop.ts) that invokes sibling scripts like gf.ts, scout.ts, and collect.ts using node:child_process.spawnSync. These executions are performed by passing arguments as an array to the bun runtime, which is a secure pattern that prevents shell command injection. The scripts are resolved relative to the skill's installation directory or via a configurable environment variable (GENFEED_SKILLS_DIR), consistent with standard practices for modular developer tools.
[CREDENTIALS_SAFE]: Instructions within SKILL.md mandate that authentication tokens must never touch the disk. Instead, they are resolved at runtime through a dedicated connector (genfeed-connector) and injected into the environment of worker processes. This practice effectively mitigates the risk of credential exposure through logs or temporary files.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface where it processes trend signals from external sources via the trend-scout worker. However, the risk is minimized as the data is handled programmatically (JSON parsing) in the helper script, and the skill documentation explicitly requires a human approval gate before any generated content is published to social platforms. The use of structured data transitions via the gf seam provides clear boundaries for external content.

content-loop-orchestrator