video-publisher

SUSPICIOUS. The skill’s browser automation scope is broadly aligned with its stated purpose and it includes meaningful guardrails against accidental final publication, but it relies on an external Ego Lite/ego-browser runtime that appears official yet unverifiable from the provided evidence. Because that runtime receives authenticated browser state and may ingest migrated cookies/passwords/sessions, the skill carries high security risk despite lacking clear evidence of malicious intent.

This fragment is primarily orchestration logic, not an obvious standalone malware implant. However, it creates a significant supply-chain/execution-integrity risk by executing a runner entrypoint chosen via process.env.VIDEO_PUBLISHER_V2_RUNNER and by forwarding the entire host environment to the child process (potentially exposing secrets to a compromised runner). Child stdout/stderr is treated as authoritative input for workflow gating and persisted verdicts, so runner compromise or tampered output would have strong downstream effects. Full risk confirmation requires reviewing runCapture/run-platform.mjs and the parsing/validation logic (parseV2Result/evaluateObservation) that consume child output.

This module primarily orchestrates a dynamic execution flow: it concatenates local .mjs source contents with a generated header containing environment-variable values, then passes the resulting fragments string into runEgo. The strongest security concern is the fragments-to-execution design (potential code/behavior injection depending on runEgo/ringify safety). Separately, it may leak sensitive evidence because it logs and returns detailed tail text from runner stdout/stderr and error stacks. No explicit malware actions (e.g., exfiltration, destructive operations) are visible in the provided snippet, but the execution pipeline and env-driven code composition make this moderately risky pending review of runEgo, ringify, and parseV2Result.