khe-decision
Warn
Audited by Gen Agent Trust Hub on Nov 22, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The validation script
scripts/decision-check.mjsdynamically executes Node.js scripts (brain-check.mjs) found within subdirectories of~/.claude/skillsor adjacent folders that match thekhe-brain-prefix. This lateral execution of code from other skills occurs without a trust verification or manual approval step for each found script, allowing for potential execution of malicious code if a deceptive skill is installed in the monitored paths. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to aggregate and synthesize data from multiple external 'brain' skills (external corpora), creating a significant attack surface for indirect prompt injection.
- Ingestion points: The skill reads
SKILL.mdandreferences/silent.mdfrom external skill folders and processes the output of parallel subagents spawned for each 'seated' brain. - Boundary markers: The skill uses structural tags like
== Seat: ... ==and== Chair ==to delineate sources, but it lacks explicit 'ignore embedded instructions' warnings when processing the corpora. - Capability inventory: The agent can execute shell commands via
spawnSyncand manage multi-step subagent chains, providing high-impact targets for successful injections. - Sanitization: While the
decision-check.mjsscript performs rigorous formatting and citation validation using regular expressions, it does not implement safety-specific filtering or sanitization of the content ingested from the external corpora.
Audit Metadata