skills/gracefullight/krds/oma-design/Gen Agent Trust Hub

oma-design

Fail

Audited by Gen Agent Trust Hub on Oct 14, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_CONTEXT_INJECTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_CONTEXT_INJECTION]: The file SKILL.md contains the pattern !bunx getdesign@latest list. This dynamic context injection executes a shell command that performs a network request to the npm registry at the moment the skill is loaded or opened by the agent.
  • [REMOTE_CODE_EXECUTION]: The skill executes remote code by using bunx and npx to run community packages (getdesign, @_davideast/stitch-mcp) that are not from trusted organizations. These packages are fetched and executed at runtime without version pinning.
  • [REMOTE_CODE_EXECUTION]: The resources/getdesign-fetcher.md file implements a shell pipeline that dynamically resolves a tarball URL from the npm registry and pipes the remote content directly into tar for extraction (curl -sL "$TARBALL" | tar -xzO ...). This pattern is high-risk as it processes unverified remote data streams into executable context.
  • [COMMAND_EXECUTION]: The skill makes extensive use of shell primitives and CLI tools including bunx, npx, curl, tar, shasum, rm, and gcloud to perform its design management tasks. This involves significant interaction with the host operating system.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests external markdown templates from a community catalog (getdesign). Because these templates are processed by the LLM to synthesize design documents, they represent a surface for indirect prompt injection. The skill provides framing instructions to mitigate this risk, but the attack surface remains.
  • Ingestion points: External design templates fetched via bunx getdesign and loaded into the context in Phase 2.
  • Boundary markers: The skill uses a framing prompt in resources/getdesign-fetcher.md to instruct the agent to treat the content as reference data only.
  • Capability inventory: The skill possesses file write, network (curl), and command execution (bunx/npx) capabilities across its scripts.
  • Sanitization: The skill uses SHA256 hash verification to ensure template integrity against the manifest, but does not sanitize the markdown content for malicious instructions.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 14, 2026, 06:51 PM