tech-stack-research
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Node.js script located at
.claude/scripts/ai-signal-scan.cjsto identify AI-related components in the tech stack. Additionally, the skill utilizes thespawn_agentcapability to authorize and manage sub-agents for complex architectural tasks. - [DYNAMIC_EXECUTION]: The skill dynamically invokes local project scripts and reads protocol files (
.claude/skills/shared/protocols/ai-feature-framing-gate.md) at runtime to adjust its behavior based on the detected technology stack components. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted data from external sources, creating a vulnerability surface where malicious instructions could be embedded in researched content.
- Ingestion points: Data enters the context via
WebSearchresults,WebFetchof external articles/benchmarks, and project-specific files likedocs/knowledge/strategy/business/andplans/. - Boundary markers: The skill uses a tiered sourcing model (Tier 1 to Tier 4) and requires weighted confidence scores (e.g., "Idea should be more than 80%") to evaluate information, which acts as a logical boundary for decision-making.
- Capability inventory: The agent has the ability to write files to the project directory (
{plan-dir}/research/), execute local shell commands (node), access the network via search/fetch tools, and spawn sub-agents. - Sanitization: No explicit programmatic sanitization or escaping of web content is defined; however, the skill mandates manual verification using multiple independent sources for critical claims to reduce the risk of following injected instructions.
Audit Metadata