software-backend

Fail

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The documentation resources in data/sources.json and several reference files include a link to https://basarat.gitbook.io/typescript/ which has been flagged as a malicious phishing URL by automated security scanners. Users should exercise caution when visiting external documentation links referenced in the skill.
  • [COMMAND_EXECUTION]: In SKILL.md, the instructions require the agent to execute a Python script (append_learning.py) located in a sibling directory (agents-skills-feedback-loop/). This dependency on unverified external files outside the skill's own package introduces a risk of arbitrary command execution if the external directory is compromised.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify user-provided source code and configuration files, creating a surface for indirect prompt injection. Since the skill has the authority to run build tools and modify files, malicious instructions hidden in processed code could influence agent behavior. • Ingestion points: User project source code, database schemas, and documentation files as described in the tech stack templates. • Boundary markers: Absent; templates do not include delimiters or specific instructions to the agent to ignore commands within user code. • Capability inventory: Extensive capabilities to modify the file system and execute shell commands through language package managers (npm, pip, cargo, dotnet, etc.). • Sanitization: Absent; the skill lacks a mechanism to sanitize or filter out instructions embedded within the ingested data before processing.
Recommendations
  • Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Oct 6, 2026, 02:03 PM