plan
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various external sources, including GitHub Actions CI logs from provided URLs, web content for conversion optimization analysis, and flattened remote repositories via the repomix tool. This ingestion of external, potentially attacker-controlled content creates a surface for indirect prompt injection, where malicious instructions could attempt to influence the agent's planning, implementation, or verification logic.
- Ingestion points: Found in
references/mode-ci.md(CI logs),references/mode-cro.md(web content/URLs), andreferences/engine-research.md(remote repository flattening). - Capabilities: The skill possesses significant capabilities, including code execution via the tester subagent and repository management via the git-manager subagent.
- Sanitization: There is no evidence of explicit sanitization or filtering of the ingested external content to prevent command injection or instruction overriding.
- [COMMAND_EXECUTION]: The skill executes various local project scripts and external CLI tools to perform its functions. While these appear to be part of the skill's intended software engineering workflow, they involve the execution of arbitrary commands.
- Local scripts: Invokes
.claude/scripts/code_graphvia Python and.claude/scripts/ai-signal-scan.cjsvia Node.js for project analysis. - External tools: Utilizes the
geminiCLI for technical research andrepomixfor repository processing. - Configuration loading: Uses
node -eto load project configuration from.claude/hooks/lib/ck-config-loader.cjs. - [EXTERNAL_DOWNLOADS]: The skill frequently fetches data from external network locations to support analysis tasks.
- CI analysis: Downloads build and test logs from external GitHub Actions URLs to identify failure causes.
- CRO analysis: Uses the
web_fetchtool to retrieve page content from provided URLs for optimization analysis. - [DATA_EXFILTRATION]: The skill's codebase understanding procedures include the analysis of environment configuration files which often contain sensitive data.
references/engine-codebase-understanding.md: Instructions explicitly state to "Analyze dotenv files and configuration" to understand the environment and dependencies. This involves accessing sensitive file paths typically associated with credential storage.
Audit Metadata