lwy-project-skill-writer
Pass
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill employs a local Node.js utility (
scripts/cli.cjs) to handle the initialization and scaffolding of new skill files. This script implements logic for parsing user problems, classifying skill types, and rendering templates into the project workspace. - [DATA_EXFILTRATION]: The skill utilizes several sub-agents (
project-scanner.md,tech-stack-analyzer.md,convention-detector.md) to read and analyze local project files. This data is used to extract coding conventions, technology stacks, and structural patterns to ensure the generated skills are context-aware and consistent with the codebase. No data is transmitted to external servers. - [PROMPT_INJECTION]: The skill presents an Indirect Prompt Injection surface (Category 8) because it processes untrusted user descriptions and project code which are then interpolated into the generated
SKILL.mdfiles. - Ingestion points: User-provided problem descriptions via the
--problemflag and project source code analyzed during the design phase. - Boundary markers: The generation template (
assets/skill.md.template) lacks explicit delimiters or 'ignore' instructions around interpolated variables like{{DESCRIPTION}}and{{USE_CASES}}. - Capability inventory: The skill can create new instruction files (SKILL.md) and supporting scripts/assets within the project's
.agents/skills/directory. - Sanitization: The provided scripts do not appear to perform sanitization or escaping of input data before it is written into the Markdown templates.
Audit Metadata