The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill collects free-text input from the user during the Phase 3 questionnaire (specifically Q19 for persona roleplay and Q20 for additional comments) and interpolates this content directly into the generated rules file. Because these files are intended to define the behavior of AI agents, unvalidated user input could be used to override the skill's generated constraints or inject malicious instructions into the agent's persistent configuration.
Ingestion points: User responses to Q19 and Q20 in SKILL.md.
Boundary markers: Absent. User input is written directly into the markdown structure.
Capability inventory: The agent writes generated content to various configuration files across the filesystem.
Sanitization: None. Input is treated as trusted content.
[EXTERNAL_DOWNLOADS]: Recommends the installation of the tool-using-virtuoso plugin from the author's GitHub repository (krzysztofsurdy/code-virtuoso) to provide the ecosystem map and decision matrix needed for full functionality.
[COMMAND_EXECUTION]: The skill performs filesystem operations to write generated instruction files to sensitive user and project locations, such as ~/.claude/CLAUDE.md, ~/.cursor/rules/, and .github/copilot-instructions.md, depending on the user's selected scope and target agent.

agentic-rules-writer