The Agent Skills Directory

[SAFE]: The skill establishes a transparent and safe workflow for creative and technical planning. No signs of malicious intent, obfuscation, or unauthorized activities were detected.\n- [PROMPT_INJECTION]: While the skill uses strong imperative language like 'You MUST' and 'HARD-GATE', these are procedural instructions designed to guide the agent through a specific brainstorming process. They do not represent attempts to override system safety protocols or bypass user-defined constraints.\n- [DATA_EXFILTRATION]: The skill references reading project context such as files, documentation, and commit logs. This data access is used locally to inform the design process and is not associated with any network exfiltration or unauthorized data sharing.\n- [COMMAND_EXECUTION]: The skill instructions involve standard development tasks such as writing markdown files to a 'docs/plans/' directory and committing them to a git repository. These operations are explicitly defined and align with the skill's functional goals.\n- [SAFE]: Analysis of potential indirect prompt injection surface: 1. Ingestion points: The skill reads project context from existing files, documentation, and git history (SKILL.md). 2. Boundary markers: No specific delimiters are used to separate user context from skill instructions. 3. Capability inventory: The skill can write to the local file system and transition to planning skills (SKILL.md). 4. Sanitization: No explicit input sanitization is defined. However, the risk is mitigated by the skill's requirement for incremental human validation and approval of all generated designs.

brainstorming