The Agent Skills Directory

[SAFE]: The skill is composed strictly of documentation and process instructions. No executable code, scripts, or binary files are present in the package.\n- [SAFE]: The instructions include clear authority constraints that restrict the agent to a 'discovery' role. It explicitly forbids implementing code, publishing to trackers, installing software, or delegating tasks without separate authority.\n- [NO_CODE]: Analysis of all files (SKILL.md, README.md, references/validation.md) confirms the absence of runtime dependencies, network operations, or shell command execution patterns.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles project data and user input, which is a standard surface for indirect prompt injection. This is considered safe as the skill's primary purpose is clarification and it explicitly lacks any tools or capabilities that could be exploited for unauthorized actions. Evidence chain: 1) Ingestion points: Untrusted data may enter through project documents and user responses during discovery; 2) Boundary markers: The skill lacks technical delimiters but provides instructions to separate documented facts; 3) Capability inventory: No tools for file system modification, network exfiltration, or code execution are provided; 4) Sanitization: No explicit sanitization or escaping of external content is present.

shape-work