The Agent Skills Directory

[METADATA_POISONING]: The skill uses deceptive and misleading references to establish false authority. Specifically, prompt-evaluation.md cites "OpenAI's guidance on skills and prompts for GPT-6 Astra" with a link to a blog post for a technology that is currently non-existent. Additionally, validation-checklist.md references a "SkillLens rubric" citing a futuristic academic paper from May 2026 (arXiv:2605.23899).
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied instructions to generate agent files, creating an attack surface for indirect injection. 1. Ingestion points: User-provided skill goals, triggers, and boundaries in SKILL.md. 2. Boundary markers: Lacks formal delimiters for untrusted input during authoring. 3. Capability inventory: Uses Write and Edit tools and executes oma and git commands. 4. Sanitization: Relies on oma skill lint and manual checklists.
[COMMAND_EXECUTION]: The skill directs the agent to execute project-specific CLI tools including oma skill lint, oma skill audit, oma skill eval, and git diff. It also manages subagent orchestration through oma agent spawn and opencode run, which involve process execution based on local configuration.

oma-skill-creation