Universal prompt addenda that make workspace files work across different LLM providers when using fallback chains. Addresses documented failure modes per mod...
Security Analysis
high confidenceInstruction-only prompt-engineering guidance that matches its description, requests no credentials or installs, and is proportionate to its stated goal of adding short guardrails to workspace files for multi-model fallback.
The skill name/description (cross-model prompt addenda) matches the instructions: it only recommends brief text blocks to add to workspace files. It requests no binaries, env vars, or installs, which is consistent with being an instruction-only prompt guide.
SKILL.md stays on-topic: it tells the operator what short guardrail snippets to add to AGENTS.md, TOOLS.md, and similar files. It does not instruct the agent to read unrelated system files, exfiltrate data, call external endpoints, or access credentials. It asks to avoid model-detection logic and to keep addenda short.
No install spec and no code files — lowest-risk distribution model. Nothing is downloaded or written by the skill itself.
The skill declares no required environment variables, no primary credential, and no config paths. The content even recommends guarding against leaking phone numbers/API keys in workspace files.
always: false and no installation artifacts. The skill does not request persistent presence or elevated privileges and does not attempt to modify other skills or system settings.
Guidance
This is an instruction-only prompt-engineering guide and appears internally consistent and low-risk. Before applying: (1) Review and edit the suggested addenda yourself — do not copy/paste blindly; (2) Keep each addendum very short (as recommended) to avoid token cost and unintended interactions; (3) Ensure your workspace files (AGENTS.md, TOOLS.md, etc.) do not contain secrets (API keys, phone numbers, JIDs, allowlists) — the skill explicitly warns about this; (4) Test changes in a staging environment to confirm the fallback model behavior improves and that your primary model is unaffected; (5) If an agent or tool proposes automated edits, review them before committing. No credentials or installs are required, so there is no hidden exfiltration vector in the skill itself.
Latest Release
v1.0.0
- Initial release of model-prompt-adapter. - Provides universal prompt addenda to ensure workspace files work reliably across different LLM providers in fallback scenarios. - Addresses known model-specific failure modes including prompt leaking, scope creep, fabricated completion, and privacy risks. - Adds concise guardrail sections for AGENTS.md, TOOLS.md, and tool-related files to improve cross-model compatibility. - Helps users avoid maintaining duplicate prompt files and reduces errors when switching between Claude, GPT, and Gemini models.
More by @globalcaos
Published by @globalcaos on ClawHub
