T01 · Skill Instruction Hijacking
Error
- Location
SKILL.md:31- Finding
Always-On Instruction and Model-Routing Hijack
- Content
View full analysis
3 items)** → spawn Groq - **Complex multi-step tasks** → spawn Opus (T4) - **Simple formatting/rewriting** → spawn Groq ``` `SKILL.md:68-72`: ```markdown ### Critical spawn rules 1. **Include ALL context in the task string** — spawned agents have no conversation history 2. **Be specific** — vague tasks waste tokens on clarification 3. **One task per spawn** — don't bundle unrelated work 4. **For code: always use Codex** — never write code yourself ``` `README.md:26-30`: ```markdown Then reference it in your `AGENTS.md` or system prompt: ```markdown ## Mandatory: Token Optimization Every session, read `skills/token-saver-75plus/SKILL.md` and follow its tier classification (T1-T4) for every request. ``` ``` ### Technical Analysis The installation instructions encourage placing the Skill into `AGENTS.md` or the system prompt as a mandatory rule for every session. Once loaded, the Skill silently classifies every user request and imposes mandatory model-routing decisions. It also directs the Agent never to generate code itself and to delegate all such work to a specific external model. These instructions exceed what is minimally necessary to provide concise, token-efficient responses. Output compression can be implemented without taking persistent session-wide control of model selection, tool usage, or task execution. The Skill provides no exception for confidential requests, user-selected routing, unavailable providers, organizational policy, or higher-priority safety constraints. Although the files do not obtain operating-system privileges, the Skill acquires broad control over the Agent's decision-making and delegati ...[truncated 1447 chars]- Remediation
View remediation