Token Saver 75+
ReviewAudited by ClawScan on May 10, 2026.
Overview
This skill has no code, but it asks your agent to silently delegate work and all context to other model agents, including a route with full tool access.
Review before installing. If you use it, keep it off sensitive conversations unless you add rules requiring confirmation before sessions_spawn, minimizing shared context, limiting spawned-agent tools, and checking provider/model costs.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Future conversations may be compressed or routed according to this skill rather than the user's immediate preference.
This asks the user to make the skill's routing policy authoritative for every session, which can change how future requests are handled even when the user did not explicitly invoke the skill.
Every session, read `skills/token-saver-75plus/SKILL.md` and follow its tier classification (T1-T4) for every request.
Only place it in AGENTS.md or a system prompt if you truly want always-on routing; otherwise keep it user-invoked or add an opt-out rule.
A spawned agent could use available tools or spawn additional agents in ways the user did not specifically approve, potentially increasing cost or affecting connected resources.
The Opus spawn pattern delegates work to another agent with 'full tool access' and permits nested spawning, without describing approval, tool limits, or containment.
task: "<full context + goal>. You have full tool access. Use sessions_spawn with Codex for code and Groq for bulk subtasks."
Require explicit user confirmation before spawning tool-capable agents, remove the 'full tool access' default, and restrict which tools and models spawned agents may use.
Sensitive conversation content or file excerpts could be sent to another model agent/provider as part of automatic routing.
The skill directs the agent to copy all relevant context into spawned-agent tasks, but does not define filtering for secrets, private files, or provider-specific data boundaries.
Include ALL context in the task string — spawned agents have no conversation history
Add a sensitive-data check before any sessions_spawn call, send only the minimum needed context, and ask the user before routing private or confidential material.
The agent may keep applying these routing and compression rules across tasks until the user removes or disables the instruction.
The skill is intentionally persistent in behavior when installed as recommended, though the artifacts do not show hidden background code or self-propagation.
You don't need a special command. The skill is meant to be **always-on** once installed.
Document how to disable it, and consider installing it as an optional skill rather than a permanent global instruction.