Token Saver Skill
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill broadly matches its token-saving purpose, but it automatically processes and caches chat context through an undeclared core module and makes restore/rollback claims the visible code does not support.
Before installing, verify the missing @token-saver/core package and treat caching/rollback claims as unproven. Avoid sensitive conversations until cache storage and retention are clear, and use /tokenoff or /tokenquality for important work.
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.
Incorrect compression or cache matching could cause the agent to answer from altered or incomplete context.
The skill automatically reads and transforms the current conversation context before AI requests. This is purpose-aligned, but it directly affects what context the model receives.
/* Main optimization method - called before each AI request */ ... const context = this.api.getConversationContext(); const result = this.saver.optimize(context);
Use /tokenquality or /tokenoff for precision-critical or sensitive conversations, and verify important answers against the original context.
The code that decides what chat content to compress, cache, or reuse is outside the reviewed artifacts.
The visible integration delegates the main optimization and cache logic to @token-saver/core, but the provided artifacts include no package manifest, lockfile, install spec, or @token-saver/core source.
import { TokenSaver } from '@token-saver/core'; ... this.saver = new TokenSaver({ enabled: true, showIndicator: true, defaultMode: 'adaptive' });Install only if you trust the publisher, or ask for the core package source plus pinned dependency/version information before using it.
Sensitive chat details or stale cached responses could be retained and reused in later similar queries.
The skill intentionally caches conversation-derived query/response information. Combined with the manifest's storage permission and missing core implementation, the artifacts do not clearly show cache storage boundaries, per-chat isolation, or retention enforcement.
| **Semantic Cache** | Multi-level caching (exact → semantic → pattern) | 30-50% | ... "cache": { "enabled": true, "ttl": 3600 }Avoid using this skill for highly sensitive chats unless cache scope and retention are documented; use /tokencache clear regularly.
Users may rely on restore and rollback protections that are not demonstrated in the provided implementation.
These safety assurances are important because the skill changes context, but the visible integration registers only a /tokens status handler and does not show a /tokens original restore path; the claimed rollback behavior depends on unprovided code.
- **Auto Rollback**: If quality drops > 15%, restores original context - **One-Click Restore**: `/tokens original` to see uncompressed version - **Snapshots**: Every compression creates rollback point
Treat the rollback and restore claims as unverified until the publisher provides matching implementation evidence or tests.