SHIFT
v1.0.4Manages multi-identity delegation to specialized AI sub-agents for coding, research, and quick tasks, routing and synthesizing responses seamlessly.
⭐ 0· 108·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The skill claims to route tasks to specialized sub-identities and indeed requests only the resources needed to do that: read/write under ~/.openclaw/workspace/.shift, read MEMORY.md and active files to build context, and spawn child sessions with model overrides. No unrelated credentials, system-level files, or external services are requested by the skill itself.
Instruction Scope
Runtime instructions explicitly tell the master to read ~/.openclaw/workspace/MEMORY.md and relevant active files, write per-run session files, attach context to sessions_spawn, and transmit those contexts to configured model providers. This is within the declared goal (giving sub-identities context) but means any secrets in workspace files or MEMORY.md may be sent to the configured model endpoints — the skill documents and warns about this.
Install Mechanism
Instruction-only install (no network downloads). Provided setup script creates ~/.openclaw/workspace/.shift, copies persona files, and initializes config and cost-tracking. No external URLs, no archive extraction from unknown hosts, and the manifest emphasizes local-only setup.
Credentials
The skill requests no environment variables or external credentials itself and uses the user's OpenClaw model-provider configuration (so provider creds are used but not stored by SHIFT). This is proportional to its function. However, it reads workspace files and transmits message content, conversation history (configurable last N turns), and active file snippets to model providers — users must consider whether those files contain secrets before delegating.
Persistence & Privilege
Does not require always:true or system-wide privileges. Writes and maintains files only under ~/.openclaw/workspace/.shift and updates cost-tracking.json there. It spawns child sessions (normal for its purpose) but does not modify other skills or system configs according to the metadata.
Assessment
SHIFT appears coherent for its purpose, but follow these precautions before enabling it: (1) Inspect the included files (especially scripts/setup.sh and personas/*.yaml) in ~/.openclaw/workspace/skills/shift to confirm behavior. (2) Be aware SHIFT will read ~/.openclaw/workspace/MEMORY.md and active project files and will send excerpts to whatever model providers you configure — do not delegate tasks that reference files containing API keys, passwords, or other secrets. (3) Start with costManagement.trackOnly: true and set contextBridge.historyTurns low (e.g., 3–5) while you test. (4) Prefer trusted or self-hosted model providers for sensitive data. (5) If you want extra assurance, review/limit which persona models are enabled and their keyword lists so delegation only occurs when you expect it.Like a lobster shell, security has layers — review code before you run it.
delegationvk973m6fvm33qwnkde7v9kwq9ss8342amidentityvk973m6fvm33qwnkde7v9kwq9ss8342amlatestvk973m6fvm33qwnkde7v9kwq9ss8342ammodelsvk973m6fvm33qwnkde7v9kwq9ss8342ammulti-agentvk973m6fvm33qwnkde7v9kwq9ss8342amroutingvk973m6fvm33qwnkde7v9kwq9ss8342am
License
MIT-0
Free to use, modify, and redistribute. No attribution required.