ReAct Loop
v1.0.0Solve complex problems by interleaving reasoning with actions. Based on the ReAct research, this skill alternates between thinking (reasoning about the situa...
⭐ 0· 47·0 current·0 all-time
bySimon@simoncatbot
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The skill name/description (ReAct Loop) matches the SKILL.md instructions: it teaches an iterative Reason→Act→Observe workflow and provides examples (debugging, research, API workflows). It does not request unrelated binaries, env vars, or installs.
Instruction Scope
SKILL.md is a generic policy/template that instructs the agent to alternate reasoning and actions and includes examples that run commands (git log, analyze_logs, benchmark_search), query web/docs, and write code. That scope is coherent for debugging/research tasks, but the instructions intentionally allow broad tool use — so the agent could access repositories, logs, databases, or other resources depending on which tools are available. This openness is expected for a ReAct helper but increases operational risk if the agent has access to sensitive tools or environments.
Install Mechanism
No install spec and no code/binaries are included (instruction-only). There is nothing written to disk or downloaded by the skill itself, minimizing supply-chain risk.
Credentials
The skill declares no required environment variables, credentials, or config paths. Example workflows mention services that typically require secrets (e.g., Stripe, databases), but the skill does not request or store those credentials itself — any credential use would come from the agent's existing tool configuration.
Persistence & Privilege
always:false and no special persistence or system configuration changes are requested. Model invocation is allowed (platform default) but the skill itself does not demand permanent presence or elevated system privileges.
Assessment
This skill is coherent and low-risk as delivered (instruction-only, no installs or credentials). However, it is intentionally a broad, tool-driving template: before enabling it, review which tools and system permissions your agent has (git, database access, shell, web access, CI/CD, etc.). Limit the agent's permissions to least-privilege, avoid exposing production secrets, and test the skill in a sandbox or with read-only access first. If you plan to let the agent run commands or access systems, require explicit confirmation from you before destructive or credential-using actions.Like a lobster shell, security has layers — review code before you run it.
actingvk97dtp4g82gdd34r3hkpm055k9842h0clatestvk97dtp4g82gdd34r3hkpm055k9842h0cproblem-solvingvk97dtp4g82gdd34r3hkpm055k9842h0creasoningvk97dtp4g82gdd34r3hkpm055k9842h0cresearchvk97dtp4g82gdd34r3hkpm055k9842h0ctoolsvk97dtp4g82gdd34r3hkpm055k9842h0c
License
MIT-0
Free to use, modify, and redistribute. No attribution required.