Chaos pivot
v1.0.0Prevents LLMs from sunk-cost pushing broken solutions. Triggers when an agent is stuck, looping, or failing repeatedly. Forces a Popperian falsification mome...
⭐ 0· 625·0 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
Name, description, and runtime instructions all align: the skill teaches an agent to recognize dead-ends, produce three radically different alternatives, probe them, and pick or escalate. No extraneous permissions, binaries, or services are requested.
Instruction Scope
SKILL.md confines itself to internal reasoning steps (declare failure, produce 3 alternative lenses, probe each). It does give the agent broad discretion to 'attempt' alternatives — which could involve external actions depending on the agent's capabilities (API calls, CLI, or other tools the agent already has). The instructions do not explicitly tell the agent to read unrelated files, exfiltrate secrets, or contact unexpected endpoints, but the open-ended 'attempt' step means behavior depends on the agent's other tool permissions; consider restricting what the agent may do when probing alternatives.
Install Mechanism
No install spec and no code files. Nothing is written to disk and there is no third-party code to fetch. Low installation risk.
Credentials
The skill requires no environment variables, credentials, or config paths. There is no request for extra secrets or unrelated service tokens, which is proportional to the skill's described function.
Persistence & Privilege
Flags: always:false and user-invocable:true (default). The skill does not request permanent inclusion or system-level configuration changes. It does not modify other skills or agent settings.
Assessment
This skill is internally coherent and low-risk: it only contains instructions for how an agent should recognize a dead end and try three very different alternatives. Before enabling it, confirm the agent's permitted actions when 'attempting' alternatives (e.g., whether it can call external APIs, run commands, or access files). Keep it user-invocable (do not set always:true), and consider adding operational guards: require explicit user approval before attempts that perform network calls or destructive actions; log pivot summaries and results for auditability; and limit randomness in safety-critical contexts. If you want tighter control, restrict the agent's tool permissions or modify the SKILL.md to explicitly disallow external side effects.Like a lobster shell, security has layers — review code before you run it.
design thinkingvk974act53vgvd5rxjvtfz7cnmh81fn22failurevk974act53vgvd5rxjvtfz7cnmh81fn22latestvk974act53vgvd5rxjvtfz7cnmh81fn22pivotvk974act53vgvd5rxjvtfz7cnmh81fn22
License
MIT-0
Free to use, modify, and redistribute. No attribution required.