Halts thrashing agent behavior and enforces a three-question diagnostic gate before work continues.
npx skills add https://github.com/tschallacka/ai-skills --skill www-tschallacka
Please help me install this Agent Skill. Skill: www Source: https://github.com/tschallacka/ai-skills/tree/main/www Command: npx skills add https://github.com/tschallacka/ai-skills --skill www-tschallacka
www, or the agent recognizes its own thrashing, and all speculative action stops until the situation is properly assessed. ## Core Features & Use Cases - Dual trigger mechanism: The human typing www stops all tool calls immediately, and the agent applies the same brake on itself when it detects thrashing behavior. - Three-question gate: Work only resumes after answering, in order: What do we have (measured facts), What are the values (concrete paths, ports, exit codes), and What are we trying to achieve (the restated goal). - Anti-guessing discipline: Answers must cite something actually checked, not beliefs; unmeasurable facts must be measured before continuing, and the next action is one reasoned step or a numbered question to the human. - Use Case: An agent has tried five variations of a build command that keeps failing. Instead of a sixth guess, it stops, confirms the real error text, lists the actual flags and exit code, restates the goal, and then takes one informed step. ## Quick Start Type www to stop the agent immediately and require it to answer what we have, what the values are, and what we are trying to achieve before continuing.High-intent search queries and answers about installing and using this skill.
Type www and the agent stops all tool calls immediately. It must then answer three questions in order - what do we have, what are the values, what are we trying to achieve - before taking a single reasoned step.
The three questions are: What do we have (measured facts versus assumptions), What are the values (concrete paths, ports, ids, exit codes), and What are we trying to achieve (the restated goal). All three must be answered in order.
Yes. When the agent notices it is thrashing - retrying command variants, re-reading the same files, or guessing at unmeasured causes - it pulls the same brake on itself without waiting for the human to type www.
Do not use it for ordinary uncertainty that a quick check resolves, or when the answers to all three questions are already known. It is a brake for thrashing, not a routine pause.
A valid answer names something actually checked, such as an error message confirmed by running a command. Guesses like 'probably a timing issue' do not count; unmeasured facts must be measured before work continues.
TalonT-OrgFetch CI logs for a failing Git branch and classify failure types.
leonardoteodorooRun fresh verification commands and validate outputs before claiming completion.
cookysRecord reusable lessons and fixes into MEMORY.md for future sessions.
danielmiesslerInvestigates failures using Five Whys, Fishbone, Postmortem, Fault Tree, and Kepner-Tregoe methods.