qwenchance — Claude Code skill: escape loops and track context
A skill for staying on track during long multi-step work: detects when you're repeating actions, stops endless second-guessing, and monitors context budget. Activates when the model is stuck in circles or running out of window space.
npx skills add thananon/9arm-skills --skill qwenchance -a claude-codeWhat qwenchance does
Long work in an editor hits three failure modes: the model repeats the same actions, deliberates without acting, or burns through context. qwenchance checks before each step and fires on one of three triggers—switch approach, stop reasoning and act, or hand off to a fresh session.
A loop fires if the model re-reads a file it has already seen (unchanged), runs the same command with identical arguments, returns to a dropped hypothesis, or gains no new information across the last two steps. Reasoning caps out around one thousand words; past that threshold, it becomes conversation with the user. Context pressure counts concrete signals: a system reminder about low memory, number of turns in the task, files and dumps already read, remaining plan steps.
When context fills up, the skill finishes the current step and hands off to a new session—artifacts (files, commits) stay in place, interruption is clean.
What it helps with
- Break out of circular re-evaluation and repeated file reads
- Stop unproductive reasoning and move to action
- Track context spend and time the handoff before running dry
- Coordinate a clean task handoff to a new session without losing work
Example prompts for the agent
- I notice I'm re-reading the same file. I'll run qwenchance to catch the loop.
- The model has been spinning on one bug for hours. I need qwenchance—it will flag when to pivot to a new hypothesis or ask for help.
- This task is complex and context is filling. I'll use qwenchance each step; it will warn before auto-compaction kicks in.
How to install qwenchance
1Skills CLI (Vercel)
npx skills add thananon/9arm-skills --skill qwenchance -a claude-code2Manually: put the files in place
~/.claude/skills/qwenchance/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/qwenchance/
1Skills CLI (Vercel)
npx skills add thananon/9arm-skills --skill qwenchance -a codex2Manually: put the files in place
.agents/skills/qwenchance/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add thananon/9arm-skills --skill qwenchance -a cursor2Manually: put the files in place
.cursor/skills/qwenchance/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install thananon/9arm-skills qwenchance2Manually: put the files in place
.github/skills/qwenchance/SKILL.md1Skills CLI (Vercel)
npx skills add thananon/9arm-skills --skill qwenchance -a gemini-cli2Platform CLI
gemini skills install https://github.com/thananon/9arm-skills --path skills/productivity/qwenchanceManually: put the files in place
.devin/skills/qwenchance/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add thananon/9arm-skills --skill qwenchance -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about qwenchance
How do I install qwenchance in Claude Code?
Run npx skills add thananon/9arm-skills --skill qwenchance -a claude-code. The skill will appear at ~/.claude/skills/qwenchance/SKILL.md.
What does qwenchance do?
Before each step, it checks for loops (repeated actions), stops over-reasoning, and watches the context budget. When memory gets tight, it signals time to hand off the task to a fresh session.
Which platforms does qwenchance support?
Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode. Each platform has its own install command.
