Unstuck

Install in one command

mkdir -p ~/.claude/skills/unstuck && curl -fsSL https://eliteaiadvantage.com/skills/unstuck/SKILL.md -o ~/.claude/skills/unstuck/SKILL.md

When you've been chasing a bug for 20+ minutes and you're no closer, the failure mode is almost never "I haven't read enough code." It's that you started spelunking before you had a hypothesis tree. Unstuck forces the structure: state the symptom, list ranked hypotheses with evidence, name the cheapest test for each, run them in order.

The skill is read-only — it produces a plan, not a fix. The fix happens after, when you know which hypothesis was right. Most bugs collapse to one of three causes once you actually rank them.

• →Breaks the "keep reading code, keep guessing" loop with forced ranking.
• →Each hypothesis includes the cheapest test (DB query, log line, single command) — not "investigate further."
• →Run-order picks the highest likelihood × cheapest test first, not the one you happened to think of.
• →Forces honest hypothesis count — if you can't generate three, you don't understand the system enough to debug it raw.
• →Includes a "zoom out" trigger: when two tests fail, the framing is wrong, look elsewhere.

• •Twenty minutes into a bug hunt and you're re-reading the same files.
• •Test failures that don't match what you'd expect.
• •Production behavior that "should work" but doesn't.
• •Anything where you've thought "let me try one more thing" three times in a row.

Without this skill, debug sessions tend to spiral — Claude keeps reading more code, suggesting more changes, never stepping back to commit to a hypothesis. Unstuck stops the spiral: produces a ranked tree, picks a starting test, names what would suggest the framing is wrong. Debug becomes a search through hypotheses, not a search through files.