hard-problems
Skill2 starsChanged 24 days agoMITView on GitHub
Attacks a problem nobody has solved yet. It writes many guesses down, runs one cheap test per guess, logs every try (failures included), keeps only the guesses that survived and proves the answer with a result someone else can rerun. Fixes "The agent tried one idea after another, and we are still stuck." Use when a bug or question has beaten the first two fixes, when nobody knows the cause, when the agent keeps saying "this should fix it", or when a session is going in circles. Also on /hard-problems.
In this repository102 other files it gives its agents.
- access-levelsSkill
- agent-failuresSkill
- agent-guardSkill
- agent-hooksSkill
- agent-limitsSkill
- agent-loopSkill
- agent-patternsSkill
- agent-wikiSkill
- ai-in-businessSkill
- answer-checksSkill
- ask-firstSkill
- automate-questionsSkill
- batch-judgeSkill
- big-archiveSkill
- brief-partsSkill
- bug-triageSkill
- build-an-evalSkill
- calls-to-keepSkill
- checkpointsSkill
- check-waysSkill
- claude-md-rulesSkill
- code-mapSkill
- command-guardSkill
- compact-safelySkill
- context-rulesSkill
- corrected-twiceSkill
- count-checkSkill
- data-leaksSkill
- day-gateSkill
- disprove-findingsSkill
- dogfood-firstSkill
- done-testSkill
- draft-only-mailSkill
- drift-checkSkill
- fail-closedSkill
- failed-readsSkill
- faster-sameSkill
- first-messageSkill
- fix-reviewSkill
- handoff-noteSkill
- handoff-rulesSkill
- harness-starterSkill
- hidden-text-checkSkill
- hooks-to-addSkill
- injection-guardsSkill
- inventory-firstSkill
- keys-and-agentsSkill
- live-checkSkill
- lock-on-failureSkill
- long-videosSkill
- mcp-partsSkill
- memory-queueSkill
- memory-sourcesSkill
- memory-trimSkill
- night-alarmSkill
- objectorSkill
- old-code-judgeSkill
- one-task-gateSkill
- past-green-testsSkill
- phone-rulesSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Keep what survived the tests, not what you hoped for.
- Distinctive: line 18
Write the problem in one sentence: what you see and what you expected to see.
- Distinctive: line 19
Write down every guess about the cause as H1, H2, H3 and on.
- Distinctive: line 27
Never write "found it" or "fixed" before the rerun matched.
- Distinctive: line 29
If ten tries in a row kill no guess, stop and ask the person before the next one.
- Distinctive: line 30
Keep the log after the problem is solved.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Hard problems
The agent tried one idea, then another, and you are still where you started. It chased the first guess, forgot what had failed and called the last try the answer. This skill makes it work like a lab: guesses written down, one cheap test each, every try logged, and a result someone else can rerun.
The principles
- A guess in the head gets chased. A guess on paper gets tested.
- One cheap test per guess. A test that cannot fail tells you nothing.
- Failed tries are data. A try that is not logged gets tried again.
- Keep what survived the tests, not what you hoped for.
- It is solved only when someone else can rerun it and get the same result.
Steps
- Write the problem in one sentence: what you see and what you expected to see.
- Write down every guess about the cause as H1, H2, H3 and on. Write at least five before you test any.
- For each guess, write one cheap test and the result that would kill it. Cheap means minutes: a log line, a small input, one setting changed, one older git commit checked out. Run the cheapest tests first.
- Log every try in a file such as tries.md, one line each: number, guess, what you did, what happened, killed or open. Failed tries go in too. Read the log before each new try so nothing runs twice.
- Narrow down. Cross out the killed guesses. If several are left, design one test that tells them apart. If none is left, write new guesses from what the log shows and go back to step 3.
- Prove the survivor. Write the exact steps and the command that show the result. Then someone else reruns it: a fresh session, a subagent that gets only the written steps, or a person. It counts only if they get the same output.
- Report the guess that survived, the test that proves it, how many tries it took, and each killed guess with the try that killed it.
Rules
- Never write "found it" or "fixed" before the rerun matched.
- Change one thing per try, so each result points to one guess.
- If ten tries in a row kill no guess, stop and ask the person before the next one.
- Keep the log after the problem is solved. The next hard problem may start there.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

