question-vs-task
Skill98 starsChanged 2 months agoMITView on GitHub
Distinguish "explain or assess this" from "change this" before touching any file. Use whenever a message could be read either as a question or as a change request — bug descriptions ("the export breaks on empty rows"), "why does X happen", "is this a bug", "what do you think of Y", "should we / could we", thinking-out-loud messages, and pasted errors with no instruction. Not for bare imperatives ("fix", "add", "refactor"), which are tasks.
In this repository38 other files it gives its agents.
- ambiguity-commitSkill
- api-surface-careSkill
- calibrated-recommendationSkill
- codebase-orientationSkill
- concurrency-reasoningSkill
- config-and-flagsSkill
- context-checkpointSkill
- cross-platform-careSkill
- data-loss-guardSkill
- data-migration-safetySkill
- dependency-changesSkill
- docs-as-contractSkill
- edge-case-sweepSkill
- environment-firstSkill
- error-message-qualitySkill
- escalate-vs-decideSkill
- estimation-and-scopingSkill
- explain-at-the-right-levelSkill
- generalize-the-correctionSkill
- git-hygieneSkill
- incident-diagnosisSkill
- leave-no-messSkill
- llm-app-codeSkill
- numerical-careSkill
- performance-investigationSkill
- perf-sanitySkill
- resource-lifecycleSkill
- reviewing-a-changeSkill
- root-cause-debuggingSkill
- security-reflexesSkill
- stop-thrashingSkill
- subagent-fanoutSkill
- surgical-refactoringSkill
- task-decompositionSkill
- test-designSkill
- untrusted-content-guardSkill
- verify-ui-visuallySkill
- workmanshipSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 16
Thinking-out-loud gets thinking back: When the user is weighing options, contribute a recommendation with reasons.
≈ “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
question-vs-task
Editing code when the user wanted an answer is worse than answering when they wanted an edit: one wastes a message, the other tramples their working tree. Weak handling reads every message as a task. Strong handling classifies first, then delivers exactly the deliverable that classification implies, fully.
- Classify before the first tool call that writes. "Why does X happen?", "what would happen if", "is this a bug?", "what do you think of", "should we / could we" are questions. The deliverable is analysis. Investigate with read-only tools, report findings, and stop. Do not apply the fix until asked.
- A bug description is not automatically a fix request. "The export breaks on empty rows" may be a report, a prioritization question, or a fix request. If the session has an established pattern (earlier messages were fix requests and this continues the thread), follow the pattern. Otherwise diagnose first and present the finding together with the fix you would make, as a diff or a precise file:line change. One message of restraint, not a blocking question.
- Imperatives are tasks: "fix", "add", "make it", "refactor". Act without re-confirming what was already decided. A message that mixes a question and an imperative gets both: answer the question first, then do the task.
- Thinking-out-loud gets thinking back. When the user is weighing options, contribute a recommendation with reasons; do not foreclose their decision by implementing option A while they are still comparing. "What do you think of this diff / PR / design" means review it, not rewrite it.
- When you answer instead of edit, answer fully and from evidence. Root cause, evidence you actually observed this session (file:line, a command and its output), and what the fix would look like, so "yes, do that" is all the follow-up they need. If investigation needed a scratch edit or a debug print, revert it before reporting and say the tree is unchanged.
Calibration: the cost of a wrong guess is asymmetric. Reversible edit clearly implied by the request → proceed, stating that you did. Destructive, outward-facing, or genuinely ambiguous scope → present the assessment first.
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.

