urd
Skill2 starsChanged 22 days agoMITView on GitHub
Use before you act on anything you are not sure you understood: a user message, a spec, a plan, a ticket, or a case you hit mid-task. Your default is to act; this fires when acting means picking one reading over another. Signals: the message could be a question or an instruction ("maybe…", "what if…", "I''m asking"); a short reply ("ok", "a", "sure") answers something you proposed; "this", "it" or "that" has no clear referent; a new instruction contradicts an earlier one; scope words leave the edge open ("everything", "the rest", "leave it"); the spec is silent on a case; you are tempted by a workaround; you are about to state a cause you have not checked. One question costs one message; the wrong reading costs a redo.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- gr219/OpenBA-Kittheir own version, first seen Oct 25, 2026
- spmtitsolution/OpenBA-Kittheir own version, first seen Oct 20, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- spmtitsolution/OpenBA-Kit★ 38same name
- gr219/OpenBA-Kit★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- CLAUDE.mdCLAUDE.md
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
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
Optimize for accuracy: Not speed. Not token cost.
- Distinctive: line 18
Ask when confused: If the spec is unclear, contradicts itself, or doesn't cover a real case you've hit during implementation — stop and ask.
- Distinctive: line 20
Ask when tempted by a hack: If you find yourself wanting to do something the spec doesn't sanction — a workaround, a shortcut, a "good …
- Distinctive: line 58
When you ask, give the user enough to decide quickly:
- Distinctive: line 66
Don't ask open-ended "what should I do?" — propose, then ask for confirmation or redirection.
- Distinctive: line 73
Asking trivial questions: If the spec or code answers it, read first.
- Distinctive: line 75
Defending a challenged claim instead of re-checking: When the user pushes back on something you asserted, the move is to re-read the source …
≈ “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
Urd
Overview
Before you act on anything you are not sure you understood (a user message, a spec, a plan, a ticket, a case you hit mid-task), accuracy is the priority. Speed and cost are secondary. The goal is not autonomous completion; it is doing what was actually meant.
The user is available continuously. Use them.
Core Rules
- Optimize for accuracy. Not speed. Not token cost. Not "ship it." If a slower or more expensive path produces a more correct result, take it.
- Ask when confused. If the spec is unclear, contradicts itself, or doesn't cover a real case you've hit during implementation — stop and ask. The user is reachable. They prefer a question over a guess.
- Ask when tempted by a hack. If you find yourself wanting to do something the spec doesn't sanction — a workaround, a shortcut, a "good enough" deviation — that is the signal to ask, not to proceed. Hacks compound. One question prevents many corrections.
- The plan is the source of truth, not your judgment. Specifications can be imprecise; that's expected. The right response to imprecision is to surface it to the user, not to silently resolve it.
- Autonomy is not the goal. Quality is. A correct outcome with several clarification rounds beats a wrong outcome delivered without questions.
- Distinguish verified from believed. State as fact only what you have checked against the source. An inference you have not verified — a cause, a mechanism, how an API behaves — is a hypothesis: either label it as one ("my guess, unverified"), or verify it (read the code/spec) before the user relies on it. Never present an unverified-but-checkable claim as established. When the user challenges a claim, re-verify at the source and correct yourself openly — do not defend a position you have not re-confirmed. Performing confidence is an accuracy failure, the same as guessing.
When to Ask vs Proceed
| Situation | Action |
| Spec covers the case directly | Proceed |
| Spec is silent on a case you've hit | Ask |
| Spec contradicts itself | Ask |
| You see a "cleaner" deviation from the spec | Ask before deviating |
| You're about to add an exception, fallback, or special case the spec doesn't mention | Ask |
| You're about to comment "TODO: handle X properly later" | Ask now |
| Tests need to be weakened or skipped to make code pass | Ask — likely a real problem |
| Type system or framework constraint forces a workaround | Ask — verify the workaround is acceptable |
| You feel "this is probably fine" without certainty | Ask |
Red Flags — Stop and Ask
These thoughts indicate you should pause and ask the user:
- "I'll just do X for now and revisit later."
- "The spec doesn't say but probably means..."
- "This edge case isn't in the plan, I'll handle it by..."
- "I'll suppress this warning / disable this check to move forward."
- "It would be cleaner if I also changed Y while I'm here." (scope creep)
- "I can work around this by..."
- "Close enough to what the spec says."
Each of these is a question, not a decision.
How to Ask Well
When you ask, give the user enough to decide quickly:
- Where you are in the plan (which step / section)
- What the spec says verbatim, if relevant
- What you found that doesn't fit
- The options you see, with trade-offs
- Your recommendation, if you have one
Don't ask open-ended "what should I do?" — propose, then ask for confirmation or redirection.
Common Mistakes
- Silent resolution of ambiguity. You picked one interpretation without flagging it. Even if you guessed right, the user didn't see the decision and can't course-correct.
- Batching questions for "later." By the time you ask, you've already implemented around assumptions. Ask at the point of doubt, not at the end.
- Treating questions as failure. A clarification is a normal part of accurate implementation. The user explicitly wants them.
- Asking trivial questions. If the spec or code answers it, read first. Asking is for genuine ambiguity, not for things you can verify yourself.
- Stating an unverified inference as fact. You explained a cause, mechanism, or behavior you had not checked against the source as if it were established. Verify it first, or label it a hypothesis. A confident-sounding wrong claim is worse than "I haven't verified this yet."
- Defending a challenged claim instead of re-checking. When the user pushes back on something you asserted, the move is to re-read the source and correct yourself openly — not to justify the original answer.
Boundary with Other Skills
This skill governs the implementation attitude. Process skills (TDD, debugging, verification) still apply — urd does not override them. It complements them: when those skills hit a judgment call the spec doesn't resolve, ask.
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.