Intent Clarity
Skill2 starsChanged 4 months agoMITView on GitHub
Decode what the user actually needs before optimizing the wrong thing. Use when a request has vague referents ("fix it", "make it better", "clean this up", "improve this"), looks like a symptom-fix ("increase the timeout", "make this function faster"), is oddly specific with missing context, or would be strange taken literally. Use when a user corrects or rephrases - the delta between versions is the intent. ALWAYS use before asking any clarifying question, to check the question isn't lazy (inferable, decision-offloading, or covered by a reasonable default).
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- Adversarial VerifySkill
- Change ControlSkill
- Code ReconnaissanceSkill
- Correction PropagationSkill
- Debugging PlaybookSkill
- Deep DecompositionSkill
- Delegation DisciplineSkill
- Disclosure Is Not A FixSkill
- Discovery LoopSkill
- Divergent IdeationSkill
- Effort CalibrationSkill
- Empirical ValidationSkill
- Error RecoverySkill
- Extract ApproachSkill
- Failure Mode AwarenessSkill
- ForesightSkill
- Frontier Workflow ModeSkill
- Human HandoffSkill
- Leverage FirstSkill
- Live State TruthSkill
- Memory HygieneSkill
- Open MandateSkill
- Output StructuringSkill
- Plan GateSkill
- Proactive RigorSkill
- Product ThinkingSkill
- Prompt EngineeringSkill
- Publish HygieneSkill
- Research MethodologySkill
- Ruthless EditorSkill
- Scope FenceSkill
- Security Pattern ReviewSkill
- Self-Improvement LoopSkill
- Session OrientationSkill
- Structured ReasoningSkill
- Verification DisciplineSkill
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 32
Step 3 — Find the divergence, if any: Ask: "If I did exactly the literal thing, is there a plausible way the user would still be unhappy?" …
- Distinctive: line 43
Step 5 — Anchor the work to the mission: When making decisions mid-task, resolve them toward the mission, not the literal wording.
- Distinctive: line 54
Ask only when the answer would materially change what you produce AND the wrong branch is expensive.
- Distinctive: line 59
Every inferred constraint traces to evidence, not stereotype.
- Distinctive: line 60
If you proceeded on an interpretation, you said so in one line — the user never discovers a silent guess after the fact.
- Distinctive: line 67
Ignoring the correction signal: when the user rephrases a request, the delta between versions is the intent.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Intent Clarity
Purpose
Serve the mission, not just the sentence. Users compress their intent into short requests; the literal words are a lossy encoding of what they actually need. Misreading intent produces work that is technically responsive and practically useless. But the fix is not to interrogate the user — it's to reconstruct intent from available evidence and proceed, asking only when a wrong guess would be expensive.
When to use this skill
- At the start of any task big enough to plan (
plan-gateDepth 1+). - The request contains vague referents ("this", "it", "better", "cleaner") or an unusual constraint you don't understand.
- The literal request seems like a symptom-fix ("increase the timeout") where the mission is probably deeper ("make this stop failing").
- You're about to ask a clarifying question — run the lazy-question check below first.
When NOT to use this skill
- The request is fully specified and unambiguous. Don't manufacture hidden depths in "fix the typo in line 12".
- Mid-task re-litigation: once you've stated an interpretation and the user hasn't objected, don't keep reopening it.
Operating procedure
Step 1 — Separate the two layers:
- Literal request: what the words say to do.
- Mission: why they want it — what outcome makes them say "yes, that's it."
Step 2 — Reconstruct constraints from evidence, in this order: explicit statements → the material they provided (its style, format, audience) → the context of the conversation → common sense for the task type. Typical inferable constraints: audience, tone, length, deadline pressure (quick draft vs. polished), compatibility with existing work, appetite for changes beyond the ask.
Step 3 — Find the divergence, if any. Ask: "If I did exactly the literal thing, is there a plausible way the user would still be unhappy?" If no — proceed. If yes — that gap is the ambiguity that matters.
Step 4 — Decide: proceed or ask.
| Situation | Action |
| Ambiguity exists but any reasonable reading leads to similar work | Proceed; note the reading in one line |
| One reading is clearly most probable | Proceed on it; state it: "Interpreting 'clean up' as X — flag me if you meant Y" |
| Readings diverge sharply AND wrong guess wastes major work or is hard to undo | Ask — one question, the one that forks the work |
| The user clearly wants momentum (quick, informal, "just...") | Proceed. A best-effort draft beats a questionnaire |
Step 5 — Anchor the work to the mission. When making decisions mid-task, resolve them toward the mission, not the literal wording.
Do not ask lazy clarifying questions
A clarifying question is lazy when:
- The answer is inferable from the material already provided.
- It offloads a decision you're better positioned to make ("do you want good code or fast code?").
- It's a bundle of 3+ questions where only one changes the work.
- It exists to cover yourself, not to change the deliverable.
- A reasonable default exists — state the default and proceed instead.
Ask only when the answer would materially change what you produce AND the wrong branch is expensive. When you do ask: one question, concrete options, and do any work that's common to all branches while waiting if the setting allows.
Quality bar
- You can state the mission in one sentence that is not a paraphrase of the request.
- Every inferred constraint traces to evidence, not stereotype.
- If you proceeded on an interpretation, you said so in one line — the user never discovers a silent guess after the fact.
Common failure modes
- Literalism: shortening an email that was actually failing because it buried the ask. Responsive; useless.
- Mind-reading overreach: inventing a mission the evidence doesn't support and delivering something other than what was asked. The mission reframes how you do the task, not whether you do it — scope changes go through
scope-fence. - Question-as-stall: asking to avoid committing. If you'd struggle to say how each answer changes your output, don't ask.
- Ignoring the correction signal: when the user rephrases a request, the delta between versions is the intent. Read it.
- Frozen-mission blindness: in long collaborations the mission itself evolves ("skills for me" becomes "public reputation" becomes "value for everyone") — serving the original mission after it has moved is literalism at session scale. Re-anchor on the current mission when the user's asks start pointing somewhere new.
- Objective-function substitution: executing the user's tasks while optimizing YOUR preferred objective — most often your risk tolerance in place of theirs. Real case (2026-07-11): operator said "race to the target; the stake is accepted tuition" and the agent kept optimizing "protect the stake," rejecting the one strategy that served the stated goal — correct analysis, wrong utility function, caught only under pressure. On any risk/return recommendation, restate WHOSE objective and risk tolerance it serves; if it's yours, say so and re-derive under theirs.
Example
Request: "Can you make this function faster?" — attached: a function called once at startup, taking 40ms.
- Literal: optimize the function. Mission check: 40ms at startup is invisible; why does the user care? Plausible mission: something feels slow and they've guessed this culprit.
- Action: verify the guess before optimizing (
live-state-truth): "This runs once at startup (~40ms), so it's unlikely to be what feels slow. Checking where time actually goes first." Then profile.
Works with sibling skills
- Runs first in
frontier-workflow-mode— everything downstream inherits the interpretation. - Feeds the mission and constraints into
plan-gateandeffort-calibration. scope-fenceenforces the boundary once intent is set; intent-clarity must not be used to smuggle in scope expansion.ruthless-editoruses the audience/tone constraints inferred here.
Provenance and maintenance
Written 2026-07 as part of a portable, project-agnostic quality pack. No repo-specific claims. Re-verify occasionally by sampling recent tasks: count how often clarifying questions were asked and whether each answer actually changed the output — if most didn't, the lazy-question bar needs raising; if wrong-guess rework is common, lower it.
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.

