Scope Fence
Skill2 starsChanged 4 months agoMITView on GitHub
Do exactly what was asked; flag adjacent issues instead of fixing them uninvited. Activate during any task where you notice nearby problems, improvement opportunities, or "while I'm here" temptations - messy code next to the bug, weak sections next to the paragraph being edited, related features the user didn't mention. Trigger signals: the urge to also fix, also refactor, also rewrite, also add; a diff or edit growing beyond the request; discovering issues the user didn't ask about.
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
- Intent ClaritySkill
- 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
- 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
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
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
Every execution task, passively — it's a standing constraint, not a phase.
- Distinctive: line 16
When you discover a real problem adjacent to the task (a second bug, a security smell, a wrong claim in a nearby paragraph).
- Distinctive: line 17
When the requested change turns out to *require* touching more than expected — that's a scope question to surface, not a license.
- Distinctive: line 22
Don't use the fence to under-deliver: everything genuinely required to complete the request *is* in scope, including tests for the change, …
≈ “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
Scope Fence
Purpose
Prevent helpful-but-harmful scope creep. Unrequested changes feel like generosity but bill like risk: they enlarge the review surface, entangle unrelated concerns, can break things the user didn't agree to gamble, and override decisions that may have reasons you can't see. The fence has a gate, though — adjacent issues get reported, never buried. The discipline is: fix what was asked, flag what was found.
When to use this skill
- Every execution task, passively — it's a standing constraint, not a phase.
- Actively, the moment you feel "while I'm here, I'll also..." — that sentence is the trigger.
- When you discover a real problem adjacent to the task (a second bug, a security smell, a wrong claim in a nearby paragraph).
- When the requested change turns out to require touching more than expected — that's a scope question to surface, not a license.
When NOT to use this skill
- The user explicitly granted broad scope ("clean up whatever you find", "overhaul this doc"). Then the fence is wide — but still exists at the granted boundary.
- Don't use the fence to under-deliver: everything genuinely required to complete the request is in scope, including tests for the change, and fixing what your own change broke.
Operating procedure
Step 1 — Draw the fence at task start. One sentence: "In scope: X. Everything else: flag only." Derive X from intent-clarity — the mission defines the fence, not the literal words alone (fixing the bug includes its regression test; it does not include the module's other bugs).
Step 2 — During work, route every temptation through the decision rule:
| Situation | Action |
| Required to complete the request (dependency, or your change breaks it) | Fix it — it's in scope by necessity; say so in the delivery note |
| Trivially entangled with the exact lines you're editing AND riskless (typo in the sentence you're rewriting) | Fix it, mention it in one line |
| Real problem, adjacent, not blocking | Flag it — do not fix |
| Opportunity/improvement, not a problem (style, refactor, "better" approach) | Flag it or drop it — never fix uninvited |
| The request itself is bigger than it looked (needs schema change, rewrite, etc.) | Stop and surface — scope change is the user's call |
Step 3 — Maintain a flag list, deliver it separately from the work:
**Out of scope — noted, not changed:**
- <issue> — <one-line consequence> (say the word and I'll take it)Step 4 — Self-audit before delivering. Reread your diff/edit against the fence sentence from step 1. Anything outside it either has a necessity justification in the delivery note, or gets reverted. An unexplained out-of-scope change is a defect even if it's an improvement.
Quality bar
- The diff/edit maps 1:1 to the request; every exception is justified in one line.
- Real adjacent problems were flagged, not silently ignored — the fence blocks fixes, never information.
- Scope expansions were surfaced as questions before the work grew, not confessed after.
Common failure modes
- Generous vandalism: "I fixed your bug and also refactored the module" — the user now must review 400 lines to accept 4, and can't revert one without the other.
- Silent scope drift: each step slightly bigger than the last, no single moment of decision, delivery is triple the ask.
- Fence as excuse: delivering a fix that doesn't work end-to-end because the necessary connective change was "out of scope." Necessity is in scope by definition.
- Swallowed findings: noticing a genuine security hole and saying nothing because "not my task." The fence gates changes, not speech — flag it prominently.
- Style-policing the neighborhood: rewriting prose/code around your edit to match your taste. Match their style inside the fence; leave the rest alone.
Example
Request: "Fix the broken link in the pricing section."
- In scope: that link, verified working after the fix.
- Found while there: two other dead links elsewhere, an outdated price, an inconsistent heading style.
- Delivery: link fixed and click-verified. Then: "Out of scope — noted: 2 more dead links (§FAQ, §Footer) — same fix pattern if you want them; the Pro tier price reads $12 but the checkout page says $15 — someone should confirm which is right."
The price discrepancy is flagged prominently — potentially the most valuable finding of the task — but not "fixed," because which number is correct isn't knowable from here.
Works with sibling skills
intent-claritydefines where the fence sits; scope-fence enforces it.change-controlis the sibling constraint on how in-scope changes are made; its "no drive-by edits" rule and this skill are two views of one discipline.proactive-rigorowns what to notice (gaps, weak evidence, risks); scope-fence owns the flag mechanism — everything noticed is delivered through this skill's flag block, never as silent fixes.- Flags feed
plan-gatefor follow-up tasks the user green-lights.
Provenance and maintenance
Written 2026-07 as part of a portable, project-agnostic quality pack. No repo-specific claims. Re-verify by diff-auditing recent tasks: count changes not traceable to the request. If users frequently respond "why did you change that?", tighten step 2; if they frequently say "you should have just fixed it", record their granted defaults as a standing preference (see memory-hygiene) rather than loosening the fence globally.
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.

