proving-change-safety
Skill1 starsChanged 11 days agoMITView on GitHub
Find what a change could break beyond the diff, then prove the one fact it is safe because of by running real code. Apply when the user asks what a change could break, whether a small diff is safe to merge, or to prove a safety fact before shipping. Use subsystem-walkthrough for how it works. Use design-rationale for motivation. Use dissect for a system-wide entity audit.
In this repository107 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/base.mdcCursor rule
- .cursor/rules/changelog.mdcCursor rule
- .cursor/rules/comments.mdcCursor rule
- .cursor/rules/release.mdcCursor rule
- .cursor/rules/seo-ai-search.mdcCursor rule
- .cursor/rules/testing.mdcCursor rule
- agent-readySkill
- authoring-skills-and-rulesSkill
- blinded-evalSkill
- cloakbrowser-fallbackSkill
- deep-deliberationSkill
- design-rationaleSkill
- dissectSkill
- documentation-writerSkill
- everything-seoSkill
- executeSkill
- generating-app-verifySkill
- inceptionSkill
- kit-workflowsSkill
- marketing-psychologySkill
- media-genSkill
- nlm-skillSkill
- opinionSkill
- proving-change-safetySkill
- refreshing-app-verifySkill
- release-deploySkill
- seo-auditSkill
- startup-marketing-brainSkill
- subsystem-walkthroughSkill
- trlSkill
- unmatched-workflowSkill
- agent-readySkill
- authoring-skills-and-rulesSkill
- blinded-evalSkill
- cloakbrowser-fallbackSkill
- deep-deliberationSkill
- design-rationaleSkill
- dissectSkill
- documentation-writerSkill
- everything-seoSkill
- executeSkill
- generating-app-verifySkill
- inceptionSkill
- kit-workflowsSkill
- marketing-psychologySkill
- media-genSkill
- nlm-skillSkill
- opinionSkill
- refreshing-app-verifySkill
- release-deploySkill
- seo-auditSkill
- startup-marketing-brainSkill
- subsystem-walkthroughSkill
- trlSkill
- unmatched-workflowSkill
- agent-readySkill
- authoring-skills-and-rulesSkill
- blinded-evalSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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 34
Read the change: symbols added, changed, deleted, including what the diff does not spell out.
- Distinctive: line 36
Look where grep stops: library source, pinned version, microtasks, wire format, other languages, flags, code three hops down.
- Distinctive: line 56
do not ship: if the fact is false; unproven only if you did not run it
≈ “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
Proving change safety
Leading words: one safety fact, prove by running, unproven stays unproven, grep is not enough.
Listing callers is not the job. Grep those in a second. The job is the breakage grep will not show.
Copy these steps into the todo list verbatim. Skip with skip: <reason>.
- Read the change: symbols added, changed, deleted, including what the diff does not spell out.
- Name the one fact it is safe because of.
- Look where grep stops: library source, pinned version, microtasks, wire format, other languages, flags, code three hops down.
- Honest risk list: chance, cost,
file:line. Cleared items listed separately. - Prove the one fact by calling the real function (script or test). Leave the prove script in the tree. Cite its path in the reply. If you cannot get to "ran it", write unproven. If you ran it and the fact is false, write do not ship — that is not unproven.
How sure
For each safety fact, get as far down this list as is cheap, and say where it stopped.
- You said so. Worthless alone.
- You pointed at
file:lineor the library's own source. - You walked the failure and it does not reach.
- You ran it. A script or test that imports the same library the app ships and calls the exact function.
- You reproduced it in the running app.
Step 4 is the bar. Unproven stays unproven.
Reply contract
- What it does (including the non-obvious part)
- The one safety fact, the step you reached, and the proof path (or unproven)
- do not ship if the fact is false; unproven only if you did not run it
- Risks that remain
- Cleared
- Cheapest check to catch the real bug before merge
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.

