ab-a-isolation
Skill8 starsChanged 11 days agoApache-2.0View on GitHub
The A/B/A stash experiment — decide in minutes whether a test failure is YOUR diff or the environment/run-history. Use when a suite fails during your change and blame is ambiguous.
In this repository39 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .cursor/rules/architecture-patterns.mdcCursor rule
- .cursor/rules/completion-discipline.mdCursor rule
- .cursor/rules/github-tracking.mdcCursor rule
- .cursor/rules/programmer-briefing.mdcCursor rule
- .cursor/rules/verification-first.mdcCursor rule
- assign-sprint-safelySkill
- audit-cascadeSkill
- brief-coding-agentSkill
- check-mailboxSkill
- cleanup-dev-activeSkill
- close-issue-properlySkill
- cohort-attention-rollupSkill
- compost-reviewSkill
- continue-narrativeSkill
- create-omnibusSkill
- create-session-logSkill
- cut-releaseSkill
- delete-module-safelySkill
- deliver-mailSkill
- discovered-work-captureSkill
- doc-sync-sweepSkill
- draft-blog-postSkill
- draft-weekly-shipSkill
- duty-cycle-tickSkill
- narrative-verificationSkill
- piper-draft-issueSkill
- piper-draft-specSkill
- piper-sprint-planSkill
- piper-stakeholder-updateSkill
- piper-synthesize-feedbackSkill
- propose-featureSkill
- publish-to-blogSkill
- query-github-boardSkill
- template-auditSkill
- trust-checkSkill
- update-calendarSkill
- update-current-stateSkill
- update-piperSkill
The file
Files
ab-a-isolation
Decide whether a failure is your diff or the world — by running the same check three times with only one variable moved. Canonized from two decisive uses in one week (2026-08-13: the e2e "regression" that was #1532-era run-history — #1602; and the pre-existing provenance failure isolated the same hour).
When to Use
- A test fails while your uncommitted change is in the tree, and "did I break this?" is not obvious.
- A previously-green suite reads red and you're about to either debug your diff (maybe wasted) or push anyway (maybe reckless).
The experiment
# The stash stack is SHARED by every worktree on the host: never bare `git stash pop` (it can pop another
# session's entry). Tag your entry uniquely, apply it by SHA, then drop it by re-finding the tag.
TAG="aba-$(git branch --show-current)-$$"
# A — the unchanged world
git stash push -u -m "$TAG" # -u if untracked files are part of the diff
SHA=$(git stash list --format='%H %gs' | awk -v t="$TAG" '$NF==t {print $1; exit}')
<run the failing check> # record the result VERBATIM
# B — your change
git stash apply "$SHA"
<run the same check> # record verbatim
# A again — the tiebreaker (THIS is the step people skip, and it decides)
git stash push -u -m "$TAG-2"
SHA2=$(git stash list --format='%H %gs' | awk -v t="$TAG-2" '$NF==t {print $1; exit}')
<run the same check>
git stash apply "$SHA2" # restore your change
# drop YOUR two entries only, each re-found by its tag (indexes shift, so look them up one at a time)
for tg in "$TAG-2" "$TAG"; do
git stash drop "$(git stash list --format='%gd %gs' | awk -v t="$tg" '$NF==t {print $1; exit}')"
done(Shared-stash fix, 2026-10-08, R6 step 4 defect P5.)
Reading the three results
| A | B | A′ | verdict |
| pass | fail | pass | your diff — debug it |
| fail | fail | fail | pre-existing — file it, don't absorb it into your change |
| pass | fail | fail | run-history/state — the run itself mutates shared state (DB rows, caches, session stores); neither you nor HEAD is clean twice. Find the state carrier (#1602 was conversation rows + fixed session ids) |
| flaky mix | margin/stochastic — name it as such; never tune-until-green |
The third run is the whole method. A/B alone cannot distinguish "my diff broke it" from "the first run planted state that breaks every later run" — the 2026-08-13 case read as a 13-test regression on A/B and was proven run-history by A′ (unchanged tree failing 12 after previously passing).
Rules
- Same command, verbatim, all three runs (shape drift invalidates the read).
git statusBEFORE the first stash; verify clean pop after each (stash -u captures untracked — know what you're carrying).- Record all three results in your log/issue — a verdict without the three numbers is an assertion, not an experiment.
- On a shared/dev DB, note that A′'s failure may be A's own residue — cleanup belongs to the run that planted 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.

