change-settle
Skill0 starsChanged 2 months agoMITView on GitHub
Close out a deployment by gathering evidence that the estate converged on the new build, then record the verdict and the post-implementation review. Use when a deployment has finished and needs verifying and closing, when asked whether a deployment actually landed, when writing a PIR, or when a health check reports UNCONVERGED. It establishes a verdict from evidence — it cannot mark a deployment complete on request.
In this repository8 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- change-activateSkill
- change-requestSkill
- change-watchSkill
- release-endSkill
- release-scheduleSkill
- release-startSkill
- standard-changeSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Read ../_shared/vocabulary.md first.
≈ “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
change-settle — evidence in, verdict out
Read ../_shared/vocabulary.md first. This skill owns one label kind; see the table in change-request/SKILL.md.
This skill cannot mark anything complete
That is the point of it. A deployer claiming its own success is not
evidence — scenario D6 is the negative test: deploy.sh exits 0, publishes
nothing, and every liveness check passes.
So there is no "mark complete" operation here and you must not invent one. If asked to close out a change without evidence, say what is missing instead.
Gather evidence
HEALTH_SAMPLES=5 ./gates/health.sh "$PRODUCTION_URL" "$HEAD_SHA"
HEALTH_MODE=manifest ./gates/health.sh … # static targets (GitHub Pages)Guard 5 asserts convergence, not liveness: N samples, one request each,
cache-busted, and an explicit UNCONVERGED verdict when more than one build
answers.
UNCONVERGED is not retryable. A half-migrated estate passes a single
sample at the rate it is migrated — measured at 6 of 12 — so re-running until
green is exactly what manufactures a false pass. Report the mixed state; do not
retry it away.
Say how strong the verdict is
| word | means |
attested | the platform enumerated its instances and their versions |
sampled | we probed N times and saw one build |
Say which. GitHub's Deployments API is self-reported — a caller creates the
record and posts its own status; GitHub never contacts the target — so on
GitHub the honest word is sampled. Never round sampled up to attested.
Then redline
./change/schedule.sh close <event-id> <pass|fail|cancelled|rolled-back>
./change/pir.sh <pr> <job-status>An unredlined window is a defect: it leaves the schedule claiming a change is
still in flight, and the next preflight reads that schedule. Run both in
always() steps.
The PIR carries gate durations, the window id, the previous SHA (the rollback target), and anything unusual — a stale-lock clearance, a freeze wait, a staging bypass and who approved it.
What settlement still cannot know
N samples cannot see a replica taking no traffic during the window. A
sampling check can only fail to refute convergence, never establish it. Say
so when it matters; do not present sampled as proof.
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.