change-watch
Skill0 starsChanged 2 months agoMITView on GitHub
Report the estate's health — which of the five invariants are failing, what the dashboard says, who holds the berth, and whether the calendar has stale windows. Use when asked how the estate is, whether anything is wrong, what is blocking deployments, why a change is stuck, or to check before or after a deployment round. Do NOT use to fix what it finds: this skill observes and never writes a label, clears a lock, or reaps a window.
In this repository8 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- change-activateSkill
- change-requestSkill
- change-settleSkill
- release-endSkill
- release-scheduleSkill
- release-startSkill
- standard-changeSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
This skill writes nothing: Not a label, not a lock, not a window.
- Very specific: line 16
If asked to fix something this skill surfaces, say what you found, name the verb that owns the fix, and let the caller decide.
- Distinctive: line 51
Clean and unread are different answers: If the audit could not run, the forge could not be reached, or the dashboard is dark, say …
≈ “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
change-watch — the verb that only looks
Read ../_shared/vocabulary.md first.
This skill writes nothing. Not a label, not a lock, not a window. That is
the whole discipline: the estate has three writers already (a human, the
workflows, and whichever driver is running), and scenarios.org D21 is what
happened when a fourth cleared a lock it did not hold while its own log said
"releasing the estate". A watcher that repairs is a fourth writer.
If asked to fix something this skill surfaces, say what you found, name the
verb that owns the fix, and let the caller decide. The owning verbs are in
change/label-owners.tsv.
The five invariants
gates/pr-state-audit.py holds them and their self-test. Do not reimplement
them here — two copies of one rule is the defect this repo keeps finding.
| invariant | why it exists | |
| I1 | at most one open PR carries deploy:staging | the berth. Guard 1 has admitted two holders (D22) |
| I2 | at most one open PR carries deploy:production | the path to production is a singleton |
| I3 | no open PR carries a terminal label | release:ended/complete/failed/backed-out claim a state an open change is not in |
| I4 | no PR is finished and moving | #52 carried release:ended through an entire successful deploy |
| I5 | the six declared exclusive groups hold | declared in label-owners.tsv; nothing read them until this gate |
How to look
gmake -s pr-audit # the five, against the live forge
python3 gates/pr-state-audit.py --json # same, for a machine
curl -s http://192.168.86.29:9999/api/status # the dashboard's own view
./change/lock.sh status # who holds the berth, and since when
./change/schedule.sh list --open # windows still open
./targets/node/switch.sh status # ask the FRONT which colour is livegmake pr-audit exits 1 when it finds something. That is its verdict, not
a failure to produce one — do not report a non-zero exit as "the audit broke".
The dashboard made exactly that mistake and rendered findings as "the audit did
not run".
Reporting rules
Clean and unread are different answers. If the audit could not run, the
forge could not be reached, or the dashboard is dark, say indeterminate and
say which. Never report an estate you could not observe as healthy — that is
spec.org defect class 1, and it is the class this pipeline has produced most
often.
Name the change, not just the count. "2 findings" is useless to someone who
has to act. "#55 and #70 carry release:ended while open" is actionable.
Distinguish a violation from a symptom. A stale release:scheduled after a
window was reaped is bookkeeping. Two PRs holding deploy:staging means a
deployment is happening that nobody is coordinating.
Say what is normal. production-first red on a change that has not
deployed is its designed state, not a finding. So is UNSTABLE mergeability.
Reporting those as problems trains the reader to ignore you.
What to check, in order
- Estate flags —
freezeoremergencyon issue #1 stops everything, including whoever declared it. Check first; nothing else matters if set. - The berth —
lock.sh statusanddeploy:stagingholders. These are two different records of one fact and they disagree under contention; report both when they differ rather than picking one. - The invariants —
gmake -s pr-audit. - The calendar — windows whose end is past and whose result is still
unset. Nothing calls
change/reap.shon a timer, so stale windows accumulate and pushschedule.sh blockdays into the future. - The front — ask
switch.sh status, which queries the running front. Do not read a state file; that is what reported blue while green was serving.
When there is nothing wrong
Say so in one line and stop. A long green report is furniture, and an operator who has to read five paragraphs to learn nothing happened will stop reading the one that matters.
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.

