gate-lifecycle
Skill88 starsChanged 3 days agoMITView on GitHub
Keeps every gate (lint rule, CI check, hook, budget, debt baseline) proven, measured, and explicitly changed for as long as it exists. Use when a gate is added over existing violations, narrowed, suspended, trialed, or removed; when a baseline or exception list churns; when a gate passed while a known-bad input existed; or when a gate fails because it cannot evaluate rather than because of a violation. Do not use to place a check at a stage, decide how a gate blocks, or design the oracle it asserts; hand those to defect-shift-left, ci-cd-reliability-architecture, and test-strategy.
In this repository52 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- alchemy-orchestrateSkill
- alchemySkill
- architecture-as-code-javascriptSkill
- architecture-as-code-pythonSkill
- architecture-as-codeSkill
- architecture-guidelinesSkill
- bring-downSkill
- ci-cd-reliability-architectureSkill
- continuous-improvementSkill
- defect-shift-leftSkill
- dependency-lifecycleSkill
- evolutionary-database-designSkill
- functionality-complexity-tradeoffSkill
- gate-lifecycleSkill
- implementation-readinessSkill
- infrastructure-fitSkill
- morphogenetic-architectureSkill
- observability-designSkill
- push-outSkill
- requirements-groundingSkill
- requirements-topologySkill
- requirements-traceabilitySkill
- structural-simplificationSkill
- system-optimizationSkill
- test-strategySkill
- zero-copy-requirementsSkill
- alchemy-orchestrateSkill
- alchemySkill
- architecture-as-code-javascriptSkill
- architecture-as-code-pythonSkill
- architecture-as-codeSkill
- architecture-guidelinesSkill
- bring-downSkill
- ci-cd-reliability-architectureSkill
- continuous-improvementSkill
- defect-shift-leftSkill
- dependency-lifecycleSkill
- evolutionary-database-designSkill
- functionality-complexity-tradeoffSkill
- implementation-readinessSkill
- infrastructure-fitSkill
- morphogenetic-architectureSkill
- observability-designSkill
- push-outSkill
- requirements-groundingSkill
- requirements-topologySkill
- requirements-traceabilitySkill
- structural-simplificationSkill
- system-optimizationSkill
- test-strategySkill
- zero-copy-requirementsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 95
Run it through the gate as configured: a rule tested in isolation passes while the assembled configuration shadows 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
Gate Lifecycle
A gate that stopped checking reports the same green as one that found nothing. Here a gate is any automated check that can stop a change, not one of Alchemy's lettered gates.
Purpose: Every gate is proven, measured, and loosened only with a criterion, an owner, and an expiry.
Core Directives
- A gate is proven by a failure, not a pass. Green from a gate never seen red is a claim about the gate, not evidence about the code.
- Effective scope is measured, never assumed. A gate checks what its resolved configuration and aggregation reach, not what it declares.
- A debt baseline only shrinks on its own. Automation may remove entries; growth is named in the change that makes it.
- Indeterminate is not failed. A gate that could not evaluate said nothing about the code: not a pass, not a violation, not a reason to remove it.
- Every suspension expires. Something other than memory enforces its criterion, owner, and expiry.
- Retirement needs a criterion met. A gate leaves because what it checks is gone or covered, never because it is inconvenient.
- The gate's own tooling obeys the gate's rules. A script that regenerates a baseline or aggregates results carries its own negative control and cannot loosen the gate as a side effect.
Boundary
gate-lifecycle is a task-matched Alchemy companion, not a qualification stage,
gate, or acronym letter. It runs in a single pass when a §2 trigger fires: after
H places a check, beside CI/CD's gate behavior, or later on its own. It decides
whether a gate still does what it claims, not where it runs, how it blocks, or
what oracle it asserts.
defect-shift-leftowns the stage. A check its §6.4 made blocking comes here forPROVE; its "remove as theatre" branch takes its criterion from §6.RESTOREnever changes stage.ci-cd-reliability-architectureowns blocking and the Gate Evidence Contract. ItsPolicy: baselineis a debt list under §5; itsResult: not-applicableis aTRIALorRETIREwith this skill's guard. Whether a trial gate warns or blocks nothing stays there.test-strategyowns the oracle and per-test quarantine, whose fields aTRIALreuses. The aggregator enforcing that oracle needs a negative control here.architecture-as-code, its stack skills, andcontinuous-improvementown a new rule; this skill takes over when §2 fires.functionality-complexity-tradeoffowns whether an obligation stands:RETIREcites its verdict rather than reaching one.
A routine baseline shrink, a new rule nothing violates, a test added to a proven gate without changing its scope, and a pure placement or blocking question do not open this skill.
1. Gate Population
A project profile names where gates live and how they run. A sweep enumerates from there; a check found elsewhere is unmanaged, which is a finding. The earliest enforceable form is a registry whose entries each name a negative-control command run in CI, so a gate added without proof fails.
2. Triggers
Most of these fire with no edit to the gate.
| Trigger | Signal | Opens |
| born | Promoted to blocking, or added over existing violations | §3; §5 when violations exist |
| narrowed | A block, glob, ignore, filter, or skip changed its reach, or it passed while a known-bad input existed | §4 |
| churn | A baseline or budget moves on most commits, or grew with no named decision | §5 |
| suspended | Dropped, downgraded, or shortened "until" something, or that date passed | §6 |
| indeterminate | It could not evaluate: identity, environment, tooling, an unreachable dependency | §6 |
| orphaned | What it checks is gone, or its obligation lapsed | §6 |
3. Proof
PROVEN names a negative control and the last commit or run where it turned the
gate red. Without that record a gate is UNPROVEN, however long its green
history. The control is a committed known-bad input (rules, schemas), a self-test
feeding the runner a synthetic failing report (aggregators, budgets), or a
mutation of a tree copy (whole-tree gates).
- Run it through the gate as configured: a rule tested in isolation passes while the assembled configuration shadows it.
- Assert the declared message, rule, or exit code. A crash is also red, and a broken rule can hide behind the block that shadows it.
- A one-off probe proves the gate once. Where configuration can change under it, keep the control in CI.
4. Scope
Declared scope is what a gate says it governs; effective scope is what it reaches once configuration resolves and results aggregate. Measure each way a gate shrinks:
- Shadow. A later or broader block replaces a narrower rule's options, with no error. Compare declared against resolved configuration; every rule fires on its control.
- Narrow. A glob, ignore, path filter, or skip marker drops files, or a new directory falls outside. Every rule governs at least one file; the governed set equals the source set.
- Aggregate. A summary counts skipped, retried, flaky, or not-run as passed. Feed the aggregator a report holding each state; it must fail. Aggregation semantics are scope.
An undeclared shrink is RESTORE. A deliberate narrowing updates the declaration
in the same change and says why.
5. Ratchet
A baseline admits known violations so a gate can block new ones. It is a debt list, and one that moves both ways hides growth. Read references/ratchet.md for the reconcile algorithm and the dated expiry step.
- Shrinking needs no decision. Growth is refused unless the change names the entry; a blanket update flag is regeneration under another name.
- A red run never writes the baseline. Regenerating from the current tree turns the gate into a recording of whatever exists.
- A stale entry is removed or fails the gate. Kept, it is headroom for the next regression.
- A budget raise is growth.
6. Decisions
Choose one decision per gate. A decision that cannot satisfy its guard is not that decision.
| Decision | Admission test | Guard |
| KEEP | Proven, in scope, baseline absent or only shrinking | None |
| PROVE | UNPROVEN | The control runs in CI, or its last red is recorded |
| RESTORE | An undeclared scope delta | The control that would have caught the shrink |
| RATCHET | A baseline or budget can grow without a named decision | Unnamed growth refused; the list has an owner |
| TRIAL | Its cost is measured and its value in doubt, or an evaluation outage outlasts what delivery can absorb | A measurable criterion, an owner, and an expiry a dated step enforces |
| REINSTATE | A trial failed its criterion or expired undecided, or the gate was removed for erroring | It blocks again and its control goes red |
| RETIRE | What it checks is gone, a proven same-scope gate catches it earlier, or M retired the obligation | Names that evidence; control and configuration go in the same change |
Past expiry the default is REINSTATE. A renewal is a new TRIAL citing data
measured in the window that ended; a moved date is not a decision.
INDETERMINATE is a status, not a decision. The gate keeps blocking under a
distinct exit code or label, so nobody reads it as a defect or a pass. Repair
evaluation; the rest of the record sets the decision. A gate that needs
target-environment state goes back to defect-shift-left for 8a. Being unable to evaluate is never a reason to
RETIRE; neither is never having fired, which is UNPROVEN, not useless.
7. Output Contract
Gate: <rule / check / hook / budget / baseline, and where it runs>
Stage: <defect-shift-left stage, or unplaced>
Trigger: born | narrowed | churn | suspended | indeterminate | orphaned
Proof: PROVEN <negative control, last red commit or run> | UNPROVEN
Scope: declared <…> / effective <…> / delta <none | what is no longer checked>
Baseline: <none | count, trend, last accepted growth and the change that named it>
Status: BLOCKING | TRIAL | INDETERMINATE | RETIRED
Decision: PROVE | RESTORE | RATCHET | TRIAL | REINSTATE | RETIRE | KEEP
Guard: <criterion, owner, expiry that ends this decision, or none>
Handoff: <skill that owns the next decision, or none>
Next action: <one concrete action>
Verification: <command, red run, or Not run + reason>In a sweep, emit one table row per gate with these fields, then the UNPROVEN
count.
8. See Also
defect-shift-left- the stage a check belongs at.ci-cd-reliability-architecture- how a gate blocks and what it records.architecture-as-code- rule recipes and the file-scope audit.
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.

