Best for
- Use this skill when writing or reviewing Go that could use a 1.
dimetron/pi-go/.pi-go/skills/go-127/SKILL.md
What changed in Go 1.27 (released August 2026) and how it changes the way Go is written in pi-go. Use this skill when writing or reviewing Go that could use a 1.27 feature, when bumping the `go` directive in go.mod, when a build or test behaves differently after a toolchain upgrade, or when `code-guidelines-go` points here. Covers generic methods, struct-literal field selectors, generalized type inference, the go command and go fix changes, the size-specialized allocator, the `goroutineleak` pro
Decision brief
Released: August 2026, six months after Go 1.26. Go 1 compatibility holds; almost all programs compile and run unchanged.
Compatibility matrix
| Platform | Status | Evidence | What to check |
|---|---|---|---|
| Codex | Not declared | No explicit evidence | Portability before use |
| Claude Code | Not declared | No explicit evidence | Portability before use |
| Cursor | Not declared | No explicit evidence | Portability before use |
| Gemini CLI | Not declared | No explicit evidence | Portability before use |
Installation
The source command is displayed only when detected. A safe inspection prompt is always available so your agent can explain every action before execution.
npx skills add https://github.com/dimetron/pi-go --skill ".pi-go/skills/go-127"Inspect the Agent Skill "go-127" from https://github.com/dimetron/pi-go/blob/4cfb2ed862714ae5604f4e4980bb1d805c4ba71d/.pi-go/skills/go-127/SKILL.md at commit 4cfb2ed862714ae5604f4e4980bb1d805c4ba71d. List every install step, command, network request, credential, file read/write, external action, and rollback step. Explain whether it fits my task. Do not install or execute anything until I approve.
Workflow
go env on this host persists GOEXPERIMENT=none, which does not mean "no extra experiments" — it clears the 1.27 baseline, which enables JSONv2, GreenTeaGC, SizeSpecializedMalloc, Dwarf5 and RandomizedHeapBase64 (src/internal/buildcfg/exp.go:81). Verified consequences here:
The headline change. Before 1.27 a generic operation on a type had to be a package-level function; now it can live on the type.
The headline change. Before 1.27 a generic operation on a type had to be a package-level function; now it can live on the type.
A key in a struct literal is no longer restricted to a top-level field name, so promoted fields from an embedded struct can be set directly.
Inference now applies in every context where a generic function is assigned to or converted to a matching function type — not just plain assignment. Composite literals and conversions now infer too.
Permission review
The documentation asks the agent to run terminal commands or scripts.
$ go build ./... # importing encoding/json/v2The documentation asks the agent to run terminal commands or scripts.
go env GOEXPERIMENT # prints "none" -> baseline is clearedEvidence record
| Signal | Value | Evidence type | Meaning |
|---|---|---|---|
| Quality score | 91/100 | Computed | Documentation, specificity, maintenance, and trust rules |
| Repository stars | 147 | Source | Repository attention, not individual Skill quality |
| Compatibility | 0 platforms | Source | Declared in the catalog source record |
| Usage guide | automated source guide | Editorial | Generated or reviewed according to the visible evidence level |
Pinned source
Released: August 2026, six months after Go 1.26. Go 1 compatibility holds; almost all programs compile and run unchanged.
Sources: go.dev/doc/go1.27 (authoritative) ·
victoriametrics.com/blog/go-1-27 (worked examples).
Every claim below is from one of those two, or from reading
$(go env GOROOT)/src on the installed 1.27.0 toolchain.
Companion skill: code-guidelines-go holds the standing style rules. This skill holds
only what 1.27 changed.
GOEXPERIMENT=none is set on this machinego env on this host persists GOEXPERIMENT=none, which does not mean "no extra
experiments" — it clears the 1.27 baseline, which enables JSONv2, GreenTeaGC,
SizeSpecializedMalloc, Dwarf5 and RandomizedHeapBase64
(src/internal/buildcfg/exp.go:81). Verified consequences here:
$ go build ./... # importing encoding/json/v2
imports encoding/json/v2: build constraints exclude all Go files in .../encoding/json/v2
$ golangci-lint --version
golangci-lint has version 2.13.1 built with go1.27.0-X:nojsonv2,nogreenteagc,
norandomizedheapbase64,nosizespecializedmalloc
So on this machine encoding/json/v2 will not import, and the 1.27 allocator and GC
work are switched off. Before benchmarking anything or trying a v2 snippet:
go env GOEXPERIMENT # prints "none" -> baseline is cleared
go env -u GOEXPERIMENT # restore the default baseline (user decision — ask first)
GOEXPERIMENT=jsonv2 go build ./... # or scope it to one command
Do not silently change the user's go env config. Flag it, and scope the experiment to
the command you are running.
The headline change. Before 1.27 a generic operation on a type had to be a package-level function; now it can live on the type.
// ✗ Before 1.27 — generic op stranded at package scope
type Box[T any] struct{ v T }
func MapBox[T, U any](b Box[T], f func(T) U) Box[U] { return Box[U]{v: f(b.v)} }
// ✓ Go 1.27 — the method declares its own type parameter U
type Box[T any] struct{ v T }
func (b Box[T]) Map[U any](f func(T) U) Box[U] { return Box[U]{v: f(b.v)} }
b := Box[int]{v: 21}
label := b.Map(func(n int) int { return n * 2 }).
Map(func(n int) string { return fmt.Sprintf("value=%d", n) })
Two hard restrictions, both compiler-enforced:
Map[U any](...) inside an
interface fails with interface method must have no type parameters.Prescription for pi-go: this is the right tool for transforms on a concrete generic
container (a typed cache, a typed result set). It is the wrong tool anywhere the type
is consumed through an interface — and pi-go's design rule is "define interfaces at the
consumer" (code-guidelines-go §1). If a call site takes an interface, keep the
package-level generic function.
A key in a struct literal is no longer restricted to a top-level field name, so promoted fields from an embedded struct can be set directly.
type Base struct{ ID int }
type User struct {
Base
Name string
}
u := User{Base: Base{ID: 7}, Name: "Mittens"} // ✗ before 1.27, the only way
u := User{ID: 7, Name: "Mittens"} // ✓ Go 1.27
go fix's new embedlit modernizer rewrites the old form for you (see §2.4).
Inference now applies in every context where a generic function is assigned to or converted to a matching function type — not just plain assignment. Composite literals and conversions now infer too.
func first[T any](s []T) T { return s[0] }
func last[T any](s []T) T { return s[len(s)-1] }
ops := []func([]int) int{first[int], last[int]} // ✗ before 1.27: explicit instantiation
ops := []func([]int) int{first, last} // ✓ Go 1.27: element type drives T=int
The old failure was cannot use generic function without instantiation
(VM blog). If you see leftover [int]
instantiations in a slice or map literal of function types, they can now be deleted.
go command, modulesgo test runs stdversion by defaultgo test now invokes the stdversion vet check, which reports use of stdlib symbols
newer than the go directive in go.mod (adjusted by build tags on the file). This is
the guard that catches "works on my machine": calling strings.CutLast from a module
still declaring go 1.26 is now a test failure, not a silent success.
Prescription: when you use a 1.27 stdlib symbol, bump go.mod in the same change.
Do not //nolint around it.
From 1.27 the go command accepts a removed GODEBUG setting in go.mod (godebug
lines) and //go:debug comments if it is set to the final default value established
before removal. Set to an old value, the go command fails. So
//go:debug asynctimerchan=0 still builds; asynctimerchan=1 is now a hard error.
go mod tidy enforces a two-block require layoutFor modules declaring go 1.27 or later, go mod tidy merges duplicate require
blocks into at most two — direct and indirect — preserving attached comment blocks. A
comment block spanning a mixed set of directives is merged onto the direct block.
Expect a large go.mod diff on the first go mod tidy after the version bump.
Review it, do not fight it.
go fix modernizersVerified against go tool fix help on 1.27.0:
| Analyzer | Rewrites |
|---|---|
atomictypes | basic types in sync/atomic calls → atomic types |
embedlit | references to embedded fields in composite literals (see §1.2) |
slicesbackward | backward loops over slices → slices.Backward |
unsafefuncs | unsafe pointer arithmetic → function calls |
waitgroupgo | wg.Add(1)/go/wg.Done() → wg.Go — renamed from waitgroup |
fmtappendf was removed (stylistic concerns). If a script or CI job names
waitgroup or fmtappendf, it breaks.
go docgo doc example.com/[email protected] — documentation at a specific module version.go doc -ex <pkg> — list a package's runnable examples (verified: go doc -ex strings
interleaves func ExampleTrimSpace() lines under each symbol).go doc bytes.ExampleBuffer — prints the example's source with comments.bzr support removed from the go command; modules on Bazaar servers can no longer
be fetched directly.go tool trace -http=:6060 now binds localhost only, matching go tool pprof.
Pass -http=0.0.0.0:6060 to listen on all interfaces. Relevant to the go-pprof
workflow in CLAUDE.md.@file) are accepted by compile, link, asm, cgo, cover
and pack, in GCC's format.-macos and -macsdk set OS/SDK versions in LC_BUILD_VERSION.
Defaults: macOS 13.0.0, SDK 26.2.0.The compiler emits calls to size-specialized allocation routines, cutting the cost of some allocations under 80 bytes by up to 30%, ~1% overall in allocation-heavy programs, at ~60 KB more binary size (workload-independent). Nothing to change in code.
Opt-out is GOEXPERIMENT=nosizespecializedmalloc, expected to be removed in Go 1.28.
Per §0, GOEXPERIMENT=none on this host already disables it — so any allocation
benchmark run here is measuring the old allocator.
goroutineleak profile is now GAThe goroutine leak profile, experimental in 1.26, is generally available. No
GOEXPERIMENT needed; the goroutineleakprofile GOEXPERIMENT setting is deleted.
// scrape in-process
pprof.Lookup("goroutineleak").WriteTo(os.Stdout, 1)
Or the net/http/pprof endpoint /debug/pprof/goroutineleak — which is what pi-go
should use, since it already serves pprof on :6060 under pi --pprof true.
A goroutine counts as leaked when it is blocked on a concurrency primitive (channel,
sync.Mutex, sync.Cond, …) that the GC proves unreachable from any runnable goroutine
or anything those could unblock — so it can never be unblocked. Known blind spot:
leaks where the primitive is reachable through a global, or through the locals of a
runnable goroutine, are not detected. Long-lived registries in pi-go fall exactly in
that gap, so a clean profile is not proof of no leak. Keep goleak.VerifyTestMain(m) in
tests.
For modules with go 1.27 or later, runtime/pprof goroutine labels appear in the
traceback header line — in panics, SIGQUIT dumps and runtime.Stack output:
goroutine 1 [running] {request: 42}:
Anything already wrapped in pprof.Do(ctx, pprof.Labels(...), ...) now leaks that
context into crash output. Check pi-go's labels for anything sensitive (session ids
are fine; API keys or prompt text are not) before bumping the go directive. Opt-out is
GODEBUG=tracebacklabels=0, added in 1.26 and expected to be kept indefinitely.
== nil; 1.27 just exposes it more.//line / /*line*/ directives resolve against the containing
file's directory (#70478), matching go/scanner.encoding/json/v2 and encoding/json/jsontext (new)import (
json "encoding/json/v2"
"encoding/json/jsontext"
)
data, err := json.Marshal(Point{X: 1, Y: 2}, opts...) // variadic Options
err = json.MarshalWrite(w, v, opts...) // to an io.Writer
err = json.MarshalEncode(enc, v, opts...) // to a *jsontext.Encoder
err = json.Unmarshal(data, &v, opts...)
err = json.UnmarshalRead(r, &v, opts...)
err = json.UnmarshalDecode(dec, &v, opts...)
jsontext is the syntactic layer: Encoder, Decoder, Token, Value, with a state
machine enforcing valid JSON.
v2 chooses stricter defaults than v1: it rejects invalid UTF-8 in strings and
rejects duplicate object names. It also does not sort map keys (v1 always did) —
pass json.Deterministic(true) when you need stable output for golden files.
encoding/json (v1) is now backed by v2. Behavior is preserved; error message
text may differ. If any pi-go test asserts on a JSON error string, that is the thing
most likely to break — assert on error identity or a substring you control instead.
Marshal is at parity; unmarshal is significantly faster. Escape hatch:
GOEXPERIMENT=nojsonv2 (expected to be removed in a future release).
Prescription: no migration is required and v1 stays supported. Keep pi-go on v1 unless a call site actually wants v2's strictness or its options; do not do a mechanical sweep.
uuid (new, top-level stdlib package)Implements RFC 9562, seeded from a cryptographically secure RNG. UUID is [16]byte,
so it is comparable with == and usable as a map key.
id := uuid.New() // suitable for most purposes; currently == NewV4
r := uuid.NewV4() // 122 bits of randomness
k := uuid.NewV7() // 48-bit timestamp prefix — sorts by creation time
u, err := uuid.Parse("f81d4fae-7dec-11d0-a765-00a0c91e6bf6")
m := uuid.MustParse(s) // panics on bad input
uuid.Nil(); uuid.Max() // functions, not vars
u.String(); u.Compare(v) // Compare is big-endian, RFC 9562 §6.11
UUID implements encoding.TextMarshaler, TextAppender and TextUnmarshaler.
Prescription for pi-go: drop github.com/google/uuid for new code. For anything
written to $HOME/.pi-go/sessions/ or used as a database/session key, prefer NewV7 —
time-ordered ids sort chronologically, which is exactly what the session directory
listing wants.
crypto/mldsa (new) — post-quantum signaturesFIPS 204 ML-DSA, in three parameter sets returned by functions: mldsa.MLDSA44(),
MLDSA65(), MLDSA87(). Parameters exposes PublicKeySize(), SignatureSize(),
String().
priv, err := mldsa.GenerateKey(mldsa.MLDSA65())
sig, err := priv.Sign(rand.Reader, msg, crypto.Hash(0))
err = mldsa.Verify(priv.PublicKey(), msg, sig, nil)
crypto/x509 handles ML-DSA keys and signatures; crypto/tls gains the MLDSA44,
MLDSA65, MLDSA87 SignatureScheme values for TLS 1.3. crypto.MLDSAMu is a new
Hash sentinel for external-μ signing (it panics if passed to RegisterHash).
simd and simd/archsimd (experimental)GOEXPERIMENT=simd at build time. simd is portable and vector-size-agnostic
(Int8s, Int32s, Float32s, …; width varies by machine), supporting a scalable
subset of simd/archsimd. archsimd is architecture-specific and unstable: 128-bit on
wasm/arm64/amd64, 256-/512-bit on some amd64. Not for pi-go — no hot vector loops
here, and the API is not stable.
Signatures below were read from the 1.27.0 GOROOT, not from the release notes prose.
// strings / bytes — replaces the LastIndex dance
func CutLast(s, sep string) (before, after string, found bool) // strings
func CutLast(s, sep []byte) (before, after []byte, found bool) // bytes
before, after, found := strings.CutLast("a/b/c", "/") // "a/b", "c", true
before, after, found = strings.CutLast("nosep", "/") // "nosep", "", false
// testing/synctest — time.Sleep + synctest.Wait in one call
func Sleep(d time.Duration)
// net/http/httptest — in-memory network, no real port, registers t.Cleanup
func NewTestServer(t testing.TB, handler http.Handler) *Server
// hash/maphash — the contract for future hash-based containers
type Hasher[T any] interface {
Hash(*Hash, T)
Equal(x, y T) bool
}
type ComparableHasher[T comparable] struct{ /* Equal is == */ }
// math/big — quotient and remainder with an explicit rounding mode
func (z *Int) Divide(x, y, r *Int, mode RoundingMode) (*Int, *Int)
// modes: big.Trunc, big.Floor, big.Round, big.Ceil
// 7/2 with Ceil -> q=4, r=-1
// 7/2 with Floor -> q=3, r=1
// math/rand/v2 — the top-level N, now a method (first stdlib generic method)
func (r *Rand) N[Int intType](n Int) Int
r := rand.New(rand.NewPCG(1, 2))
r.N(100) // int in [0, 100)
r.N(5 * time.Second) // works for any integer-kind type, incl. Duration
// net/url
func (u *URL) Clone() *URL
func (vs Values) Clone() Values
Also new: database/sql.ConvertAssign(scanCtx driver.ScanContext, dest any, src driver.Value) error;
database/sql/driver.RowsColumnScanner; go/constant.StringLen(x Value) int64;
(*go/scanner.Scanner).End() token.Pos; (*go/token.File).String();
go/types.Hasher and HasherIgnoreTags.
Prescriptions:
strings.CutLast / bytes.CutLast replace LastIndex + two slice expressions.
Grep pi-go for strings.LastIndex when touching those call sites.synctest.Sleep(d) replaces the time.Sleep(d); synctest.Wait() pair that
code-guidelines-go §6 currently spells out. Prefer it inside a bubble: when the test
and the code under test sleep the same duration, plain time.Sleep leaves the wake
order unpredictable.httptest.NewTestServer(t, h) binds no TCP port and self-cleans, so defer srv.Close()
goes away and round-trips can run in synthetic time under testing/synctest. Worth
trying against the internal/cli sandbox trap in CLAUDE.md, where
httptest.NewServer panics in newLocalListener — but that has not been demonstrated
here, so verify before relying on it.maphash.Hasher[T] is a contract for future containers; there is no stdlib hash set
yet. Use it only if pi-go writes its own hash-based container, and honour the rule that
Equal values must Hash identically.| Package | Change | What to do |
|---|---|---|
net/http | HTTP/1 Response.Body auto-drains unread content on Close (to a conservative limit) so the connection can be reused | Usually a win. If you Close early to abort a large download, set Transport.DisableKeepAlives = true |
net/http | HTTP/2 server honours RFC 9218 client priorities | Server.DisableClientPriority = true restores round-robin |
net/http | New Server.MaxHeaderValueCount caps accepted header values; default http.DefaultMaxHeaderValueCount (500) | Set it on pi-go's servers if a client legitimately sends many values |
net/http | ALPN negotiation now works on a user-supplied net.Conn that has ConnectionState() tls.ConnectionState | — |
compress/flate | Compression speed improved; exact encoded bytes differ from 1.26 — propagates to archive/zip, compress/gzip, compress/zlib, image/png | Any golden file holding compressed bytes will fail. Regenerate; do not assert on compressed output |
unicode | Unicode 15 → 17 | Code points unassigned before (e.g. U+1FADC) now report IsSymbol/IsGraphic true. Golden files of width-measured or classified text may shift |
net | UnixConn read methods return io.EOF directly instead of wrapping in net.OpError | Replace net.OpError unwrapping with errors.Is(err, io.EOF) |
crypto/x509 | SystemCertPool respects SSL_CERT_FILE/SSL_CERT_DIR on Windows and Darwin, switching to the Go verifier | GODEBUG=x509sslcertoverrideplatform=0 restores platform APIs |
crypto/x509/pkix | RDNSequence.String / Name.String render string-typed values as strings even for unrecognized OIDs (previously hex DER) — #33093 | Certificate-subject golden strings may change |
crypto/ecdsa | PrivateKey.Sign validates hash length when SignerOpts is non-nil | Passing a mismatched digest now errors |
runtime/secret | Goroutines created in secret mode now run in secret mode | — |
Removed permanently — these now fail the build or the go command:
| Removed | Since | Replacement / effect |
|---|---|---|
asynctimerchan GODEBUG | 1.23 | time channels are always unbuffered. asynctimerchan=1 is now a hard error; =0 still parses (§2.2) |
gotypesalias GODEBUG | 1.22 | go/types always produces an Alias node for alias declarations |
tlsunsafeekm, tlsrsakex, tls10server | 1.22 | no escape hatch left |
tls3des, x509keypairleaf | 1.23 | no escape hatch left |
goroutineleakprofile GOEXPERIMENT | 1.26 | the profile is GA (§3.2) |
bzr support in the go command | — | vendor the module, or move it |
fmtappendf go fix analyzer | — | none (dropped on style grounds) |
Renamed: go fix waitgroup → waitgroupgo.
Newly deprecated: crypto/tls.Config.Rand — for deterministic testing use
testing/cryptotest.SetGlobalRandom. Post-quantum hybrid key exchanges can now be
enabled explicitly via Config.CurvePreferences even under tlsmlkem=0 /
tlssecpmlkem=0; those GODEBUGs were only ever meant to affect the default set used when
CurvePreferences is nil.
Temporary opt-outs, on a clock — do not build a workflow on these:
GOEXPERIMENT=nosizespecializedmalloc (expected gone in 1.28) and
GOEXPERIMENT=nojsonv2 (expected gone in a future release).
Work through this when bumping the go directive to 1.27:
go.mod: bump the go line in the same change as the first 1.27 stdlib symbol —
go test now runs stdversion and will fail otherwise (§2.1).go mod tidy: expect the require blocks to be consolidated into two (§2.3).
Review the diff once, then move on.go.mod godebug lines and //go:debug comments for
asynctimerchan, gotypesalias, tlsunsafeekm, tlsrsakex, tls3des,
tls10server, x509keypairleaf. Any of them set to a non-default value now fails.pprof.Labels value carries a secret before tracebacks
start printing them (§3.3).CGO_ENABLED=0 for a static binary from a cgo-optioned program.//go:linkname: 1.27 tightens unsanctioned linknames (a linknamestd directive
restricts some to the stdlib, and cmd/link now checks linkname access to assembly
symbols). A dependency reaching into runtime internals is the likely breakage; test
the dependency tree early. (This one is from the VM blog's "hidden gems", not the
release notes — treat it as a heads-up to verify, not as a documented contract.)golangci-lint v2.13.1 parses generic methods and struct-literal field
selectors cleanly — verified locally on both constructs, 0 issues, exit 0.| Concern | Go 1.27 | Was |
|---|---|---|
| Generic op on a concrete container | method with own type params | package-level generic func |
| Setting a promoted field | User{ID: 7} | User{Base: Base{ID: 7}} |
| Generic func in a slice literal | []func([]int) int{first, last} | {first[int], last[int]} |
| Split on last separator | strings.CutLast / bytes.CutLast | strings.LastIndex + slicing |
| Sleep inside a synctest bubble | synctest.Sleep(d) | time.Sleep(d); synctest.Wait() |
| HTTP test server | httptest.NewTestServer(t, h) | httptest.NewServer(h) + defer Close() |
| UUIDs | stdlib uuid (NewV7 for keys) | github.com/google/uuid |
| Bounded random from own source | r.N(100) | rand.N(100) (global) |
| Rounded integer division | big.Int.Divide(x, y, r, big.Floor) | Quo/Mod + manual adjust |
| Deep-copy a URL | u.Clone() | manual struct copy |
| Leaked goroutines in prod | /debug/pprof/goroutineleak | experiment / goleak in tests only |
| Deterministic JSON map order | json.Deterministic(true) (v2) | implicit in v1 |
wg.Add/Done modernizer | go fix -waitgroupgo | -waitgroup |
Frequently asked questions
Released: August 2026, six months after Go 1.26. Go 1 compatibility holds; almost all programs compile and run unchanged.
The source record exposes this install command: npx skills add https://github.com/dimetron/pi-go --skill ".pi-go/skills/go-127". Inspect the command and pinned source before running it.
Static rules flagged exec-script in the source; the page lists the matching lines and excerpts.
Alternatives
coreyhaines31/marketingskills
When the user wants to plan, design, or implement an A/B test or experiment, or build a growth experimentation program. Also use when the user mentions "A/B test," "split test," "experiment," "test this change," "variant copy," "multivariate test," "hypothesis," "should I test this," "which version is better," "test two versions," "statistical significance," "how long should I run this test," "growth experiments," "experiment velocity," "experiment backlog," "ICE score," "experimentation program
garrytan/gbrain
End-to-end discipline for turning any large data source (audio libraries, email takeouts, document corpora, chat exports, API dumps) into brain pages at scale. The lifecycle spine: SCHEMA → ACCESS → TRIAL → EVALUATE → IMPROVE → CODIFY → TEST → SKILLIFY → BULK → MONITOR. State is tracked in a durable JSON manifest (see MANIFEST-PATTERN.md) so any crash, session boundary, or subagent fan-out resumes from ground truth instead of memory.
alirezarezvani/claude-skills
App Store Optimization (ASO) toolkit for researching keywords, analyzing competitor rankings, generating metadata suggestions, and improving app visibility on Apple App Store and Google Play Store. Use when the user asks about ASO, app store rankings, app metadata, app titles and descriptions, app store listings, app visibility, or mobile app marketing on iOS or Android. Supports keyword research and scoring, competitor keyword analysis, metadata optimization, A/B test planning, launch checklist
dotnet/skills
Migrates .NET test projects from VSTest to Microsoft.Testing.Platform (MTP). Use when user asks to "migrate to MTP", "switch from VSTest", "enable Microsoft.Testing.Platform", "use MTP runner", set OutputType=Exe only for test projects in Directory.Build.props, or mentions EnableMSTestRunner, EnableNUnitRunner, or UseMicrosoftTestingPlatformRunner. USE FOR: MTP behavioral differences vs VSTest (exit code 8, zero tests discovered, --ignore-exit-code, TESTINGPLATFORM_EXITCODE_IGNORE); centralizing