Bump kin to oasdiff/kin-openapi @ e14b38a (the open getkin/kin-openapi#1185
branch with multi-error + simple-leaf conversions + plural-examples
typing fix). Lets oasdiff exercise the changes end-to-end before the kin
PR merges; the replace will drop and the dep will return to upstream
master once #1185 lands.

Wire-up:

 - internal/validate.go: pass openapi3.EnableMultiError() to Validate so
   simple-leaf and container errors aggregate rather than fail-fast.
 - internal/validate.go (fieldLoc): when the dotted Field name on a
   cluster error (e.g. "info.version") doesn't match a Fields key, fall
   back to the suffix after the last dot ("version"). Kin's Origin.Fields
   is keyed by the leaf name as it appears in the YAML mapping, while
   cluster errors use a dotted form for rule-ID disambiguation; without
   this fallback findings under aggregated leaves resolved to the parent
   object's Origin.Key instead of the precise field.

Concrete result on /tmp/multi-problem.yaml (empty info.title, empty
info.version, missing operation.responses): three findings at the exact
field lines and columns where the value is missing, instead of one
finding at the info section start.

Tests updated:

 - missing-required-info.yaml fixture: title now set, only version left
   empty so single-finding tests stay single.
 - Test_ValidateCmd_LineColumnFromOrigin: now asserts line 4 col 3 (the
   version line in the fixture).
 - Test_ValidateCmd_DocRootFieldHasLineColumn (renamed from
   ...HasNoLineColumn): asserts doc-root findings carry line:1 col:1 now
   that T.Origin is populated (kin #1184).
 - Test_ValidateCmd_TextFormatLocation: location string updated to 4:3.

kin getkin/kin-openapi#1185 (multi-error aggregation + simple-leaf
conversions + plural-examples typing + precise example-value Origin)
merged today as e56c2c7. Drop the temporary replace directive that
pointed at the oasdiff fork branch and bump to a pseudo-version of
upstream master that includes the merge.

Five fmt.Errorf call sites in the openapi3 validators returned bare
errors that consumers could only inspect by substring matching. Wrap
each in a typed cluster so errors.As / errors.Is consumers can dispatch
on the error kind and pull structured fields out:

- PathParameterRequiredError  (parameter.go)        Param + Origin
- DuplicateOperationIDError   (paths.go)            OperationID + Endpoint1/Endpoint2
- ExtraSiblingFieldsError     (extension.go, refs)  Fields []string
- SchemaTypeError             (schema.go)           Type + Origin

Error() strings are unchanged byte-for-byte (existing string-comparison
consumers, apis_guru golden fixtures, validation_error_test assertions
all pass without edits). The benefit is purely additive: callers gain
the typed-cluster dispatch they already use for the RequiredFieldError /
FieldVersionMismatchError / etc. clusters added in getkin#1183.

Promised in getkin#1185's heads-up comment after running oasdiff validate
across ~500 fixtures and finding these as the remaining 22% of
findings that fell through to a downstream catchall.

* feat(validate): new subcommand wrapping kin-openapi's Validate()

Adds 'oasdiff validate <spec>' that flags per-RFC OpenAPI / JSON
Schema spec violations. Wraps kin-openapi's openapi3.T.Validate()
walker and dispatches each typed error to a stable kebab-case rule
ID via errors.As against kin's RequiredFieldError /
FieldVersionMismatchError clusters (introduced in kin-openapi#1166).

Output format follows the changelog command's shape (id/text/level/
source) so a single CI script can parse both. Two formats: text
(default, with a header summary line + per-finding block matching
ApiChange.MultiLineError style) and yaml (-f yaml).

Exit code 0 on no findings, 1 if any.

10 tests cover: happy path, typed dispatch (info-version-required,
openapi-required, identifier-field-for-3-1-plus, webhooks-field-for-
3-1-plus, const-field-for-3-1-plus through 3 levels of %w wrapping),
untyped fallback (kin-validation-error for sites kin hasn't migrated
to a typed cluster yet), text + yaml format shapes, load-failure
exit code distinct from validation-finding exit code.

DO NOT MERGE - depends on kin-openapi#1166 landing and a kin release
shipping. The go.mod replace directive pins kin to the fork's rfc
branch.

Cleanup before merge:
  1. Remove the replace directive
  2. Bump kin-openapi to the released version
  3. go mod tidy

* fix(validate): use validateCmd const to silence unused-const lint

* feat(validate): surface line + column from kin's typed Origin

Loader now starts with IncludeOrigin=true. The Finding struct gains
Line/Column fields, populated from the kin cluster errors' Origin.Key
when available (info, license, server, schema). Document-root fields
(openapi, webhooks, jsonSchemaDialect) have nil Origin and emit
findings without line/column (yaml omitempty).

Text format renders <file>:<line>:<column> when origin is set, plain
<file> otherwise — matches the changelog command's location shape.

Three new tests pin: line/column populated for info-version-required,
both fields absent for openapi-required (doc-root), text format
includes :line:column when available.

* chore: drop accidentally-committed .DS_Store

* fix(validate): indent every line of multi-line finding messages

kin errors like *SchemaError embed newlines in their Error() output
(Schema and Value dumps). Without indenting continuation lines, those
broke the finding's visual grouping in text format. Every \n in the
message now becomes \n\t so the whole finding stays under the same
tab indent.

* chore: gitignore .DS_Store

* chore: gitignore .DS_Store

* fix(validate): leave blank lines blank when indenting multi-line text

Previous indent logic prefixed every \n with \t, including blank
lines — leaving stray tabs on otherwise-empty separator lines and a
trailing \t at the end of the message. Switch to a line-by-line
walk that skips blanks and trims trailing whitespace.

* feat(validate): dispatch SchemaValueError cluster (example/default-violates-schema)

kin#1166 added a *SchemaValueError cluster wrapping the *SchemaError
returned by VisitJSON when a schema's example or default doesn't
satisfy its own constraints. Add a third dispatch arm so findings of
that shape get a specific rule ID derived from the cluster's
ValueKind: 'example' → 'example-violates-schema',
'default' → 'default-violates-schema'.

The cluster also carries the parameter/media-type/schema's Origin, so
Line+Column now populate for these findings as well.

* deps: bump kin-openapi to include long-tail RequiredFieldError leaves

* deps: switch kin-openapi from oasdiff fork to upstream master

All four typed-validation-error PRs now merged into upstream
getkin/kin-openapi:
- #1162 (openapi3conv 3.0→3.x canonicalization)
- #1166 (ValidationError framework)
- #1170 (long-tail RequiredFieldError leaves)
- #1180 (combined long-tail PR collapsing #1171-#1179)

Drop the replace directive; pin to upstream-master pseudo-version
e8145f8f4d2b (the #1180 merge commit). When getkin tags a release
containing this work, the require line will move from pseudo-version
to a stable v0.138.x.

* feat(validate): dispatch the 8 remaining kin clusters to typed rule IDs

After bumping kin to upstream master post-#1180, eight cluster types
became available that the validate dispatch was still routing to the
'kin-validation-error' catchall. Wire each:

- *PathParametersError -> path-parameters-mismatch (static, since the
  cluster carries Path/Method/Missing not a single derivable field)
- *MutuallyExclusiveFieldsError -> <f1>-<f2>-mutually-exclusive
- *ForbiddenFieldError -> <field>-forbidden
- *ServerURLTemplateError -> server-url-template-invalid (static, the
  cluster carries only the offending URL)
- *EitherFieldRequiredError -> <f1>-or-<f2>-required
- *SchemaBothFormsExclusive -> <field>-both-forms-exclusive
- *ExactlyOneFieldError -> <f1>-or-<f2>-exactly-one
- *SingleEntryContentError -> <subject>-content-single-entry
- *WebhookNilError -> webhook-nil

keyOriginForKinError extended in the same way so line/column flow
through for clusters that carry Origin (all except WebhookNilError,
whose offending key sits on the document root that the loader
doesn't track per-key).

Updated the previously-flipping Test_ValidateCmd_UntypedKinErrorFallback
(server-url-mismatched-braces is now typed) and added a test for the
user-reported PathParametersError case.

* deps: temporarily point kin-openapi at oasdiff fork with DisableTimestamps

Switches the kin-openapi dep to the oasdiff fork's
feat/yaml-disable-timestamps branch via a replace directive.

This is the same branch underlying getkin/kin-openapi#1181 (still
in review). It includes the typed validation errors from #1166 and
#1180 plus the DisableTimestamps integration with oasdiff/yaml
v0.1.0, which prevents YAML 1.1 implicit-timestamp resolution from
mangling date-shaped map keys in real-world specs.

Verified with the canonical case: unicourt.com/1.0.0/openapi.yaml
(originally cited in invopop/yaml#10 four years ago) now loads
cleanly and produces actual schema-validation findings rather than
the time.Time map-key explosion.

REMOVE BEFORE MERGE: replace this directive with a bump to the
released kin version once #1181 lands and a kin tag is cut.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(validate): align Finding shape with locked JSON schema

Restructures the validate output to match the design's locked JSON
schema (mirrors oasdiff changelog --format json with a few additions).

Finding struct now exposes:
- source as an object {file, line, column} (was: string source + top-level
  line/column)
- section: which top-level doc section the finding belongs to
  ("info", "paths", "components", "webhooks", "servers", "security",
  "tags", or "" for unscoped doc-root findings). Determined per-cluster
  with a light Field-prefix check.
- fingerprint: stable 12-char sha256-prefix derived from
  "{id}:{operation}:{path}:{args}" — mirrors the existing
  formatters/changes.go:computeFingerprint scheme so the Pro
  PR-comment can partition findings into new/pre-existing/fixed via
  set membership on fingerprint across the base/revision spec pair.
- comment, operation, path: present but omitempty (Phase 1 leaves
  operation/path empty; extracting them from kin Origins requires
  walking the spec structure, deferred to Phase 2).
- All fields have both yaml: and json: tags.

Adds --format json support (encoder + enum value); existing yaml and
text paths unchanged in behaviour.

Tests:
- Existing yaml-format + origin-tracking tests updated to the new
  source-object shape.
- New JSON-format test pins the locked shape.
- New fingerprint-stability test pins determinism across runs (the
  property Pro PR-comment partitioning depends on).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(validate): colorize text output via --color flag

Adds the locked --color auto|always|never flag to oasdiff validate,
matching the convention used by changelog and breaking. Text output
now renders the severity in red/purple/cyan (error/warning/info,
via Level.StringCond) and the rule ID in yellow, matching the
established color scheme. Source path stays uncolored.

Auto mode disables color when stdout is piped or redirected; the
auto-detect helper lives in checker/piped_output.go and is shared
across subcommands so behaviour is consistent.

Exports checker.IsColorEnabled as a thin wrapper around the
package-internal isColorEnabled. Lets oasdiff packages outside
checker (validate, future subcommands) gate their own color logic
without duplicating the auto-detect convention.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(validate): pin findings to the offending field line, not the enclosing object

kin's Origin model carries two location handles:
- Origin.Key       — the start of the enclosing collection
                     (for a license-identifier error, the `license:` key)
- Origin.Fields[X] — the specific scalar field X inside that collection
                     (for a license-identifier error, the `identifier:` line)

The previous locator returned Origin.Key for every cluster, so a finding
in `data/validate/license-identifier-in-3-0.yaml` was pinned to line 5
(`license:`) instead of line 7 (`identifier: MIT`). Reviewers reading
the output had to scan from the parent key to find the actual offender.

Reworks locationForKinError (renamed from keyOriginForKinError) to prefer
Origin.Fields[Field] when the cluster carries a Field, falling back to
Origin.Key when the per-field entry is missing (e.g. empty values, which
kin doesn't track per-field). New fieldLoc helper centralises the lookup.

Per-cluster strategy:
- RequiredFieldError, FieldVersionMismatchError, ForbiddenFieldError,
  SchemaBothFormsExclusive — use Fields[Field]
- MutuallyExclusiveFieldsError — use Fields[Field1] (either field pins
  to the right object)
- SchemaValueError — use Fields[ValueKind] (e.g. "example", "default")
- SingleEntryContentError — use Fields[Subject]
- EitherFieldRequiredError, ExactlyOneFieldError — use Key (cluster fires
  when NONE of the fields are present, so per-field lookup wouldn't match)
- PathParametersError, ServerURLTemplateError — use Key (no Field metadata)
- WebhookNilError — no Origin (doc root)

Adds a regression test asserting the License-identifier fixture pins to
line 7:5 (the identifier line) rather than 5:3 (the license: line).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(validate): extract path + operation from kin's error chain

kin wraps validation errors in nested fmt.Errorf("...: %w") layers
that carry path/operation context as plain text rather than typed
fields:

  invalid paths: invalid path /thing: invalid operation GET: <inner>

These three layers (or any subset) are now stripped off the rendered
message and surfaced as Finding.Operation and Finding.Path, matching
the changelog command's convention of presenting them as discrete
fields. Finding.Text holds the cleaned inner message.

Text output gains a second header line "in API <op> <path>" rendered
in green when color is enabled, mirroring changelog / breaking. Doc-
root findings (info, openapi, license — no path/operation) skip the
line entirely.

Before:
  error [const-field-for-3-1-plus] at spec.yaml:18:21
        invalid paths: invalid path /thing: invalid operation GET: field const is for OpenAPI >=3.1

After:
  error [const-field-for-3-1-plus] at spec.yaml:18:21
        in API GET /thing
        field const is for OpenAPI >=3.1

Side benefit: Finding.Operation / Finding.Path are now populated for
operation-scoped findings, which feeds into the fingerprint (per
formatters/changes.go scheme: sha256(id:op:path:args)) and makes
Pro PR-comment partitioning stable when the same finding moves to a
new path between base and revision.

Component-scoped findings (kin's "invalid components: schema X:"
wrapper) are still inline in Text. Section is set correctly via
sectionForKinError, but extracting the schema name into a discrete
field is a separate enhancement.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Revert "feat(validate): extract path + operation from kin's error chain"

This reverts commit f9bfe39.

The reverted commit used regex to strip kin's "invalid paths: invalid
path X: invalid operation Y:" prefixes off the rendered error message.
That was always meant to be interim — message-text parsing is exactly
the brittleness kin's typed-error work eliminates.

kin PR getkin/kin-openapi#1183 adds typed SectionContextError /
PathContextError / OperationContextError wrappers that carry the
context as structured fields. Once that merges and a kin release is
cut, the extraction becomes three errors.As calls with no string
parsing.

Carrying the regex in the interim isn't worth it: the whole
feat/validate-command branch is gated on a kin release anyway, so
there's no window where the regex would ship to users. Reverting now
keeps the branch honest and avoids a "delete the regex" follow-up
commit later.

The Finding struct keeps its Operation / Path fields (added in the
schema-alignment commit, not here) — they're simply unpopulated until
the typed-error extraction lands. omitempty elides them from output
in the meantime. Validate findings temporarily render the full
wrapped message inline again, as they did before f9bfe39.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(validate): typed path/operation/section extraction from kin's error chain

Now that getkin/kin-openapi #1183 is merged, validate lifts the
structural scope of each finding out of the message text and into the
typed fields:

- Operation / Path come from PathValidationError + OperationValidationError
  via errors.As (replacing the reverted regex approach).
- sectionForKinError now prefers SectionValidationError.Section, kin's
  authoritative section name, falling back to the pre-#1183 cluster
  heuristics only for doc-root errors that aren't section-wrapped.

Adds data/validate/operation-missing-responses.yaml and a test asserting
an in-operation error surfaces operation=GET, path=/things, section=paths.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(validate): strip the redundant context prefix from finding text

Section/path/operation scope now lives in the Finding's typed fields,
so the "invalid paths: invalid path X: invalid operation Y:" prefix that
kin's context wrappers add to Error() was pure duplication in Text.
unwrapContext peels those typed wrappers off the front of the chain so
Text carries only the underlying leaf message.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* validate: pin kin to PR #1185 branch for multi-error + precise locations

Bump kin to oasdiff/kin-openapi @ e14b38a (the open getkin/kin-openapi#1185
branch with multi-error + simple-leaf conversions + plural-examples
typing fix). Lets oasdiff exercise the changes end-to-end before the kin
PR merges; the replace will drop and the dep will return to upstream
master once #1185 lands.

Wire-up:

 - internal/validate.go: pass openapi3.EnableMultiError() to Validate so
   simple-leaf and container errors aggregate rather than fail-fast.
 - internal/validate.go (fieldLoc): when the dotted Field name on a
   cluster error (e.g. "info.version") doesn't match a Fields key, fall
   back to the suffix after the last dot ("version"). Kin's Origin.Fields
   is keyed by the leaf name as it appears in the YAML mapping, while
   cluster errors use a dotted form for rule-ID disambiguation; without
   this fallback findings under aggregated leaves resolved to the parent
   object's Origin.Key instead of the precise field.

Concrete result on /tmp/multi-problem.yaml (empty info.title, empty
info.version, missing operation.responses): three findings at the exact
field lines and columns where the value is missing, instead of one
finding at the info section start.

Tests updated:

 - missing-required-info.yaml fixture: title now set, only version left
   empty so single-finding tests stay single.
 - Test_ValidateCmd_LineColumnFromOrigin: now asserts line 4 col 3 (the
   version line in the fixture).
 - Test_ValidateCmd_DocRootFieldHasLineColumn (renamed from
   ...HasNoLineColumn): asserts doc-root findings carry line:1 col:1 now
   that T.Origin is populated (kin #1184).
 - Test_ValidateCmd_TextFormatLocation: location string updated to 4:3.

* validate: bump kin pin to 9b85457 (example value origin fix)

* validate: drop the kin fork replace; bump to kin master with #1185

kin getkin/kin-openapi#1185 (multi-error aggregation + simple-leaf
conversions + plural-examples typing + precise example-value Origin)
merged today as e56c2c7. Drop the temporary replace directive that
pointed at the oasdiff fork branch and bump to a pseudo-version of
upstream master that includes the merge.

* validate: dispatch typed clusters from kin #1187

Wires oasdiff validate against the four new typed validation
clusters added in getkin/kin-openapi #1187 plus the Origin tracking
this branch pushed back to that PR for the duplicate-operation-id
and extra-sibling-fields sites.

Before this change, the four corresponding error sites all fell
through to the kin-validation-error catchall and lost line:column
precision. After this change each surfaces under its own stable
rule ID with file:line:col when the loader tracks origins.

Dispatcher updates in internal/validate.go:

* ruleIDForKinError — four new errors.As cases returning the
  stable rule IDs path-parameter-required, duplicate-operation-id,
  extra-sibling-fields, schema-type-unsupported.

* locationForKinError — four new cases:
  - PathParameterRequiredError → parameter object's Key.
  - SchemaTypeError → the offending type field via fieldLoc.
  - DuplicateOperationIDError → fieldLoc(Origin, "operationId")
    so the finding pins to the duplicate operationId scalar inside
    the second operation, not the operation block start.
  - ExtraSiblingFieldsError → parent object's Key (parser does
    not track Origin for unknown sibling field names).

* sectionForKinError — PathParameterRequiredError and
  DuplicateOperationIDError both resolve to "paths"; the other two
  fall through to the existing schema/generic logic which already
  handles them correctly.

* argsForKinError — fingerprint disambiguation for all four
  (Param / OperationID / joined Fields / Type).

Tests + fixtures:

* data/validate/{path-parameter-not-required, duplicate-operation-id,
  extra-sibling-fields, schema-type-unsupported}.yaml — minimal
  specs that each trigger exactly one of the four clusters.

* internal/validate_test.go — one test per fixture asserting the
  stable rule ID, the kin-side error message, and that Origin
  resolves to a file:line:col suffix.

go.mod pins to the oasdiff fork's branch via pseudo-version. Swap
to the upstream tag once getkin/kin-openapi cuts a release that
includes #1187.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* validate: dispatch InvalidParameterInError and SchemaPatternRegexError

Wires oasdiff validate against the two additional typed clusters
added to getkin/kin-openapi #1187 after corpus testing surfaced
them as the dominant remaining catchall sites.

Corpus result on testdata/APIs-guru-openapi-directory (~4,000 specs):
catchall hits dropped from 385 to 10 after this change. The remaining
10 cluster into three untyped sites (two in security_scheme.go, one
in ref.go) that are out of #1187's scope.

Dispatcher updates in internal/validate.go:

* ruleIDForKinError — two new errors.As cases returning the stable
  rule IDs parameter-in-invalid and schema-pattern-regex-invalid.

* locationForKinError — two new cases pinning to the offending
  field via fieldLoc: parameter.in for InvalidParameterIn,
  schema.pattern for SchemaPatternRegex.

* sectionForKinError — InvalidParameterIn → "paths". The other
  falls through to the existing schema-deep logic.

* argsForKinError — fingerprint disambiguation by the structured
  field (Value / Pattern).

Tests + fixtures: data/validate/{parameter-in-invalid,
schema-pattern-regex-invalid}.yaml minimal specs and corresponding
tests in internal/validate_test.go asserting typed rule ID, the
kin-side error message, and Origin file:line:col resolution.

go.mod bumped to the latest oasdiff/kin-openapi fork pseudo-version
v0.0.0-20260517110407-8e7311f2c94f (corresponds to #1187 HEAD with
the InvalidParameterIn + SchemaPatternRegex additions and the
SchemaPatternRegex message revert that preserves Error() string).

* validate: dispatch remaining 8 typed clusters from kin #1187; rename catchall to spec-validation-error

Adds errors.As cases (rule ID, section, args, location) for the
typed clusters in kin #1187 that didn't yet have oasdiff
dispatchers: InvalidSecuritySchemeTypeError, InvalidHTTPSchemeError,
UnresolvedRefError, APIKeyInInvalidError, PathMustStartWithSlashError,
ConflictingPathsError, DuplicateParameterError,
InvalidSerializationMethodError.

Also extends unwrapContext to strip the 10 new typed context wrappers
kin #1187 added (ComponentValidationError, ExternalDocsURLValidationError,
HeaderFieldValidationError, MediaTypeExampleValidationError,
WebhookValidationError, ParameterFieldValidationError,
ParameterExampleValidationError, SecuritySchemeFlowValidationError,
OAuthFlowValidationError, OAuthFlowFieldValidationError) so Finding.Text
carries just the underlying typed message without the wrapper prefixes.

Corpus result on testdata/APIs-guru-openapi-directory (~4,000 specs):
zero catchall hits across the entire corpus.

Renames the catchall rule ID from 'kin-validation-error' to
'spec-validation-error'. The old name leaked an implementation detail
(kin-openapi is an oasdiff dependency users don't know about); the new
name is meaningful in the user's domain (their OpenAPI spec failed
validation). The constant is renamed in lockstep
(kinUnknownID -> unknownValidationID).

go.mod bumped to kin fork pseudo-version with the full PR.

* validate: dedupe defects reported once per $ref site

A defect in a shared components definition (e.g. a bad example in
components/schemas/Status) is reported once by kin's components walk
AND once per operation that $refs it. From the user's perspective
that's one thing to fix; from the Pro PR-comment workflow's
perspective it should be one approve/reject decision.

dedupePreferringComponents groups findings by their defect identity
(Id + Source location + Text — Text carries the args discriminator)
and keeps one representative per group. When a group includes a
components-rooted finding (Section == "components"), prefer it: the
components-rooted version points at the definition site and has
empty Operation/Path, giving a stable fingerprint across reference-
graph changes (adding a 7th endpoint that $refs the schema doesn't
churn the fingerprint).

Covers all components/* sub-sections (schemas, parameters, headers,
requestBodies, responses, examples, links, callbacks,
securitySchemes, and 3.1+ pathItems) because the dedup looks at
Section, not the specific bucket. Path-level shared parameters don't
need handling here because kin only validates them once at the
PathItem level (no per-operation re-validation).

Regression test pins a fixture with components/schemas/Status that
has a bad example referenced from 3 operations: produces one finding
located at the components-schema definition's line:col, not three
finding-sized copies pointing at $ref sites.

* validate: rewrite Long help text to drop implementation leaks

Dropped 'kin-openapi', 'Phase 1', 'errors.As / RequiredFieldError /
FieldVersionMismatchError clusters' — internal implementation details
users don't care about. Replaced with user-facing content: what the
command catches, the output format options, and a clear exit-code
table including 102 for load failure.

* text formatter: drop trailing-tab artifacts; context-aware indent in validate

MultiLineError for ApiChange, SecurityChange, ComponentChange carried a
trailing "\t" on the header line (in security/component, a space-tab
where "at source" never went). Visible as stray whitespace. Removed.

validate's text formatter now mirrors the changelog shape when an
operation context exists ("in API METHOD /path" plus a two-tab message
body) and stays single-indent when it does not (doc-root findings,
components-rooted findings after dedup). Continuation lines (Schema:,
Value:, JSON blocks in nested-schema messages) inherit the same indent
so multi-line messages stay visually grouped under the operation line.

indentContinuation now takes the prefix as a parameter so the first
line and continuation lines always agree.

Tests updated to match the corrected strings.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore: bump kin-openapi pseudo-version

Pick up the latest HEAD of the upstream fork branch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore: track kin-openapi upstream master directly

Drop the oasdiff/kin-openapi fork replace. The fork's master is 109
commits behind upstream and all the typed-cluster work
(#1166, #1180, #1187) landed upstream directly via Pierre's merges,
so the fork is no longer carrying anything we need.

Bumps the kin pseudo-version to 8381bfc (the #1187 merge commit,
2026-05-21 14:53 UTC). Phase 1 of the validate subcommand can now
errors.As against the full typed-cluster surface (clusters from
#1166 + #1180 + #1187 + the 24 new clusters/wrappers from #1187).

Switch back to a tagged kin release version before merging this
branch to main, once Pierre cuts one containing all three PRs.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* formatters: update text-changelog golden after trailing-tab removal

Commit f04519e (2026-05-17) dropped the trailing "\t" artifact from
ComponentChange.MultiLineError (and the parallel ApiChange and
SecurityChange formatters), but didn't update the only golden test
that exercises this path. The branch has been failing
TestTextFormatter_RenderChangelog ever since.

Update the expected string to match the now-clean output —
`error\t[change_id]\n` instead of `error\t[change_id] \t\n`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* build: pin kin-openapi to released v0.139.0

#894 was wired against a master pseudo-version while the typed-validation
PRs were unreleased. Pierre tagged v0.139.0 (2026-05-23), which includes
them, so pin the real release. Build + full test suite green.

* docs: document the validate command

Add docs/VALIDATE.md (usage, output formats, flags, exit codes, rule
IDs) and list `validate` in the README Commands section (eight commands
now). Mirrors the existing per-command doc pattern.

* validate: document git-ref as a supported input; tidy help and comment

Note git refs (e.g. main:openapi.yaml) alongside file/URL/stdin in the
validate Long help and docs/VALIDATE.md, since the loader supports them.
Also tidies the Long help and the catchall-ID comment.

* validate: render output through the formatters package; add githubactions

Move validate's output off the inline yaml/json/text marshaling and onto the
shared formatters, the same path changelog/breaking/flatten use: a new
Formatter.RenderValidate method renders a plain formatters.Finding, looked up
by format via formatters.Lookup. text, yaml, and json implement it; the
command's -f options come from SupportedFormatsByContentType(OutputValidate)
so the advertised and implemented format sets stay in sync.

Add githubactions support so the upcoming oasdiff-action validate wrapper can
emit one CI annotation per finding (anchored to file/line/column) inline on
the pull request, plus error_count/warning_count/info_count step outputs.

Validate and changelog now share formatters.ComputeFingerprint so a downstream
tool can match findings across spec versions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* validate: classify finding severities and add --fail-on

Findings were all reported as errors. Classify them by impact via
severityForKinError (errors.As dispatch, default ERR so any unrecognised or
newly-typed kin cluster stays an error):

  WARN  version-portability (3.1 field in an older doc), $ref siblings that
        are silently ignored, conflicting paths, duplicate parameters, and a
        default value that violates its schema
  INFO  an example that violates its schema (the contract itself is valid)

duplicate-operation-id stays ERR: it violates the spec's uniqueness MUST and
breaks code generators.

Add -o, --fail-on (default ERR), mirroring changelog: the command exits 1
only when a finding is at or above the threshold, so warnings and info print
but don't fail CI by default. Severity flows through to githubactions
annotations (::warning / ::notice / ::error) and the text summary counts.

The mapping is a hardcoded dispatch; a future per-rule override (like
changelog's --severity-levels) can sit on top of it.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: escape GitHub Actions annotation messages and properties

The githubactions formatter only escaped newlines, so a message containing
'%' (decoded by GitHub) or a property value containing ':' or ',' (e.g. a
git-ref source path like main:openapi.yaml) would produce a malformed
annotation. Add escapeData (% / CR / LF) and escapeProperty (+ ':' / ',')
per GitHub's workflow-command rules, escaping '%' first so the introduced
%0A/%0D sequences are not double-escaped. Applied to both the changelog
(RenderChangelog) and validate (RenderValidate) output.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* docs(validate): add Feedback section

Invite users to report issues with a [validate] title prefix, mirroring the
OpenAPI 3.1 page. Pairs with the catchall spec-validation-error rule ID,
which we want users to report.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* validate: reject --color with a non-text format; clarify docs wording

--color only affects the text output (yaml/json/githubactions ignore it), so
pairing it with a non-text format now errors instead of being silently
dropped, matching the changelog command. Also soften the VALIDATE.md intro:
findings are spec-defined violations classified by severity, not "hard
violations only", now that some are warnings/info.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test: cover validate formatter rendering and more kin clusters

Codecov patch coverage was low because CI runs `go test ./...` without
-coverpkg, so the new formatters code (RenderValidate, Findings helpers) got
no credit from the internal tests that exercise it, and validate.go's typed
dispatch had several kin clusters with no fixture.

- formatters-package tests for RenderValidate (text/yaml/json), the
  not-implemented fallback, and the Findings helpers (GetLevelCount,
  HasLevelOrHigher, indentContinuation).
- fixtures + tests for previously-untested clusters: conflicting-paths,
  duplicate-parameter, path-must-start-with-slash, and the three invalid
  security-scheme clusters; plus a multi-cluster test over openapi-test3
  (EitherFieldRequired / ExactlyOneField via joinFieldsForRuleID, OAuth-flow
  required field).
- a checker test for the exported IsColorEnabled.

No production code change. validate.go coverage 82.8% -> 88.6%; the new
formatters validate rendering is now fully covered in-package.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>