Two problems, one cause. Before 62131e7 the promotion step renamed the
Unreleased heading instead of consuming it, appending "(carried forward)" on
every build - one heading had accumulated the suffix seven times. Notes were
left stranded under 103 such headings while the version sections the release
workflow extracts held only their build line, so a release would have published
nothing. 62131e7 fixed the mechanism but never repaired the backlog, and
Unreleased was empty besides.
Consolidated the stranded notes under Unreleased and collapsed duplicate
version sections: 1441 lines to 570, with every one of the 144 bullets and 16
sub-headings still present, checked by diffing the sets before and after.
Described the seeding work of the last several commits, which was undocumented:
the enrollment protocols and where they are configured, the dynamic SCEP
challenge, auto-renew by default, and the four defects the first live run
exposed.
Verified the mechanism end to end - a build promotes the notes into its version
section, leaving Unreleased empty, and the workflow's extraction returns 245
lines rather than a bare build line. 340 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The line still interpolated $enrollmentType, which went away when profile
creation stopped varying by enrollment method. Every run since had adopted
existing profiles, so the create path never executed and never surfaced it.
A run against an emptied project failed there immediately.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The enrollment methods were not appearing in the UI because they were being set
in the wrong place. A profile's own enrollmentType is one value, but that is
only the base; what the UI lists under an application - "configure how this
application will issue certificates via API, EST, ACME, or SCEP" - is a
separate configuration on the application-to-profile link, and those are
independent. A single profile can answer all four at once.
That replaces the previous approach. The mecm-...-scep and ec-...-acme profiles
are gone; there are three profiles again, one per policy, and each declares an
Enrollment block naming the protocols it should answer. Applied with PUT to
/applications/{id}/profiles/{id}/enrollment/{api,scep,acme}, which is idempotent
by nature, so a re-run restates the same settings. The routes take PUT, not
POST, and a POST there returns 404.
SCEP now uses a dynamic challenge, so there is no shared secret to distribute
or rotate; each request collects a one-time password from the challenge
endpoint. -ScepChallengePassword is kept for anyone who sets ChallengeType to
static, and is otherwise unused.
Enabling a protocol is what makes Infisical mint its endpoint - a SCEP URL,
challenge URL and RA certificate, or an ACME directory URL - so the run reads
them back and prints them, and returns them under Enrollment. None of it can be
derived from the configuration alone.
Verified against a live instance: scepConfigured and acmeConfigured come back
true on the intended profiles, the dynamic challenge settings round-trip, and a
re-run restates without creating anything.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A profile carries exactly one enrollment method - Infisical rejects a mixed one
with "API enrollment type cannot have EST, ACME, or SCEP configuration" - so
reaching a policy over more than one method means a profile per method, sharing
the CA and policy. Added mecm-server-client-auth-scep and
ec-server-client-auth-acme next to the three API profiles.
SCEP for the RSA policy because that is how domain-joined Windows and mobile
clients enrol without a token, which is the NDES role SCCM/MECM would otherwise
need; RSA because SCEP client support for EC keys is patchy. ACME for the EC
server/client policy, for workloads that already speak it. Code signing stays
API only: SCEP and ACME prove control of a device or a DNS name, and neither
says anything about who may sign code.
A static SCEP profile needs a challenge password of at least eight characters.
Added -ScepChallengePassword, and when it is omitted the run generates one and
prints it once at the end, because Infisical will not hand it back afterwards.
ACME sets skipDnsOwnershipVerification. These are internal names from an
internal CA, and the DNS-01 challenge proves control of a public zone, so
leaving it enforced makes the profile unusable rather than safer. The comment
says so and says when to turn it off.
Auto-renew is now on for every API profile including code signing, at fourteen
days. It is an api-only setting; neither SCEP nor ACME has an equivalent,
because both have the client drive renewal itself.
Verified against a live instance: profiles create with their config ids
populated, a re-run creates nothing and does not reissue the challenge.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ran the script against a live Infisical instance for the first time. It failed
at four separate points, each fixed here.
Responses are read through Get-ApiProperty/Get-ApiCollection/Get-ApiObject
rather than directly. Set-StrictMode makes an absent property terminating, and
the cert-manager routes return bare payloads - a naked array for a list, the
resource itself for a create - where the secrets and pki routes wrap theirs in
a named property. All eighteen reads now handle either shape. The
application-to-profile listing is a join row carrying profileId, not a profile
object with an id, so it is read accordingly.
Renamed the local $configuration inside the CA loop to $caConfiguration.
PowerShell variable names are case-insensitive, so from the second CA onward it
shadowed the script-level $Configuration block and every lookup into it failed.
Renamed the $profile loop variables for the same reason: $profile is an
automatic variable.
Policies now carry subject, sans, and signature algorithms, and profiles carry
defaults. The previous comment claimed omitting subject and sans left them
unconstrained; the opposite is true. Infisical refuses any attribute with no
policy entry ("no subject policies defined"), refuses a policy that defines no
signature algorithm, and refuses a request missing a usage the policy marks
required - Request-InfisicalCertificate sends none of its own, so the profile
default has to supply it. The enumeration list in the header dropped
domain_component, upn, and any_purpose, which the API does not accept.
Verified end to end: a clean run builds both CA hierarchies including the
subordinate CSR-sign-import sequence, and a re-run creates nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Seeding a real instance failed with "Route GET:/api/v2/organizations not found".
The route genuinely does not exist: /api/v2/organizations mounts only
/:organizationId/* sub-routes, and the listing lives on the singular
GET /api/v1/organization, which returns the same { organizations: [...] } shape.
The module had the same defect, so Get-InfisicalOrganization would have 404d on
any current server. ListOrganizations now prefers /api/v1/organization with the
v2 route retained as a fallback candidate, and the organization client invokes
with candidate fallback so the fallback can actually be reached. The endpoint
shape test expected the broken template and now expects the working one.
The script reported this as "The supplied -AccessToken was rejected ... it may
have expired", which sent the reader after a token that was seconds old and
perfectly valid. A 404 or a connection failure says nothing about the token, so
only a 401 or 403 now produces that message; anything else reports that the
instance could not be reached and states plainly that it is not a token problem.
The token probe also moves to /api/v1/projects, which is the next call the
script makes anyway, so a successful probe proves both the token and a route the
run actually depends on rather than one chosen only for validation.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The seeding script was never in the repository. .gitignore carries "scripts/"
for local helper scripts, core.ignorecase is true on Windows, so Scripts/
matched it and git add -A silently skipped the file. The commit that claimed to
add the script contained only its CHANGELOG and README entries; the script
itself existed on disk and nowhere else. It now lives in Tools/, which is not
ignored, and the ignore rule says why so this cannot swallow shipped tooling
again.
The script also gains an -AccessToken parameter set alongside
-ClientId/-ClientSecret, so a throwaway instance can be seeded with the bearer
token from a signed-in browser session rather than by first creating a machine
identity. The two are mutually exclusive parameter sets, so supplying both is
rejected at binding rather than silently preferring one.
A supplied token is proven with a single read before anything is created. An
expired browser token is the common case, and failing there is much better than
failing midway through building a CA hierarchy and leaving a half-seeded
project behind; the error says as much and suggests copying a fresh one.
Renaming the internal token variable was not cosmetic. The script held the
bearer token in $Script:AccessToken, the same name as the new parameter, so
PowerShell applied the parameter's [ValidateNotNull()] to every assignment and
both authentication paths failed with "the value $null is not a valid value for
the AccessToken variable" before reaching the network. It is now
$Script:BearerToken.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The release body was built solely from the CHANGELOG section, which is a
terse changelog entry rather than the account of what changed. Gitea's
pull_request webhook payload carries the description - PullRequest.Body is a
documented field on the API struct - so it is now passed to the release step as
PR_BODY and leads the release when present.
The CHANGELOG entry is kept behind a fold rather than dropped, so a release
records both the narrative and the versioned entry. With no description the
body falls back to the CHANGELOG exactly as before, and with neither it says so.
Co-author and generation trailers are stripped: they belong on the commit, not
on a published release page. The description supplies its own headings, so no
"## Changes" wrapper is added around it; the fallback branches emit one because
a bare changelog fragment needs it.
Verified by lifting the run: block straight out of the workflow and executing
it, so the test exercises the shipped code rather than a copy: a description
plus changelog produces the folded form, description-only omits the fold,
changelog-only falls back, neither produces the placeholder, and the trailers
are removed while the real content survives. Both workflows still parse as YAML
with the same three jobs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The module was read-only for PKI configuration, which is why standing up an
environment meant raw REST rather than cmdlets. Adds 15: New-, Set-, and Remove-
for certificate authorities, policies, profiles, applications, and PKI
subscribers.
New-InfisicalCertificateAuthority -Type Intermediate returns an authority that
can actually issue. Infisical creates a subordinate with status
pending-certificate and never invokes generateIntermediateCaCertificate from the
create path, so the cmdlet performs the remaining sequence: read the certificate
signing request, sign it with -ParentCaId, import the signed certificate and
chain back.
Policy and profile bodies are deeply nested, so the constraint objects and
per-enrollment-type config blocks are taken as dictionaries rather than as
dozens of parameters, matching how -Subject and -Metadata already work. Two
conversion details matter and are now pinned by tests:
- PowerShell callers capitalise hashtable keys, and @{ Required = ... } was
reaching the API as "Required", which its schema does not recognise. The
constraint vocabulary is emitted lower-cased while every other key keeps its
camelCase, since those are API field names supplied verbatim and lowercasing
ttlDays or isCA would silently drop them.
- An empty collection is omitted rather than sent. "allowed": [] reads to
Infisical as "allow nothing", never what @{ Allowed = @() } was meant to
express, and an entry carrying only "type" is rejected outright.
-EnrollmentConfig routes to the block matching -EnrollmentType so EST, ACME, and
SCEP settings arrive through one parameter. -AutoRenew is sent only when bound,
because a switch is false when absent and sending it unconditionally would
disable renewal on an update that never mentioned it.
-ProjectId is optional on all 15 and resolves as it does elsewhere. Every one
supports -WhatIf; Remove- cmdlets default to high confirm impact.
Also fixes Update-Changelog, which inserted the version heading above the notes
so the section the release workflow extracts held only the build line while the
notes stayed under Unreleased - every release published an empty changelog.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Update-Changelog inserted the version heading above the notes, so the section
the release workflow extracts contained only the build line while the actual
notes stayed under "Unreleased". Every release published an empty changelog. It
now promotes whatever sits under Unreleased into the version section and leaves
a fresh empty Unreleased above it, with the provenance line italicised and last
so it does not read as another entry in whichever section the notes ended on.
Re-running the same version remains a no-op, and a build with nothing to say
still produces a section.
SignCertificateByCa now prefers /api/v1/cert-manager over the older /api/v1/pki
route.
RetrieveCertificate and GetCertificateBundle deliberately keep the older route
first, which is the opposite of what it looks like. The /api/v1/pki route
resolves a certificate by serial number, which is what callers supply; the
cert-manager route takes a certificate ID and passes it through as
getCert({ id }), so it cannot answer a serial. Both remain registered so either
identifier resolves - which is what lets the metadata path look a certificate up
by ID - and a comment now records why reordering them would break serial
lookups.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The script created a Certificate Manager project whenever the configured slug
did not match an existing one. Against an organization that already had one
that produced a second, which would then be unable to hold applications at all -
they are served only from the organization's active project. It now adopts the
existing project and says so, creating one only when the organization has none,
and notes when an organization has several that applications work in one.
It also seeds a certificate application and attaches the seeded profiles to it.
Applications are how the console groups profiles and how
Get-InfisicalCertificateProfile -ApplicationId scopes a lookup, so seeding
profiles without one left an environment the documented workflow could not
navigate. Profiles are attached at creation via profileIds, and an application
that already exists has only its missing profiles added.
Project discovery moves to /api/v1/projects, matching the module, and tolerates
either response shape.
Separately, the environment-variable discovery table listed ProjectId,
Environment, and SecretPath as discoverable Connect-Infisical parameters.
InfisicalEnvironmentResolver defines patterns for BaseUri, OrganizationId,
ClientId, ClientSecret, AccessToken, and ApiVersion only - the other three are
per-call parameters and were never resolved from the environment. Same defect as
the Quick Start, in the table immediately below it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Several examples still passed -ProjectId ($Project.Id) from a project lookup the
walkthrough no longer performs, so they read as though the parameter were still
required. Removed from the profile, policy, subscriber, and certificate
authority examples, and from the offline-fallback snippet.
The end-to-end example loses the site-specific application name, and the profile
selection collapses to one line. Its "Example output" section still showed the
project and certificate authority records the example stopped fetching, so it now
matches what the script actually emits, including the project-resolution line.
Adds a renewal example directly beneath, reusing the parameters built above so
the difference is visible as two added keys rather than a second wall of setup.
It notes that the call is safe to repeat, that reuse is matched on the issuing
profile and the SAN set so a changed profile or added name reissues, and that
-Force ignores both.
Also drops the note about $CertificateProfile versus the $Profile automatic
variable, which explained a naming choice the reader has no reason to care
about, and points the extension docs at ResolveCertManagerProjectId rather than
the ResolveProjectId name that no longer exists.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An organization that has not set up Certificate Manager has nothing to list, so
resolution now returns nothing instead of throwing, and the PKI Get-* cmdlets
return quietly. -Verbose explains what happened and what to do:
-ProjectId was not supplied and this organization has no Certificate Manager
project, so there is nothing to resolve to. Create one in Infisical
(Certificate Management), or pass -ProjectId to target a specific project.
Without the guard the null would have travelled to the client layer and
surfaced as "ProjectId is required" from somewhere unrelated to the cause, so
each Get-* cmdlet returns as soon as resolution comes back empty.
Request-InfisicalCertificate deliberately proceeds, because profile issuance
does not need a project: only the reuse search does, and that already degrades
to a local match with a warning when the search cannot run.
Resolution is now attempted once per cmdlet instance rather than once per
non-null result, so an organization with no project does not re-query on every
call.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Several Certificate Manager projects in one organization is a normal
configuration, not an ambiguity to reject. Infisical designates one as the
organization's active project and serves certificate applications only from it:
if (req.internalCertManagerProjectId !== activeProjectId) {
throw new BadRequestError({ message: "Applications are only available on
this organization's active Certificate Manager project." });
}
So an application-centric workflow is single-project by design, and resolving to
the active project is what makes it work. Resolution now picks that project when
several exist, falling back to the first - saying so on the verbose stream -
when the organization designates none. Only an organization with no Certificate
Manager project at all still errors, because there is genuinely nothing to
resolve to.
Adds InfisicalOrganization.DefaultCertManagerProjectId, which is what the
organization record calls its active project, so the choice is read rather than
guessed.
Moves Get-InfisicalProject onto /api/v1/projects. /api/v1/workspace mounts
Infisical's deprecated project router; it is retained as a fallback candidate so
older servers keep working, and the endpoint shape test now expects the current
route.
The end-to-end README example drops to the four calls that actually do the work:
find the application, pick its profile, gather SANs, request. The project lookup
is gone because -ProjectId resolves itself, and the CA lookup is gone because the
profile already binds its issuing CA and -InstallChain installs the chain
regardless.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Infisical console never asks which Certificate Manager project to use. The
project is in the URL, but it is chosen implicitly - the service resolver takes
the single cert-manager project when an organization has exactly one - and
everything below it is presented as applications. Requiring -ProjectId on every
PKI cmdlet made the module stricter than the service it wraps.
-ProjectId is now optional on Get-InfisicalCertificateApplication,
-ApplicationEnrollment, -Authority, -Certificate, -CertificatePolicy,
-CertificateProfile, Get-InfisicalPkiSubscriber, and
Request-InfisicalCertificate. Omitting it applies the same rule the server
applies and reports the resolved project on the verbose stream; an organization
with several produces an error naming the candidates rather than guessing.
Resolution is client-side rather than deferred to the server because several PKI
endpoints carry the project in the URL path - /api/v1/projects/{projectId}/
pki-subscribers and .../certificates/search - and cannot fall back to the
server's resolver at all. Doing it in one place keeps path-scoped and
query-scoped endpoints behaving identically.
Get-InfisicalCertificate -SerialNumber no longer resolves a project, since
addressing a certificate by serial does not need one.
Adds -ApplicationId and -CaId to Get-InfisicalCertificateProfile, which the
profiles endpoint already supports as query filters, so a listing can be scoped
the way the console groups profiles. Get-InfisicalCertificate already accepted
-ApplicationId. The existing five-argument ListCertificateProfiles overload is
retained.
Also fixes the README Quick Start, which did not run: it passed -ProjectId and
-Environment to Connect-Infisical, which has neither, failing with "A parameter
cannot be found that matches parameter name 'ProjectId'". Project, environment,
and secret path are per-call parameters. Documents that a project contains
applications, since listing projects returning one entry while the console shows
several applications is the confusing part.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Takes any IDictionary, so a hashtable, an [Ordered] dictionary, or a generic
Dictionary[String,String] all bind. Verified against all three issuance
parameter sets under Windows PowerShell 5.1.
The reconcile is client-side by necessity. Infisical's PATCH /certificates/{id}
replaces a certificate's metadata wholesale - certificate-v3-service deletes
every resource_metadata row for the certificate before inserting what it was
sent - so sending just the caller's keys would silently discard everything else
attached to it. The module reads the current set, merges the supplied keys over
it, and writes back the union. Keys it was not asked about survive, and when
nothing would change no request is sent at all.
Reconciliation runs on the reuse path too, so changing metadata does not force
a reissuance to take effect.
Values are flattened to strings because the API accepts only strings: 443
becomes "443", $True becomes "True", $Null becomes "". Keys are trimmed and
compared case-insensitively, matching how PowerShell callers supply them, and
blank keys are dropped since the API rejects them.
Metadata never fails an issuance that otherwise succeeded. By the time it is
applied the certificate exists and may already be installed in the store, so a
failure is reported as a warning and the certificate is still emitted.
Adds InfisicalCertificate.Metadata and InfisicalCertificateResult.Metadata as
case-insensitive dictionaries, the PATCH endpoint under both the cert-manager
and pki route namespaces, and models the metadata array already returned by the
certificate response.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Stands up a complete Certificate Manager environment from one declarative
configuration block, taking only a base URI, client id, and client secret.
Standalone: the module consumes the result but is not needed to produce it,
since InfisicalConnection deliberately does not expose its access token.
Seeds an RSA hierarchy carrying server and client authentication for SCCM/MECM,
and an ECDSA P-384 hierarchy carrying server, client, and code signing. Objects
are cross-referenced by name in the configuration and resolved to ids at run
time, so adding a policy or profile means adding an entry rather than editing
code.
Creating a subordinate CA needed more than one call. Infisical only self-signs
on creation for a root, and only when given an expiry; a subordinate is created
with status pending-certificate and generateIntermediateCaCertificate is never
invoked by the create path. The script performs the sequence itself: create,
GET the CSR, sign it with the parent, then import the certificate and chain
back onto the subordinate.
Idempotent by lookup on each natural key, so a re-run reports what exists and
creates only what is missing, and -WhatIf shows the whole plan without
contacting anything beyond authentication.
Also corrects a documentation error this research surfaced. The README claimed a
newly created CA defaults to direct issuance enabled, on the strength of the
database column default. The creation service passes enableDirectIssuance:false
explicitly, so every CA created through the API or UI has it disabled and
nothing can enable it afterwards - which makes certificate profiles the only
workable issuance path, not merely the recommended one.
Validated by parsing the script, exercising the policy payload builders against
the shapes Infisical's zod schemas accept, confirming every enum literal matches
a defined Infisical value, checking that all name cross-references resolve, and
running -WhatIf to the point of authentication. Not yet exercised against a live
Infisical instance.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Extending -DnsName or -IpAddress and re-running returned the existing
certificate, which lacked the name that had just been added - the reuse check
compared only the common name. A candidate must now carry every requested name,
and the name that disqualified it is reported so the reissue is explainable.
Reading SANs back off an installed certificate needs a decoder: netstandard2.0
has no X509SubjectAlternativeNameExtension, and X509Extension.Format produces
localized text that cannot be compared. The extension is decoded from its DER
bytes with BouncyCastle, already carried for CSR generation.
The rule is coverage rather than equality, since a certificate carrying more
names than requested still satisfies the request. DNS names compare
case-insensitively and IP addresses are normalized through IPAddress, so ::1 and
0:0:0:0:0:0:0:1 are the same name. Trimming the SAN set therefore still reuses;
-Force covers that case.
FindMatch keeps its original four-argument overload so existing callers are
unaffected, and only reports a rejected name when no candidate qualified.
Verified against a live CurrentUser\My store with a certificate carrying
SANPROBE, SANPROBE.contoso.com and 10.20.30.40: identical and subset requests
reuse, a new DNS name or IP forces reissue naming the missing entry, differing
case reuses, and an empty request reuses.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Switching certificate profiles returned the existing certificate instead of
issuing a new one. The reuse check searched Infisical by common name and status
only, so a host already holding a server-authentication certificate for its own
name was handed that certificate back when asking a client-authentication
profile - same subject, wrong extended key usages.
The search is now scoped by -CertificateProfileId or -CertificateAuthorityId.
Both filters already existed on InfisicalCertificateSearchQuery and serialize as
profileIds/caIds; the reuse path simply never set them. The subscriber path
needs no filter because a subscriber pins its own common name, so matching the
name is already equivalent to matching the subscriber.
A second defect compounded it: InfisicalLocalCertificateLookup.FindMatch only
applies its serial filter when the candidate set is non-empty, so a search that
legitimately returned nothing degraded into a name-only local match - exactly
the case that hands back another issuer's certificate. A completed search that
finds nothing is now a definite "nothing to reuse".
The lenient fallback is kept for the case where Infisical cannot be reached,
since failing a renewal because the API is down is worse, but it now announces
itself as a warning rather than being silent.
Reuse still does not compare subject alternative names; that gap is documented
with -Force as the workaround rather than half-addressed here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The publish job failed on PR #19 with a 409 for version 2026.7.30.2309, but
that version is live on the gallery (created 23:10:04) - the push landed and
the client still surfaced an error, so the retry collided with the upload that
had just succeeded. The run went red over a publish that actually worked.
The step now checks the gallery before pushing and skips when the version is
already there, and on a publish error it re-checks before failing. This mirrors
the release job, which already looks for an existing tag and skips.
Version comparison normalizes each segment the way NuGet does, since the
manifest carries zero-padded segments (2026.07.30.2309) while the gallery lists
the stripped form (2026.7.30.2309); comparing the raw strings would never match
and the guard would never fire.
Verified: both workflows still parse as YAML with the same three jobs, the
normalizer reproduces the gallery form for four published versions, and a live
lookup confirms the guard would have exited 0 on the run that failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Chain members are now installed before the leaf, so the certificate is
chainable the moment it appears in the store rather than momentarily orphaned.
After -InstallChain the chain is validated against the machine's own stores.
An incomplete result is reported as a warning naming the certificate whose
issuer is missing, which is the exact condition Windows surfaces as "The issuer
of this certificate could not be found" - previously that was only discoverable
in certmgr after the fact.
Chain routing is unchanged and already handles arbitrary depth: a self-signed
certificate is a root and goes to the trusted-root store, anything with an
issuer above it is a subordinate CA and goes to the intermediate store. Only
the leaf honours -StoreName (default My). This is now stated in the docs,
because the split was not obvious.
The installed certificate's Windows friendly name defaults to the common name
in upper case, which is what operators look for in certmgr. -FriendlyName
overrides it and moves from the ByCa parameter set to all of them; the CA path
still forwards the same value to Infisical as the issued certificate's
friendlyName.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Request-InfisicalCertificate -InstallChain could hang indefinitely. Adding a
root certificate to CurrentUser\Root makes Windows raise a modal trust
confirmation dialog, and X509Store.Add blocks until it is answered. When that
dialog was hidden or the session non-interactive (scheduled task, MECM task
sequence) the cmdlet appeared to stop right after installing the intermediate,
with no indication why. A warning is now emitted before the blocking call.
-StoreLocation now defaults to the process elevation when the caller does not
supply it: LocalMachine when elevated, CurrentUser otherwise. This is what most
callers want, and it sidesteps the trust prompt entirely because writing
LocalMachine\Root already required elevation. Applied to both
Request-InfisicalCertificate and Install-InfisicalCertificate; the resolved
value is reported on the verbose stream and an explicit -StoreLocation wins.
Chain routing is unchanged and already correct: self-signed certificates go to
the Root store and everything else to CertificateAuthority, within whichever
location was resolved.
When the resolved location is LocalMachine and -KeyStorageFlags was not
supplied, the private key is written to the machine key store. Without this the
key lands in the calling user's profile while the certificate sits in
LocalMachine\My, which is the usual cause of an installed certificate that
reports no usable private key to a service.
Reuse detection now searches the store location the install will write to
rather than always searching CurrentUser, so -AllowRenewal and the existing
certificate short-circuit behave consistently with where certificates land.
Elevation detection moved to InfisicalCmdletBase (evaluated through the engine,
since the module targets netstandard2.0 and carries no
System.Security.Principal.Windows reference) and is shared with
Write-InfisicalScepMdmProfileToWmi, which loses its private copy.
README gains the fuller worked example, a genericized output transcript, and a
"Where certificates get installed" section; cmdlet help updated to match.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The subscriber guidance shipped in #18 was wrong for fleet enrollment.
signSubscriberCert rejects any CSR whose CN differs from the subscriber's
commonName, and allowlists the subscriber's subjectAlternativeNames, so a
subscriber is a single named identity rather than a template. Enrolling N
machines through subscribers would require N subscribers.
Certificate profiles are the correct path: they accept a per-request common
name constrained by policy allowed/required/denied lists, and profile issuance
is the only path that skips the CA direct-issuance gate
(!isFromProfile && !ca.enableDirectIssuance && !certificateTemplate), so a
profile issues against a CA whose EnableDirectIssuance is False.
Also documents that enableDirectIssuance cannot be changed after CA creation:
it appears in no Infisical create or update schema (the generic CA schemas
accept only name and status). Migration 20250521110635_add-external-ca-pki.ts
renamed requireTemplateForIssuance to enableDirectIssuance and inverted every
existing value, so CAs that previously required a template now read False
permanently. The remedies are a profile, or a new CA (column defaults to true).
README end-to-end example switched from subscriber to profile issuance, and the
issuance-path table now leads with whether the common name varies per request.
Cmdlet help for Request-InfisicalCertificate and Get-InfisicalCertificateAuthority
updated to match.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
BREAKING: operation failures are now non-terminating errors, so -ErrorAction
(and $ErrorActionPreference) decides the outcome. try/catch around these cmdlets
now requires -ErrorAction Stop or $ErrorActionPreference = 'Stop'. A failing
pipeline item no longer aborts the batch.
Cmdlets no longer report "The pipeline has been stopped." as an error.
Select-Object -First, and Where-Object feeding it, stop the upstream cmdlet by
design; the shared error path in InfisicalCmdletBase now lets pipeline-control
exceptions propagate untouched instead of logging them and raising an error.
Error-level diagnostics moved off the warning stream to verbose. Every
Logger.Error call site logs and then throws, so the failure already reaches the
caller as an ErrorRecord; emitting it again as eight warning lines put failures
under -WarningAction instead of -ErrorAction. One error per failure now.
Request-InfisicalCertificate:
- -CommonName accepts the RDN form (CN=WEB01) and reduces it to the bare value,
which previously produced a CN=CN=WEB01 subject plus a bogus DNS SAN.
- -DnsName routes IP literals to iPAddress SAN entries, so the mixed output of
Get-InfisicalSANList can be splatted in as documented.
- The CA path sends the normalized common name to the signing endpoint.
- The issuance path is resolved and reported before a keypair is generated, and
a CA with direct issuance disabled fails fast with guidance naming
-PkiSubscriberSlug and -CertificateProfileId. Infisical exposes no
template-based issuance route, so no -CertificateTemplateId is added.
Get-InfisicalCertificateAuthority table output gains a DirectIssue column
(EnableDirectIssuance) so CAs eligible for -CertificateAuthorityId are visible.
README, about_PSInfisicalAPI, and cmdlet help document the stream/-ErrorAction
contract, subscriber discovery, and direct-issuance setup. Adds regression tests
for pipeline-stop propagation, logger stream routing, SAN splitting, common-name
normalization, and the non-terminating convention across all cmdlets.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
InfisicalSecretsClient.List now merges dto.Imports[].Secrets with dto.Secrets using local-wins precedence, restoring imported secrets that were previously dropped when -IncludeImports was set.
Get-InfisicalSecret, ConvertTo-InfisicalSecretDictionary, Export-InfisicalSecrets, Import-InfisicalSecret, New/Update/Remove-InfisicalSecret, and Start-InfisicalProcess now emit Information-level counts for retrieved/processed/injected items.
All collection-returning Get-* cmdlets (Folder, Project, Organization, Environment, Tag, SubOrganization, Certificate, CertificateAuthority, CertificateApplication, CertificatePolicy, CertificateProfile, PkiSubscriber) now log returned counts.
Get-InfisicalEnvironmentVariable gains an optional -Scope (EnvironmentVariableTarget) parameter plus per-scope verbose tracing and Information-level found/not-found outcome lines.
Mirrors the Gitea workflow with GitHub-specific adaptations: ubuntu-latest runner, actions/upload-artifact and actions/download-artifact v4, Bearer auth with X-GitHub-Api-Version header, /pull/ URL path, upload_url URI template handling on uploads.github.com, contents:write permission on the release job, and on-demand Install-Module of Microsoft.PowerShell.PSResourceGet for CurrentUser.
Renames -Prefix to -SecretsPrefix and -ForcePrefix to -ForceSecretsPrefix across ConvertTo-InfisicalSecretDictionary, Import-InfisicalSecret, Export-InfisicalSecrets, and Start-InfisicalProcess. Start-InfisicalProcess also renames the pipeline parameter -Secret to -Secrets. The previous names remain available as parameter aliases (Prefix, ForcePrefix, Secret) for backward compatibility. Internal InfisicalProcessOptions properties renamed to match.
Auto-generated by build.ps1 -CommitArtifacts. Build 2026.06.07.1435. Module DLL and manifest embed BuildCommitHash=97193d46f2ff, matching the source commit they were produced from.
Adds 8 cmdlets for Organization and Sub-Organization CRUD (Get/New/Update/Remove for each), targeting /api/v2/organizations and /api/v1/sub-organizations. Get cmdlets default to List parameter set and switch to Single when -OrganizationId or -SubOrganizationId is supplied. New/Update/Remove honor -WhatIf/-Confirm; Remove defaults to High ConfirmImpact and supports -PassThru. No project context required.
Adds Get-InfisicalSANList: emits a deduplicated SAN candidate set containing the local device name, the device name suffixed with each non-empty DNS suffix found across operational adapters and the system primary domain, every IPv4 unicast address falling within RFC 1918 or CGNAT, and the IPv4/IPv6 loopback addresses. Supports optional case-insensitive -InclusionExpression and -ExclusionExpression regex filters applied in fetch -> include -> exclude -> output order. Output is a single strongly-typed System.String[] array emitted non-enumerated so List<string>.AddRange consumes it directly.
Registers 10 new endpoints, adds InfisicalOrganization/InfisicalSubOrganization models with DTOs, mappers, and clients, full MAML help for all 9 new cmdlets, mapper unit tests, EndpointRegistry inline-data coverage, and docs/DesignSpec.md sections 16.7 and 16.8. build.ps1 CmdletsToExport and Test-ModuleImports expected list now contain 51 cmdlets. README updated with Organization/Sub-Organization tables, the new Get-InfisicalSANList entry, and an end-to-end certificate request example using splatted OrderedDictionary blocks.
Auto-generated by build.ps1 -CommitArtifacts. Build 2026.06.06.2229. Module DLL and manifest embed BuildCommitHash=207e7429e448, matching the source commit they were produced from.
- New cmdlet Start-InfisicalProcess: launches a child process with InfisicalSecret
objects decrypted directly into ProcessStartInfo.Environment (optional -Prefix),
additional -EnvironmentVariables, stdout/stderr capture, -AcceptableExitCodeList,
-ParsingExpression regex parsing, -ExecutionTimeout / -ExecutionTimeoutInterval,
-NoWait, -WindowStyle / -CreateNoWindow parameter sets, -Priority,
-StandardInputObjectList, -SecureArgumentList, -LogOutput, -ContinueOnError, and
ShouldProcess support. Secret plaintext is never written to user or machine scope.
- Stream capture uses event-based OutputDataReceived/ErrorDataReceived with
BeginOutputReadLine/BeginErrorReadLine (no Task / ReadToEndAsync /
GetAwaiter().GetResult()) to avoid PowerShell SynchronizationContext deadlocks.
- Restored the do { log; sleep } while (!HasExited) polling pattern using
Thread.Sleep(pollInterval) so verbose "has been running for X" / "Checking again
in Y" messages fire at the configured cadence even when no -ExecutionTimeout is
supplied.
- TimeSpan values in verbose logs and on the result now use a friendly format
("7 seconds, and 364 milliseconds", "1 minute, and 30 seconds", "N/A" when zero)
matching the legacy Start-ProcessWithOutput GetTimeSpanMessage scriptblock.
- Added DurationFriendly property to InfisicalProcessResult and a "The command
execution took X" verbose line at completion.
- build.ps1 CmdletsToExport and Test-ModuleImports expected list contain 42 cmdlets.
- Added 9 xUnit tests covering FormatFriendly singular/plural, multi-unit joining,
zero, sub-millisecond, and skip-zero-components behavior.
Auto-generated by build.ps1 -CommitArtifacts. Build 2026.06.06.2138. Module DLL and manifest embed BuildCommitHash=318db7048017, matching the source commit they were produced from.
Adds an optional [string] -Prefix parameter that prepends the supplied
string to every emitted variable name, regardless of -Format
(Json/Yaml/Xml/Env/EnvironmentVariables). When omitted or empty the
exporter buffer is forwarded unchanged (no-op).
Implementation clones each InfisicalSecret with SecretName = Prefix +
SecretName so the caller's pipeline objects are never mutated; the
SecureString and Tags/SecretMetadata array references are shared
(read-only usage downstream).
Also updates the cmdlet help XML description + adds a -Prefix example,
and reflects the new parameter in docs/DesignSpec.md.
gsadmin
GraceSolutions