The intermediate migration added lease_generation only for the next
migration to remove it. Fold the drops into 20260608T090000Z with IF
EXISTS so fresh installs and databases that already ran 20260607 both
migrate cleanly.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Replace the single-case switch in the agent-run handler with an if,
and add the blank lines wsl_v5 requires around statements in the
approval checkpoint, worker tests, and policy-set registration.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Drop the isType helper that merely discarded the value already
returned by errors.AsType and inline the suspend and interrupt
checks directly into the result-handling branch.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Drop the lease, heartbeat, and stale-recovery test coverage that no
longer applies, and add coverage for the new behaviour: graceful
suspend returning a run to PENDING, an approval interruption parking it
in AWAITING_APPROVAL, and SubmitApproval merging decisions and
requeuing the run. Exercise the submitAgentRunApproval mutation end to
end, including the authorization paths for the relocated policies.
Signed-off-by: Bryan Frimin <bryan@probo.com>
The worker leaned on a lease plus a heartbeat goroutine and a stale
recovery sweep to reclaim runs from crashed workers. That machinery
raced with long LLM and tool calls and conflated graceful stops with
failures. Remove the lease columns, heartbeat, and stale recovery, and
rely on FOR UPDATE SKIP LOCKED for single-claim plus explicit state
transitions: a graceful suspend returns the run to PENDING and a crash
now leaves it RUNNING for manual recovery.
Treat an approval interruption as a known stop that parks the run in
AWAITING_APPROVAL, and add SubmitApproval to merge human decisions into
the checkpoint and requeue the run to PENDING. The decisions must cover
exactly the pending approvals, since a missing one would resume as an
implicit denial. Expose this through the submitAgentRunApproval
mutation.
Signed-off-by: Bryan Frimin <bryan@probo.com>
The agent-run actions and policies lived in the core probo policy set,
which forced every authorization change for the agent-run domain to
touch unrelated core files. Move the actions and the OWNER/ADMIN and
VIEWER/AUDITOR policies into the agentrun package and have it expose a
PolicySet that probod registers into the authorizer at composition
time, so the rules live alongside the domain logic they govern.
Signed-off-by: Bryan Frimin <bryan@probo.com>
The nested-restore test only checked that an unresolvable inner agent
left its tool call in the checkpoint for a later retry. It did not
verify what happens to a sibling inner agent that does resolve and
complete during the same restore.
Add a resolvable done-agent alongside the missing inner-agent and
assert that its progress is persisted: its inner checkpoint is dropped
and its result recorded as a completed call, so a later retry replays
only the still-unresolved branch instead of re-running finished work.
Signed-off-by: Bryan Frimin <bryan@probo.com>
The single- and multi-level suspend/restore tests gated their slow
leaf tool on a manual release channel closed 50ms after cancel().
That sleep was a guess at how long the suspend signal takes to reach
the running sub-agent, so the post-tool turn boundary could observe
the release before cancellation and complete the run instead of
checkpointing, making the assertions timing-dependent.
Expose the per-run suspend signal through SuspendSignalFrom in an
export_test shim and have the leaf tools block on it directly. The
tool now returns only once the graceful-suspend signal has actually
propagated to its agent, so suspension is observed deterministically
without sleeps or release channels.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Each package that exercises the database against a real Postgres
carried its own copy of the connection bootstrap and schema setup.
Those copies had already drifted: some keyed off PROBO_TEST_PG_ADDR
with hardcoded defaults, others off PROBO_TEST_PG_URL, and the
agentrun/coredata suites hand-applied individual agent_runs
migrations to ensure the table existed.
Introduce a single test.PGClient helper that parses PROBO_TEST_PG_URL
(falling back to the local compose database), runs the full coredata
migration set once per process, and skips when no database is
reachable so make test stays a pure unit-test run. Migrate the
agentrun, coredata, cookiebanner, iam, and thirdparty suites onto it
and delete the duplicated helpers so the bootstrap can no longer
diverge.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Propagate graceful-suspend signals through detached run contexts and let
only opt-in suspendable tools re-attach cancellation, so AsTool sub-agents
can checkpoint and restore across nested trees while leaf tools keep
running detached.
Add focused agent and worker tests for single and multi-level suspend/
restore flows, plus heartbeat lease-loss and nested-restore error paths to
harden functional behavior under failure conditions.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Move agent-run orchestration from the legacy supervisor path into the new
agentrun worker/service package and wire it through coredata, server,
policies, and GraphQL resolvers.
This consolidates run lifecycle handling around lease-aware workers and
aligns API surface with the new agent-run domain model so reviewers can
follow one coherent execution path.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Auditors could not see the Context page in the console because
AuditorPolicy was missing core:organization-context:get. Grant the
read action (mirroring ViewerPolicy) so the sidebar item appears and
the context resolver succeeds for auditors.
Signed-off-by: Sacha Al Himdani <sacha@getprobo.com>
Replace the hand-rolled credential chain (static, ECS endpoint, EC2
role) with config.LoadDefaultConfig, which handles the full standard
AWS credential resolution chain automatically.
Co-authored-by: Neil McGibbon <code@neilmcgibbon.com>
Signed-off-by: Bryan Frimin <bryan@getprobo.com>
Remove schema objects that are no longer referenced by any query code,
GraphQL resolver, or frontend and only survive in old migration files:
- users_organizations: superseded by iam_memberships (data was migrated in
20251006T220024Z).
- organizations.logo_object_key: pre-files-table logo storage; replaced by
logo_file_id / horizontal_logo_file_id (data migrated in 20251009T140000Z).
- trust_center_accesses.has_accepted_non_disclosure_agreement,
has_accepted_non_disclosure_agreement_metadata, nda_file_id and
last_token_expires_at: the NDA-acceptance flow now uses
electronic_signature_id together with the state column.
Signed-off-by: Sacha Al Himdani <sacha@getprobo.com>
Instead of a toast, expired magic link tokens now return a TOKEN_EXPIRED
GraphQL error code and redirect users to a dedicated /magic-link-expired
page with a clear CTA to request a new link.
Signed-off-by: Bryan Frimin <bryan@getprobo.com>
Signed-off-by: Bryan Frimin <bryan@probo.com>
Using localhost as the EHLO identity is the Go net/smtp default but is
rejected by strict relays such as Google Workspace. os.Hostname() is
the conventional SMTP client fallback and returns the actual machine or
pod name in production environments.
Signed-off-by: Bryan Frimin <bryan@probo.com>
Google Workspace SMTP relay (smtp-relay.gmail.com) rejects generic
EHLO identifiers such as "localhost", which is Go net/smtp's default.
Operators can now set SMTP_HELLO_NAME to a valid hostname so that
c.Hello() is called before StartTLS, satisfying strict relay policies.
Fixes https://github.com/getprobo/probo/issues/1284
Signed-off-by: Bryan Frimin <bryan@probo.com>
Rejecting one audit report via Slack could look like a blanket denial
when HIPAA and SOC 2 reports shared a filename. Use framework and
audit name in rejection emails.
Signed-off-by: Sacha Al Himdani <sacha@getprobo.com>
BuildTokenURLForDomain and BuildTokenURLForSite cannot be merged: a
provider's signed-state Site (e.g. Datadog's region key) is not always the
string its token host needs (the API domain). Spell that out at the switch
so the split is not mistaken for accidental duplication.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Drop the zendeskRole helper and use the raw API role, matching the other
drivers. Factor the duplicated RFC3339 timestamp parsing (parseZendeskTime
and parseDatadogTime were identical) into a shared parseRFC3339Ptr in
driver.go.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
A Registration that sets both BuildTokenURLForDomain and
BuildTokenURLForSite would pass startup but silently use only the first
at the OAuth callback (CompleteWithState checks them in order). Reject
the combination in Register so the misconfiguration fails loud at
process start, matching the existing APIKeyBasicAuth/APIKeyHeader guard.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Zendesk is a multi-tenant OAuth connector keyed by the customer
subdomain. The customer enters it at connect time; it rides the signed
state to the callback, is re-validated, and is stored on the connector
settings to build the API host.
List staff (agents and admins) via GET /api/v2/users.json with cursor
pagination, mapping role, active/suspended, and 2FA status; end-users
are excluded. The subdomain is validated as a single DNS label at every
trust boundary to close the SSRF vector, and the data client keeps the
SSRF-protected transport.
Zendesk OAuth across customer subdomains requires a Zendesk-approved
global OAuth client; the connector goes live once those credentials are
supplied via bootstrap.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Subdomain-based OAuth providers (e.g. Zendesk) authenticate against a
per-customer host, <subdomain>.zendesk.com, for both the authorize and
the token endpoint. The subdomain is known at initiate but, unlike
Datadog's region, the provider does not echo it back on the callback,
so it must survive the round-trip another way.
Carry the chosen site on the HMAC-signed OAuth state (OAuth2State.Site)
and add a BuildTokenURLForSite closure, symmetric with the existing
BuildAuthURLForSite, that builds the token URL from it at callback time.
The state is signed, so a tampered site is rejected before use; the
closure still re-validates the format. Datadog's domain-echo path is
unchanged and a provider sets at most one of the two closures.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
OktaProvider was dead: the provider-string const is only matched by
OAuth providers (SlackProvider in oauth2.go / connector.go), and an
API-key connector never matches on it.
Wrap the url.Parse error in nextPageURL with %w, like every sibling
driver, instead of dropping it.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
The rel="next" link is taken from the provider response, so the host
check alone let a response downgrade the next request to http or move
it to another port on the same host. Reject a non-https scheme and any
explicit port in addition to a host mismatch, so the crawl stays on
the same https origin.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
The driver test cassette was regenerated by running the driver in
record mode against a live Okta org, which validated the field
mapping (id, status, created/activated/lastLogin, profile.*) and the
rel="self" Link header format against the real GET /api/v1/users
response.
The recording is then fully anonymized: the real domain, member
names, emails, user IDs, session cookie, and request-id are replaced
with synthetic values, and a second page with a rel="next" Link
header is added back to keep pagination coverage (the source org had
a single user). TestCassettesUseSyntheticEmails guards the result.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Register the Okta provider (SupportsAPIKey, APIKeyAuthScheme SSWS, a
required "domain" extra setting, and the driver/name-resolver
factories) and add it to the builtin registry.
The create-API-key resolver normalizes and validates oktaDomain into
OktaConnectorSettings, returning a static INVALID error that never
echoes operator input, and stamps the SSWS scheme onto the
connection. No picker, OAuth metadata, or probe URL: the token plus
domain identify exactly one org and the host is per-connection.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
The driver lists GET /api/v1/users (limit=200) on the customer's org
host and follows the RFC 5988 Link header, pinning pagination to the
configured host so a response cannot redirect the crawl off-tenant.
User status maps to the three-valued Active flag (SUSPENDED and
DEPROVISIONED are inactive); ExternalID is the stable Okta user id.
The name resolver reads /api/v1/org and returns ("", nil) on any
non-2xx so a read-only token lacking org-settings read does not loop
the source-name worker.
The org domain is operator-supplied and feeds the URL host, so it is
the one SSRF-sensitive input: NormalizeOktaDomain validates and
strips it on the write path and IsValidOktaDomain re-checks it at
driver construction, on top of the transport's SSRF protection.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Okta API tokens authenticate as "Authorization: SSWS <token>", a
scheme none of the existing API-key modes (Bearer, custom header,
Basic) can express. Add Registration.APIKeyAuthScheme, plumb it onto
APIKeyConnection.Scheme, and send it via a new schemeAuthTransport.
The three API-key presentations (BasicAuth, Header, Scheme) are
mutually exclusive; Register rejects setting more than one so a
misconfiguration fails at process start rather than silently.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Introduce the OKTA value to the connector_provider enum (Go const,
ConnectorProviders slice, IsValid switch, GraphQL @goEnum, and the
Postgres ALTER TYPE migration) and an OktaConnectorSettings struct
holding the customer's Okta org domain.
Okta is a per-tenant IdP with no central API gateway, so the
connector is keyed on the org domain rather than a shared host. The
oktaDomain field on CreateAPIKeyConnectorInput lets the API-key flow
carry it.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
GET /v1/users returns a bare JSON array, not a {"data":[...]} envelope
(total_count is a separate endpoint), so decode directly into a slice.
The previous envelope-first decode errored on the array and never
reached the fallback, so ListAccounts failed against the real API.
Treat deprovisioned users as inactive. Add the missing Clerk
third-party logo and its wiring, and document why the registration
sets no probe URL or name resolver.
Record the driver-test cassette against a live Clerk development
instance, scrubbed of PII and instance identifiers (emails to
example.com, image_url payloads and CF transport headers dropped).
A locked account verifies the inactive path.
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Align the Clerk driver tests with the rest of the access-review\ndriver suite by exercising VCR cassette replay instead of a\ncustom transport stub.\n\nThis keeps test coverage consistent with existing providers and\nensures request matching behavior is validated against a checked-in\nfixture.
Signed-off-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Bryan FRIMIN <bryan@frimin.fr>
Wire Clerk in as a supported connector provider for access\nreviews and expose it through the console GraphQL provider enum.\n\nAdd a dedicated Clerk driver that lists users from the Clerk\nBackend API, maps account state and authentication signals into\nAccountRecord fields, and covers the behavior with focused driver\nand provider tests.\n\nInclude a migration that appends CLERK to the connector_provider\nenum so environments can persist Clerk connectors safely.
Signed-off-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Bryan FRIMIN <bryan@frimin.fr>
Signed-off-by: Cursor Agent <cursoragent@cursor.com>
The teammate detail endpoint returns a bare object, not a {"result":...}
envelope, so the N+1 scope fetch silently discarded scopes and MFA was
always Unknown for every teammate. Decode the bare object.
Found by re-recording the driver cassette against the live API; the
hand-written cassette had hidden the mismatch. Also from that review:
- Map is_sso/is_partner_sso to AuthMethod (SSO vs Password) instead of
hardcoding Unknown
- Log the swallowed teammate-detail fetch error (wire the driver logger,
matching github.go)
- Add a restricted teammate + a missing-email row to the cassette so the
N+1 decode, the empty-email skip, and the SSO path are load-bearing
- Drop the stale Content-Length response header; document ExternalID
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
- Add SendGrid third-party logo and wire it into ThirdPartyLogo
- Add SendGrid name resolver (account company name, graceful fallback)
- Fix MFA detection: full-access teammates carry both 2fa_exempt and
2fa_required, so report Unknown unless exactly one is present
- Re-record the driver cassette against the live API
- Use a random time suffix for the migration filename
Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
Add a best-effort MFA status check to the SendGrid access-review
fetch pipeline by querying teammate details and inspecting 2FA scopes.
When teammate scopes include 2fa_required or 2fa_exempt, map those
to ENABLED or DISABLED MFA status values; otherwise keep UNKNOWN.
Extend the SendGrid cassette and tests to cover the detail lookups and
MFA scope mapping behavior.
Signed-off-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Bryan FRIMIN <bryan@frimin.fr>
Implement a SendGrid access-review driver that fetches teammates
from the SendGrid API and maps them into AccountRecord values.
Register SendGrid as a connector provider, expose it through the
connector provider enum, and add a migration that extends the
connector_provider type with SENDGRID.
Cover the new driver with a VCR-backed fixture test and helper
tests for role and response-shape handling to keep parsing robust.
Signed-off-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Bryan FRIMIN <bryan@frimin.fr>
