Skip to content

feat(#7481): allow OpenAI-only repositories without GCP configuration - #7718

Merged
waynesun09 merged 30 commits into
fullsend-ai:mainfrom
amastbau:aisdlc-127/openai-no-gcp-replacement
Sep 30, 2026
Merged

waynesun09 merged 30 commits into
fullsend-ai:mainfrom
amastbau:aisdlc-127/openai-no-gcp-replacement

Conversation

@amastbau

@amastbau amastbau commented Sep 25, 2026 •

Copy link
Copy Markdown
Contributor

Summary

AISDLC-127 / #7481: repositories can now install without Vertex credentials, and each agent run selects inference from that agent's effective runtime and model. Agents in one repository may use different providers.

  • github setup and repos install accept no GCP inference inputs and write no GCP credential secrets. When GCP inputs are provided, the existing Vertex setup remains. repos status accepts both secrets absent but reports a partial pair as drift.
  • One fullsend run handles the selected agent. On GitHub Actions, Google credentials are chosen before harness validation and the pre-script:
    • Both GCP inputs set: the run prepares WIF credentials. The GitHub OIDC token uses the https://iam.googleapis.com/<provider> audience that provisioned providers allow.
    • Neither set: the run uses an existing GOOGLE_APPLICATION_CREDENTIALS file after checking it; a URL-source external_account file is rejected.
    • One set: error.
  • An OpenAI parent skips GCP unless both inputs are set; then WIF is prepared for Vertex sub-agents, and a failure is only a warning. Dummy runtimes follow the same rule. GitLab/local Vertex keep their ADC path.
  • Companion agents#1502 uses generic host_files[].optional for six inference harnesses. Scribe is unchanged.

Documentation

This PR updates the GitHub/GitLab setup guides, repo-management and inference guides, CLI references, and workflow-contract notes. Docs are included here; no follow-up docs PR is planned. The guides explain that installing without GCP does not make a Vertex agent runnable without its credentials.

Verification

  • make lint and commit hooks passed.

  • make go-test passed with race detection; overall statement coverage: 82.9%. Changed functions: runGitHubSetupPerRepo 85.4%, resolveInferenceReuse 100%, Converge 83.0%, ProbeComponents 95.7%.

  • New tests cover GitHub and GitLab installation without GCP, status with both secrets absent, partial-pair drift, and existing Vertex setup behavior.

  • Live runs at 75e9279 (and its merge parent); fbfd01d then restores WIF for dummy runtimes, which the behaviour suite needs. The GitHub Actions runs used a disposable per-repo install whose WIF provider came from fullsend inference provision; the local runs used OpenShell 0.1.2:

    Where Runtime and model GCP Result
    Actions claude, opus WIF ✅ triage completed
    Actions pi, xai-vertex/xai/grok-4.6 WIF ✅ triage completed
    Actions pi, openai/gpt-5.6-luna, Vertex subagents.default WIF for sub-agents ✅ triage completed
    Actions pi, openai/gpt-5.6-luna, fleet harness on agents main none ❌ as expected: host_files[1]: GOOGLE_APPLICATION_CREDENTIALS is empty; mark the mount optional or provide a credential file (needs agents#1502)
    Local pi, openai/gpt-5.6-luna, agents#1502 harness none ✅
    Local pi, xai-vertex/xai/grok-4.6 ADC ✅
  • At 2f113b6 the same Actions install failed with host variable GOOGLE_APPLICATION_CREDENTIALS is not set. With the OIDC audience that head sent, the provisioned provider would also have rejected the token exchange.

Generated with Codex

waynesun09 and others added 7 commits September 22, 2026 14:15
…l in github setup

A repository that runs every agent on GPT — through OpenAI Workload
Identity Federation (fullsend-ai#6689, ADR 0092) or the FULLSEND_OPENAI_API_KEY
secret (fullsend-ai#7295) — still had to provide a GCP project and WIF provider it
never used, because per-repo setup treated the GCP pair as the
definition of "inference is configured".

Add "openai" to ValidProviders so inference.provider can declare a
GPT-only repository (vertex stays the default; the runner keeps picking
the provider per model — this value only drives install-time checks).

In `fullsend github setup <owner/repo>`, when the effective provider is
openai and neither --inference-project nor --inference-wif-provider is
supplied, skip the GCP secret lookups and write no GCP secret; the
values were previously written unconditionally, which under openai would
have stored an empty FULLSEND_GCP_PROJECT_ID. Supplying one GCP value
under openai keeps the partial-pair error. Provider openai must have a
route: the --openai-* trio in the composed config or an existing
FULLSEND_OPENAI_API_KEY secret, otherwise setup refuses and names both,
the same way a Vertex install refuses without a project. The two GCP
"is required" errors now point at --inference-provider openai.
A partial inference.openai block in an existing layer is refused as
well, even when the secret exists: the runner errors on a partial trio
instead of falling back to the key, so the secret cannot make it a
route (Codex review).

Tests that used "openai" as the example of an invalid provider now use
"bedrock".

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
…ge, repos install

probe/converge/status defined "inference is configured" as the GCP pair
(requiredSecrets), so a repository on the openai route was reported
incomplete and converge insisted on --inference-project for it.

Add InferenceRoute, read from the committed per-repo config layers
(.fullsend/config.yaml over config.base.yaml): provider, whether the
OpenAI WIF trio is complete, and whether any layer exists. Required
secrets become route-aware: vertex keeps requiredSecretsForForge; openai
needs FULLSEND_OPENAI_API_KEY on GitHub unless the trio is committed,
and on GitLab (no WIF there) the project's own OPENAI_API_KEY variable
plus the forge token. That GitLab variable is probed only — never
provisioned, never deleted — and a test pins it out of every uninstall
list. requiredSecretsForForge stays the managed superset for orphan
detection and uninstall.

ProbeComponents takes a variadic ProbeOption; converge passes
WithDefaultProvider(manifest provider) for a repo with no config yet, so
a fresh install probes the secrets it is about to need. An openai-route
repo with no GCP flags skips WIF derivation and the project requirement,
but converge refuses it when neither the trio nor the OpenAI secret
exists, the same way github setup does. GCP flags passed for the batch
still apply to openai repos (both routes). An unparsable config layer is
an error, not a silent vertex default.
A partial inference.openai block on the openai route is an error from
ProbeInferenceRoute, so probe, status and converge cannot call such a
repository healthy on the strength of the secret. Converge resolves the
route once and hands it to ProbeComponents through WithInferenceRoute,
so its own decisions and the component probe read the same value;
status passes the manifest provider through WithDefaultProvider the
same way converge does. A repair that regenerates config.yaml (missing
workflow file) keeps a committed openai route instead of reverting the
repository to vertex when its manifest entry predates
inference_provider; vertex repositories still get no provider line.
(Codex review.)

repos.yaml gains inference_provider (defaults and per-repo, validated
like runtime, "none" stops the chain); ResolvedConfig and InstallConfig
carry it and BuildScaffoldFiles writes inference.provider only when set,
so vertex installs are unchanged. `repos install --inference-provider`
records it on repos the command adds.

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
…GCP secrets

The setup-gcp composite action declared gcp_wif_provider required and
ran google-github-actions/auth unconditionally, and every reusable
workflow declared the GCP secret pair `required: true`. A repository on
the openai route (inference.provider openai) has neither secret, so its
agent jobs failed at Google auth before the runner started, and
prepare-sandbox-credentials.sh would have read an empty
GOOGLE_APPLICATION_CREDENTIALS.

Make gcp_wif_provider optional and guard all four setup-gcp steps with
`if: inputs.gcp_wif_provider != ''`; declare both GCP secrets
`required: false` in reusable-{dispatch,triage,code,review,fix,retro,
prioritize}.yml. Callers are unchanged: the scaffold shims keep passing
the secrets, which are simply empty on such a repository. The runner's
own GCP refresh loop is already conditional on FULLSEND_GCP_OIDC_URL.
A first "Check GCP inputs" step fails fast when exactly one of the two
inputs is populated: a Vertex repository missing one secret is refused
there, where the message names both, rather than at the first model
call — and no ambient runner credential can stand in for the missing
one (Codex review). Both empty is the openai route and skips the rest.

TestGCPSecretsOptionalForOpenAIRepos pins both halves so a future
`required: true` or an unguarded step is caught.

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
…al GCP settings

- OpenAI guide: new section D, "A repository with no GCP project at
  all" — how to declare the openai route with github setup, what setup
  refuses, how probe/status/install judge it, and that the runner still
  selects the provider per model.
- getting-inference: point GPT-only repositories past the GCP steps.
- repo-management: partial-secret section scoped to the Vertex route;
  new section for openai repositories including the refusal message and
  the GitLab OPENAI_API_KEY ownership rule.
- CLI reference: --inference-provider on github setup and repos
  install; defaults.inference_provider in the manifest key table.
- layered-config, operations and infrastructure references: the GCP
  secrets are not needed / not written for openai repositories; the
  OpenAI secret is the required one there.

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
Keep a committed OpenAI WIF trio when scaffolds are regenerated, carry the selected provider into preset overlays, and implement the documented manifest default editor key.

Assisted-by: Codex
Signed-off-by: Amos Mastbaum <amastbau@redhat.com>
Signed-off-by: Amos Mastbaum <amastbau@redhat.com>
Assisted-by: Codex
Signed-off-by: Amos Mastbaum <amastbau@redhat.com>
Assisted-by: Codex
@amastbau
amastbau requested a review from a team September 25, 2026 15:37
@github-actions

github-actions Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

E2E tests are running

Authorization passed for this commit. See the E2E Tests workflow for results.

@qodo-code-review

qodo-code-review Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

PR Summary by Qodo

Allow repositories to install without GCP and route inference per agent

✨ Enhancement 🧪 Tests 📝 Documentation 🕐 40+ Minutes

Grey Divider

AI Description

• Allow GitHub and GitLab installations without GCP secrets while flagging incomplete credential
 pairs.
Diagram

graph TD
  D["Reusable dispatch"] --> A["Fullsend action"] --> R["Agent run"] --> P{"Inference route"}
  P -- OpenAI --> H["Agent harness"]
  P -- Vertex --> W["GitHub WIF"] --> H
Loading
High-Level Assessment

The following are alternative approaches to this PR:

1. Resolve provider in a workflow preflight step
  • ➕ Could reject missing Vertex secrets before invoking the CLI.
  • ➖ Duplicates runtime and model resolution outside the run.
  • ➖ Risks choosing a different provider from the agent's effective configuration.

Recommendation: Keep provider selection inside the single agent run, where effective runtime, model, and aliases are already resolved. This avoids a second resolver while retaining early Vertex credential validation before the pre-script.

Files changed (33) +1033 / -199

Enhancement (8) +392 / -78
reusable-dispatch.ymlForward optional GCP inputs to agent runs +15/-50

Forward optional GCP inputs to agent runs

• Makes the GCP project secret optional and removes unconditional GCP setup across dispatch jobs. Each stage forwards the inputs to the Fullsend action.

.github/workflows/reusable-dispatch.yml

action.ymlAccept run-scoped GCP inputs +13/-3

Accept run-scoped GCP inputs

• Adds optional GCP project and WIF inputs and passes them to the CLI run step.

action.yml

render.goMake generated GCP mount optional +1/-1

Make generated GCP mount optional

• Generated harnesses can omit the GCP credential file for OpenAI runs.

internal/agentnew/render.go

github.goSkip absent GCP secrets during GitHub setup +7/-4

Skip absent GCP secrets during GitHub setup

• Allows neither GCP input to be configured without secret lookups or empty secret writes. A configured partial pair retains the existing reuse and validation path.

internal/cli/github.go

run.goRoute and validate inference within each run +63/-6

Route and validate inference within each run

• Selects OpenAI or Vertex from the effective runtime and model. GitHub Vertex runs prepare WIF credentials, Vertex optional mounts are validated before pre-scripts, and GCP routing inputs are excluded from the sandbox.

internal/cli/run.go

vertexauth.goPrepare GitHub Vertex WIF credentials +274/-0

Prepare GitHub Vertex WIF credentials

• Adds run-scoped GitHub OIDC retrieval, Google STS validation, temporary credential files, and cleanup. Validates inputs and limits sensitive response exposure.

internal/inference/vertexauth/vertexauth.go

converge.goPermit convergence without GCP inputs +1/-12

Permit convergence without GCP inputs

• Removes the rejection of fresh repositories with no inference project or WIF provider.

internal/repos/converge.go

probe.goTreat absent GCP pair as current +18/-2

Treat absent GCP pair as current

• Keeps both secret components observable while marking a wholly absent pair as matching. A partial pair remains drift.

internal/repos/probe.go

Tests (13) +560 / -66
render_test.goTest generated credential mount +24/-0

Test generated credential mount

• Checks that a rendered, loadable harness marks the GCP credential host file optional.

internal/agentnew/render_test.go

triage.txtRefresh triage harness golden output +1/-0

Refresh triage harness golden output

• Records the optional GCP credential mount in generated triage output.

internal/agentnew/testdata/golden/triage.txt

validation-loop.txtRefresh validation harness golden output +1/-0

Refresh validation harness golden output

• Records the optional GCP credential mount in generated validation-loop output.

internal/agentnew/testdata/golden/validation-loop.txt

github_test.goTest GCP-free and partial GitHub setup +28/-27

Test GCP-free and partial GitHub setup

• Covers setup without GCP secrets and incomplete-pair errors; adjusts secret-check tests for the optional path.

internal/cli/github_test.go

prescript_run_test.goTest credential checks before pre-scripts +87/-0

Test credential checks before pre-scripts

• Verifies OpenAI bypasses GCP setup and Vertex rejects missing inputs or credential files before a pre-script runs.

internal/cli/prescript_run_test.go

preset_test.goTest partial GCP preset composition +3/-3

Test partial GCP preset composition

• Checks that a preset supplying only a GCP project still requires a WIF provider.

internal/cli/preset_test.go

repos_test.goAdjust installation expectations +2/-1

Adjust installation expectations

• Updates manifest-bootstrap expectations for GCP-free installs and preserves a failure-path test through a simulated variable-write error.

internal/cli/repos_test.go

run_test.goPin single-run routing and secret isolation +10/-0

Pin single-run routing and secret isolation

• Checks that no separate provider-resolution flag exists and that GCP routing inputs remain reserved from sandbox exposure.

internal/cli/run_test.go

vertexauth_test.goTest GitHub WIF preparation +241/-0

Test GitHub WIF preparation

• Exercises OIDC and STS exchange, credential contents and cleanup, invalid inputs, response limits, and secret-safe errors.

internal/inference/vertexauth/vertexauth_test.go

converge_test.goTest GCP-free forge convergence +44/-27

Test GCP-free forge convergence

• Verifies fresh GitHub and GitLab installs complete without writing GCP secrets.

internal/repos/converge_test.go

install_test.goAccept installed repos without GCP secrets +3/-3

Accept installed repos without GCP secrets

• Changes the component-check expectation when both optional GCP secrets are absent.

internal/repos/install_test.go

probe_test.goTest absent-pair and partial-pair status +21/-5

Test absent-pair and partial-pair status

• Verifies that status accepts neither GCP secret but rejects a single-secret configuration.

internal/repos/probe_test.go

workflow_call_alignment_test.goPin single-run workflow credential contract +95/-0

Pin single-run workflow credential contract

• Checks optional dispatch and action inputs, one CLI run, removal of separate GCP setup, and functional-test credential forwarding.

internal/scaffold/workflow_call_alignment_test.go

Documentation (11) +79 / -55
github.mdDocument optional GCP setup flags +5/-2

Document optional GCP setup flags

• Clarifies that per-repo setup can omit both GCP flags, although Vertex runs still need credentials.

docs/cli/github.md

repos.mdDocument GCP-free repository installation +4/-1

Document GCP-free repository installation

• Explains when manifest installations can omit GCP inference flags.

docs/cli/repos.md

run.mdExplain per-run inference routing +8/-0

Explain per-run inference routing

• Describes provider selection, GitHub Vertex WIF preparation, and the existing GitLab and local ADC path.

docs/cli/run.md

workflow-contracts.mdUpdate optional-secret workflow contract +1/-1

Update optional-secret workflow contract

• Records that per-repo GCP inputs are optional and Vertex credential failures are checked by the run.

docs/contributing/workflow-contracts.md

configuring-github.mdExplain OpenAI-only GitHub setup +7/-2

Explain OpenAI-only GitHub setup

• Distinguishes optional installation-time GCP flags from credentials required by later Vertex runs.

docs/guides/getting-started/configuring-github.md

configuring-gitlab.mdExplain OpenAI-only GitLab installation +9/-2

Explain OpenAI-only GitLab installation

• Makes Vertex prerequisites conditional and identifies the API-key path for GitLab OpenAI runs.

docs/guides/getting-started/configuring-gitlab.md

getting-inference.mdMake Vertex provisioning guidance conditional +8/-4

Make Vertex provisioning guidance conditional

• Directs OpenAI-only users past GCP WIF provisioning and toward OpenAI credentials.

docs/guides/getting-started/getting-inference.md

repo-management.mdClarify optional GCP secrets and drift +11/-16

Clarify optional GCP secrets and drift

• Documents installations with neither GCP secret and status drift when only one is present.

docs/guides/getting-started/repo-management.md

openai-workload-identity.mdClarify GCP-free OpenAI credential setup +19/-12

Clarify GCP-free OpenAI credential setup

• Explains per-agent routing and documents importing OpenAI identifiers without rerunning repository setup.

docs/guides/infrastructure/openai-workload-identity.md

building-custom-agents.mdRemove unconditional GCP setup from example +2/-9

Remove unconditional GCP setup from example

• Updates the custom-agent workflow example to pass optional GCP inputs to the run instead.

docs/guides/user/building-custom-agents.md

install.goClarify optional installation secrets +5/-6

Clarify optional installation secrets

• Documents that no GCP secrets are written when an inference project is absent and that the managed pair is optional.

internal/repos/install.go

Other (1) +2 / -0
functional-tests.ymlPass GCP inputs to functional runs +2/-0

Pass GCP inputs to functional runs

• Exposes E2E GCP project and WIF secrets to functional tests that exercise Vertex runs.

.github/workflows/functional-tests.yml

@qodo-code-review

qodo-code-review Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

Grey Divider


Action required

1. Per-org and custom Vertex jobs now fail at credential setup ✓ Resolved
Description
runAgent calls vertexauth.PrepareGitHubWIF for every Vertex run on GitHub Actions, while
action.yml populates its required GCP values only from inputs that default to empty. Per-org
reusable stage workflows and custom workflows following the old guide can authenticate before
invoking the action without those inputs, but PrepareGitHubWIF rejects the empty values and stops
the run before the pre-script or inference.
Code

internal/cli/run.go[R1132-1135]

+	if provider == runProviderVertex {
+		if os.Getenv("GITHUB_ACTIONS") == "true" {
+			vertexEnv, cleanup, err := vertexauth.PrepareGitHubWIF(ctx, vertexauth.Config{
+				ProjectID:                os.Getenv(vertexinference.SecretProjectID),
Relevance

●●● Strong

Recent run.go precedents accept regressions where new credential setup breaks previously supported
authentication paths.

PR-#7510
PR-#7394

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The standalone code and prioritize workflows invoke the action without the new GCP inputs, despite
retaining their earlier GCP setup steps. The action therefore exports empty
FULLSEND_GCP_PROJECT_ID and FULLSEND_GCP_WIF_PROVIDER values; the Vertex path reads them and
rejects an empty project ID rather than using the credentials already established by setup.

.github/workflows/reusable-code.yml[148-183]
action.yml[434-435]
internal/inference/vertexauth/vertexauth.go[81-86]
.github/workflows/reusable-code.yml[183-188]
.github/workflows/reusable-prioritize.yml[160-180]
.github/actions/setup-gcp/action.yml[18-42]
internal/cli/run.go[1132-1144]
internal/inference/vertexauth/vertexauth.go[80-85]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Vertex runs on GitHub Actions call `PrepareGitHubWIF` even when the job has already authenticated. Standalone reusable stage workflows retain their GCP setup but omit the action’s new GCP inputs, causing the run to fail before inference.

## Fix Focus Areas
- internal/cli/run.go[1132-1153]
- action.yml[434-435]
- .github/workflows/reusable-code.yml[148-190]
- .github/workflows/reusable-prioritize.yml[160-180]

## Recommended Fix
Skip `PrepareGitHubWIF` when both `FULLSEND_GCP_*` variables are empty and `GOOGLE_APPLICATION_CREDENTIALS` points to a regular file, allowing `validateVertexGCPCredentials` to handle existing credentials. Otherwise, pass `gcp_wif_provider` and `gcp_project_id` as action inputs in every standalone per-org reusable stage workflow, including prioritize, while retaining existing setup behavior where required.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


2. Health checks approve jobs that fail ✗ Dismissed
Description
requiredSecretsForRouteMode removes both GCP secrets from OpenAI-route probing and never inspects
them for pair consistency. An OpenAI repository with its required OpenAI credential plus one
leftover GCP secret is therefore reported healthy by probe, status, and converge, although the
workflow's GCP setup action rejects that same state.
Code

internal/repos/route.go[R111-114]

+	for _, s := range base {
+		if s == forge.SecretGCPProjectID || s == forge.SecretGCPWIFProvider {
+			continue
+		}
Relevance

●●● Strong

Filtering GCP secrets hides partial leftovers even though reusable workflows still pass them to
setup-gcp, creating inconsistent health and runtime behavior.

PR-#6960

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The route helper filters the GCP names before ProbeComponents performs existence checks, while the
workflow action rejects exactly-one configurations. Consequently all repository-management health
paths can miss a state that deterministically fails CI.

internal/repos/route.go[105-123]
internal/repos/probe.go[281-301]
internal/repos/converge.go[487-506]
.github/actions/setup-gcp/action.yml[22-27]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Route-sensitive health checks omit both GCP secrets for OpenAI repositories, so they cannot detect a partial optional GCP pair that the workflow rejects at runtime.

## Fix Focus Areas
- internal/repos/route.go[105-123]
- internal/repos/probe.go[281-301]
- .github/actions/setup-gcp/action.yml[22-27]

## Recommended Fix
Probe the optional GCP pair separately from route-required secrets and return an unhealthy or error state when exactly one exists. Keep neither and both valid for an OpenAI route, and add status and converge tests for each partial state.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


3. Installs accept an unusable route ✓ Resolved
Description
Converge checks for an OpenAI credential only when both GCP command-line values are absent, even
though the repository's generated provider remains OpenAI. Supplying GCP configuration for a batch
therefore lets an OpenAI repository without either its key or WIF trio install successfully, after
which GPT calls have no OpenAI credential.
Code

internal/repos/converge.go[R564-565]

+		openAIOnly := d.route.OpenAI() && cfg.InferenceProject == "" && cfg.WIFProvider == ""
+		if openAIOnly && !d.route.OpenAIWIF && !hasComponent(d.components, "secret:"+openAIRouteSecret(d.resolved.Forge)) {
Relevance

●●● Strong

Supplying unrelated GCP flags bypasses required OpenAI credential validation, permitting an unusable
generated route.

PR-#6960

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The validation guard is disabled by any GCP configuration, while scaffold generation still writes
provider: openai and secret installation only provisions the GCP pair. The route helper confirms
that an OpenAI repository independently requires the OpenAI key unless its committed WIF trio is
complete.

internal/repos/converge.go[560-578]
internal/repos/converge.go[794-799]
internal/repos/install.go[588-603]
internal/repos/route.go[117-122]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Converge makes OpenAI credential validation conditional on the absence of GCP flags, even though adding a secondary GCP route does not satisfy the configured OpenAI route.

## Fix Focus Areas
- internal/repos/converge.go[560-578]
- internal/repos/converge.go[794-799]
- internal/repos/install.go[588-603]

## Recommended Fix
Validate the OpenAI key or complete WIF trio whenever the effective route is OpenAI, independently of whether GCP flags are also supplied. Continue provisioning the optional complete GCP pair, but fail before writes if the OpenAI route itself is missing, and add a regression test with GCP flags and no OpenAI credential.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


View high (4)
4. Routine sync reverts OpenAI-only repos ✗ Dismissed
Description
driftInstallConfig renders expected scaffold content from resolved.InferenceProvider, which is
manifest-only, while the new route discovery correctly treats a committed `inference.provider:
openai as authoritative. When an existing repository configured through github setup` has no
manifest inference_provider and content-drift comparison runs, status reports its config as stale
and converge rewrites it without the OpenAI provider, causing later health checks and installs to
require GCP credentials again.
Code

internal/repos/install.go[R366-369]

		UpstreamRef:       ref,
		UpstreamTag:       ref,
		Runtime:           resolved.Runtime,
+		InferenceProvider: resolved.InferenceProvider,
Relevance

●●● Strong

Drift rendering uses manifest-only provider data and can overwrite a committed OpenAI route;
preserving route identity is correctness-critical.

PR-#215

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The new discovery path records a committed route in convergeDiscovery, and missing-component
repair explicitly preserves that route. In contrast, both status and regular convergence
content-drift rendering build expected files from driftInstallConfig, which receives only
ResolvedConfig and therefore uses an empty legacy manifest field instead of the committed OpenAI
route.

internal/repos/converge.go[487-506]
internal/repos/converge.go[974-980]
internal/repos/converge.go[1914-1920]
internal/repos/converge.go[2008-2073]
internal/repos/install.go[365-369]
internal/repos/status.go[464-478]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Content-drift rendering uses only the manifest's resolved inference provider, not the repository route discovered from committed config. Therefore an existing OpenAI-only repository whose manifest predates `inference_provider` is rendered as the vertex default and a routine sync can replace its committed OpenAI configuration.

## Fix Focus Areas
- internal/repos/install.go[365-369]
- internal/repos/converge.go[974-980]
- internal/repos/converge.go[2008-2073]
- internal/repos/status.go[248-252]

## Recommended Fix
Carry the discovered `InferenceRoute` into content-drift rendering and status expected-file generation. Render the provider and complete OpenAI WIF identifiers from a committed OpenAI route when present, falling back to the manifest only for repositories with no committed configuration.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


5. Setup accepts credentials that break jobs ✓ Resolved
Description
resolveInferenceReuse returns immediately for an OpenAI route with no newly supplied GCP values,
without checking whether the repository already contains exactly one GCP secret. When such a partial
legacy pair exists, setup succeeds but every reusable workflow forwards the mismatched values to
setup-gcp, whose new validation exits before the agent starts.
Code

internal/cli/github.go[R766-767]

+	if effectiveInferenceProvider(cfg, effective) == config.InferenceProviderOpenAI && project == "" && wif == "" {
+		return false, false, nil
Relevance

●●● Strong

The early return bypasses partial-secret detection, allowing a state that setup-gcp explicitly
rejects; analogous credential validation defects were accepted.

PR-#6960

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The early return bypasses both RepoSecretExists checks. Reusable workflows pass both repository
secrets to the composite action, and that action explicitly exits when exactly one value is
populated.

internal/cli/github.go[763-790]
.github/workflows/reusable-triage.yml[153-158]
.github/actions/setup-gcp/action.yml[22-27]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
OpenAI setup skips all existing GCP-secret checks when no GCP values are supplied, allowing a repository with exactly one legacy GCP secret to pass setup even though its workflows will fail.

## Fix Focus Areas
- internal/cli/github.go[763-790]
- .github/actions/setup-gcp/action.yml[22-27]

## Recommended Fix
Before taking the OpenAI no-GCP return path, probe both existing GCP secrets and reject the repository when exactly one exists. Allow neither or both, and add tests covering both partial legacy states.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


6. OpenAI preset installs are rejected ✓ Resolved
Description
Converge calls ProbeInferenceRoute before loading resolved.Config, so the new OpenAI-only
validation sees neither the WIF trio that the preset will write nor the static-key secret. A newly
added repository with inference_provider: openai, no GCP flags, and a complete OpenAI trio in its
manifest preset is rejected even though its generated layered configuration would provide a working
route.
Code

internal/repos/converge.go[R564-566]

+		openAIOnly := d.route.OpenAI() && cfg.InferenceProject == "" && cfg.WIFProvider == ""
+		if openAIOnly && !d.route.OpenAIWIF && !hasComponent(d.components, "secret:"+openAIRouteSecret(d.resolved.Forge)) {
+			repoFullName := d.repo.Owner + "/" + d.repo.Repo
Relevance

●●● Strong

Route probing precedes preset application, so valid OpenAI preset credentials are invisible and
installation can be incorrectly rejected.

PR-#5976

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The route is probed exclusively from committed config layers during discovery, whereas the preset is
only loaded in the later convergence phase. The added OpenAI-only check rejects a route without a
committed WIF trio or key, and the subsequent install configuration receives its WIF fields only
from that earlier route.

internal/repos/converge.go[487-496]
internal/repos/converge.go[536-574]
internal/repos/converge.go[794-802]
internal/repos/route.go[45-88]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
OpenAI-only repositories cannot use a complete OpenAI WIF trio supplied by their manifest config preset. Route discovery and validation happen before the preset is loaded, so the validation requires a static OpenAI key even though installation would commit the preset as `.fullsend/config.base.yaml`.

## Fix Focus Areas
- internal/repos/converge.go[487-496]
- internal/repos/converge.go[536-574]
- internal/repos/converge.go[794-802]
- internal/repos/route.go[45-88]

## Recommended Fix
Load and compose the declared preset before evaluating the effective inference route for a fresh repository, or extend route resolution to accept the preset layer. Use the resulting layered OpenAI WIF identifiers both for the OpenAI-route health check and for `InstallConfig.InferenceOpenAI`, while retaining rejection of partial trios.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


7. OpenAI agent workflows cannot start ✓ Resolved
Description
The new GCP setup conditions skip authentication and credential preparation when the provider input
is empty, but the first-party agents harnesses still require ${GOOGLE_APPLICATION_CREDENTIALS} as
a host file. OpenAI-only repositories using standard roles such as triage or review therefore reach
sandbox creation with an empty source path and stop before the agent starts.
Code

.github/actions/setup-gcp/action.yml[35]

+      if: inputs.gcp_wif_provider != ''
Relevance

●●● Strong

The PR explicitly documents this harness incompatibility as a known gap; agent startup remains
broken until the companion change lands.

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The PR gates both Google authentication and sandbox credential preparation on a nonempty GCP
provider and explicitly documents that this leaves the credential variable unset. The agents
repository selects the affected harnesses as its standard roles, while both representative triage
and review harnesses declare the variable as a required host-file source without optional: true.

fullsend -> agents
.github/actions/setup-gcp/action.yml[34-39]
.github/actions/setup-gcp/action.yml[52-57]
docs/guides/infrastructure/openai-workload-identity.md[329-334]
External repo: fullsend-ai/agents, config.yaml [4-11]
External repo: fullsend-ai/agents, harness/triage.yaml [23-31]
External repo: fullsend-ai/agents, harness/review.yaml [30-38]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
OpenAI-only workflows skip GCP authentication, leaving `GOOGLE_APPLICATION_CREDENTIALS` unset, while the pinned first-party agents harnesses require that value as a sandbox host file. This prevents standard agent workflows from starting.

## Fix Focus Areas
- .github/actions/setup-gcp/action.yml[35-53]
- /cross_repos/agents/config.yaml[4-11]
- /cross_repos/agents/harness/triage.yaml[23-31]
- /cross_repos/agents/harness/review.yaml[30-38]

## Recommended Fix
Update every first-party harness in the agents repository so the `${GOOGLE_APPLICATION_CREDENTIALS}` host-file entry is optional, validate an OpenAI-only workflow end to end, and coordinate that agents change before enabling the no-GCP route here.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools



Remediation recommended

8. OpenAI users are told GCP is required ✓ Resolved
Description
The static-key guide says fullsend github setup and fullsend repos install require a GCP project
and WIF provider pair, although both commands now support setup without either value. An OpenAI-only
repository updating older workflows is therefore directed to obtain unnecessary credentials to sync
OpenAI key forwarding, despite the guide later saying the pair is optional.
Code

docs/guides/infrastructure/openai-workload-identity.md[R282-283]

+   for a manifest-managed repository) can sync them when the repository already has the GCP
+   project and WIF provider pair; those setup commands still require the pair. The caller shim must
Relevance

●●● Strong

Recent docs corrections consistently accept inaccurate user-facing behavior guidance, especially
credential and setup requirements.

PR-#7074
PR-#7705

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The guide’s claim that both setup commands require the GCP pair conflicts with its later
instructions and with the setup implementation’s early return, which permits neither GCP value and
leaves existing secrets untouched.

Rule 2748504: Update docs/ references when changing user-facing behavior
docs/guides/infrastructure/openai-workload-identity.md[280-284]
internal/cli/github.go[746-754]
docs/cli/github.md[40-43]
docs/guides/infrastructure/openai-workload-identity.md[330-332]
internal/cli/github.go[752-755]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The static-key guide incorrectly tells OpenAI-only repositories that syncing older workflows to forward OpenAI credentials requires a GCP project and WIF provider pair.

## Fix Focus Areas
- docs/guides/infrastructure/openai-workload-identity.md[280-284]

## Recommended Fix
Remove the claim that `fullsend github setup` and `fullsend repos install` require the GCP pair. Explain that users can sync workflow files for OpenAI credential forwarding without GCP inference flags, while Vertex runs still need the pair.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


9. Vertex runs lose service account impersonation ✓ Resolved
Description
externalAccountConfig never sets service_account_impersonation_url, and the environment it
returns replaces any existing GOOGLE_APPLICATION_CREDENTIALS on every GitHub Actions Vertex run.
Setups where only a service account holds Vertex permission pass the STS-only validation, then fail
on the first model call.
Code

internal/inference/vertexauth/vertexauth.go[R195-199]

+func externalAccountConfig(audience, tokenURL, tokenFile string) externalAccountCredentials {
+	return externalAccountCredentials{
+		Type:             "external_account",
+		Audience:         audience,
+		SubjectTokenType: subjectTokenType,
Relevance

●●● Strong

Recent authentication precedents accept credential-flow correctness and security fixes affecting
token exchange and credential propagation.

PR-#6695
PR-#7780

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The old script copied service_account_impersonation_url into the sandbox credentials; the new struct
has no such field. Functional tests authenticate through a service account and now also export the
FULLSEND_GCP pair.

internal/scaffold/fullsend-repo/scripts/prepare-sandbox-credentials.sh[34-39]
.github/workflows/functional-tests.yml[229-248]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The generated WIF credential file omits service_account_impersonation_url, whereas prepare-sandbox-credentials.sh preserved it. Impersonation-based setups lose Vertex access.

## Fix Focus Areas
- internal/inference/vertexauth/vertexauth.go[38-45]
- internal/inference/vertexauth/vertexauth.go[195-209]

## Recommended Fix
Add an optional ServiceAccount field to Config (for example from FULLSEND_GCP_SERVICE_ACCOUNT). When it is set, emit service_account_impersonation_url in both the host and sandbox credential files.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


10. Workflow review evidence is missing ✗ Dismissed
Description
The PR changes reusable-workflow secret contracts and composite authentication behavior without
stating that docs/contributing/ci-workflows.md was consulted. This affects every workflow whose
GCP requirements were relaxed, so the protected-path review lacks the explicit checklist evidence
required for these changes.
Code

.github/workflows/reusable-dispatch.yml[R81-83]

+      # Both GCP secrets are optional: a repository whose inference runs on
+      # OpenAI (inference.provider openai, #7481) has neither, and the
+      # setup-gcp action skips Google auth when the provider is empty.
Relevance

●●● Strong

Protected workflow and secret changes require explicit guide-review evidence; the PR provides
neither checklist nor CI guide reference.

PR-#3039

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
Rule 3201588 requires an explicit PR-description, checklist, or inline reference when workflows or
secrets are modified. The cited changes alter reusable secret requirements and authentication
routing, while the supplied PR description and comments contain no reference to the CI workflow
guide.

Rule 3201588: Consult CI workflow guide when modifying GitHub Actions workflows or secrets
.github/workflows/reusable-dispatch.yml[81-87]
.github/actions/setup-gcp/action.yml[19-27]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The PR modifies GitHub Actions workflows and secret contracts but contains no explicit evidence that the repository's CI workflow guide was consulted.

## Fix Focus Areas
- .github/workflows/reusable-dispatch.yml[81-87]
- .github/actions/setup-gcp/action.yml[19-27]

## Recommended Fix
Review `docs/contributing/ci-workflows.md`, verify the modified contracts and composite-action context follow it, and add an explicit consultation statement to the PR description or an appropriate inline review comment.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


View medium (3)
11. Vertex jobs start without credentials ✓ Resolved
Description
reusable-dispatch.yml now marks both GCP secrets optional, while the generated per-repository shim
calls it without a secrets: mapping. When a Vertex repository uses that shim, workflow-call
validation no longer stops the run and each inline stage reaches setup-gcp with empty values,
which skips authentication.
Code

.github/workflows/reusable-dispatch.yml[87]

+        required: false
Relevance

●●● Strong

Optional secrets let the generated shim bypass validation while Vertex jobs receive empty
credentials, matching accepted dispatch secret-threading concerns.

PR-#3903
PR-#6960

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
Rule 1062045 requires equivalent secret threading across both dispatch paths. The changed
declaration permits empty GCP values, the generated shim forwards no secrets, and the workflow
contract documents that this condition is a correctness failure rather than valid inheritance.

Rule 1062045: Keep jq payload, stage routing, and secret threading logic in dispatch workflows in sync
.github/workflows/reusable-dispatch.yml[84-87]
internal/scaffold/fullsend-repo/templates/shim-workflow-call.yaml[48-57]
docs/contributing/workflow-contracts.md[3-10]
.github/actions/setup-gcp/action.yml[19-35]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The generated per-repository shim does not forward inference secrets to the reusable dispatch workflow, and making the GCP secrets optional allows Vertex jobs to proceed without authentication.

## Fix Focus Areas
- .github/workflows/reusable-dispatch.yml[84-87]
- internal/scaffold/fullsend-repo/templates/shim-workflow-call.yaml[48-57]

## Recommended Fix
Add explicit secret mappings to the shim call for both GCP secrets and the OpenAI API key. Do not use `secrets: inherit`; preserve optional OpenAI-only operation while ensuring configured Vertex repositories pass their credentials.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


12. Deprecated installs gain new behavior ✓ Resolved
Description
reusable-code.yml and its sibling stage workflows relax their GCP secret contracts even though the
workflow contract classifies these standalone stages as serving the deprecated per-organization
chain. The change extends that installation mode to OpenAI-only credentials without identifying ADR
0044 in the PR description, obscuring whether continued legacy support is intentional.
Code

.github/workflows/reusable-code.yml[49]

+        required: false
Relevance

●●● Strong

This directly changes deprecated workflow contracts, and the PR neither acknowledges ADR 0044 nor
explains intentional legacy support.

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
Rule 2795055 prohibits extending deprecated per-organization installation behavior and requires any
touch to be called out with ADR 0044. The workflow contract says the standalone stage workflows
serve only that deprecated chain, while the PR changes their credential behavior and contains no ADR
0044 acknowledgement.

Rule 2795055: Flag and avoid changes to deprecated per-org installation mode content (ADR 0044)
.github/workflows/reusable-code.yml[45-51]
docs/contributing/workflow-contracts.md[3-9]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Standalone stage workflows used only by the deprecated per-organization chain are being extended with OpenAI-only credential behavior without the required ADR 0044 callout.

## Fix Focus Areas
- .github/workflows/reusable-code.yml[45-51]
- .github/workflows/reusable-fix.yml[57-63]
- .github/workflows/reusable-review.yml[45-51]
- .github/workflows/reusable-retro.yml[45-51]
- .github/workflows/reusable-triage.yml[46-52]

## Recommended Fix
Remove the new behavior from workflows that exclusively serve deprecated per-organization installations. If touching that mode is unavoidable, explicitly identify ADR 0044 and the deprecated scope in the PR description without broadening legacy functionality.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


13. Guides still require GCP for OpenAI ✓ Resolved
Description
The changed --inference-provider behavior makes GCP project and WIF settings optional for OpenAI,
but three existing setup and prerequisite sections still present GCP infrastructure as mandatory.
Users following the repository-management, GitHub-configuration, or advanced-setup paths therefore
receive instructions that exclude the newly supported route.
Code

internal/cli/github.go[190]

+	cmd.Flags().StringVar(&cfg.inferenceProvider, "inference-provider", "", "inference provider: vertex (resolved if unset) or openai; openai makes --inference-project and --inference-wif-provider optional and needs an OpenAI route (the --openai-* trio or the FULLSEND_OPENAI_API_KEY secret)")
Relevance

●●● Strong

User-facing provider behavior changed, but several cited setup guides still present GCP as
mandatory; documentation consistency findings are typically accepted.

PR-#7057
PR-#5763

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
Rules 2748504 and 2806787 require all references to changed CLI behavior to remain current. The
implementation and dedicated reference now allow OpenAI-only setup, while the cited guides continue
to impose unconditional GCP prerequisites.

Rule 2748504: Update docs/ references when changing user-facing behavior
Rule 2806787: Update CLI-related docs according to docs/contributing/documentation.md when changing CLI surface
internal/cli/github.go[190-190]
docs/guides/getting-started/repo-management.md[24-26]
docs/guides/getting-started/configuring-github.md[10-14]
docs/guides/infrastructure/advanced-setup.md[21-21]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Several setup guides still state or imply that every installation requires a GCP project and WIF provider, despite the new OpenAI-only route.

## Fix Focus Areas
- docs/guides/getting-started/repo-management.md[24-26]
- docs/guides/getting-started/configuring-github.md[10-14]
- docs/guides/getting-started/configuring-github.md[60-69]
- docs/guides/infrastructure/advanced-setup.md[21-21]

## Recommended Fix
Qualify the GCP prerequisites as Vertex-specific and add the OpenAI-only alternative, including its required OpenAI credential route and the fact that no GCP project or provider is needed.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


Grey Divider

Context sources
✅ Compliance rules (platform): 72 rules
✅ Cross-repo context — repo relationships
  Explored: repo: fullsend-ai/experiments (sha: 3f1614fa) — View relationship
  Explored: repo: fullsend-ai/.fullsend (sha: 7c163cad) — View relationship
  Explored: repo: fullsend-ai/metrics (sha: b66b1655) — View relationship
  Explored: repo: fullsend-ai/pi-xai-vertex (sha: 8f50bdda) — View relationship
  Explored: repo: fullsend-ai/scribe (sha: 0f3eb528) — View relationship
  Explored: repo: fullsend-ai/mlflow-assessments (sha: 46f42987) — View relationship
  Explored: repo: fullsend-ai/pi-anthropic-vertex (sha: 960eff35) — View relationship
  ⚠️ Failed to retrieve: fullsend-ai/agents
Review mode: 🧠 Deep: This push contains substantial, independently risky logic across inference authentication, repository lifecycle/configuration, CLI paths, CI scaffolds, and multiple runtime/workflow integrations, making redundant review materially valuable.

Grey Divider

Tip of the day
💡 Did you know, you can route each severity your way: inline, summary, both, or drop

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Grey Divider

Qodo Logo

Comment thread .github/workflows/reusable-dispatch.yml Resolved
Comment thread internal/cli/github.go Resolved
Comment thread .github/workflows/reusable-dispatch.yml Resolved
Comment thread .github/workflows/reusable-code.yml Resolved
Comment thread internal/cli/github.go Resolved
Comment thread internal/repos/route.go Resolved
Comment thread internal/repos/converge.go Resolved
Comment thread internal/repos/converge.go Resolved
Comment thread internal/repos/install.go Resolved
Comment thread .github/actions/setup-gcp/action.yml Resolved
Allow fleet harnesses to opt a required host file into an OpenAI-only empty-source skip. Keep required-file errors for Vertex runs and unannotated mounts, including Scribe Drive credentials.

Signed-off-by: Amos Mastbaum <amastbau@redhat.com>
Assisted-by: Codex
@waynesun09

Copy link
Copy Markdown
Member

/fs-review

@fullsend-ai-review

fullsend-ai-review Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 9:20 PM UTC · Completed 9:47 PM UTC

Commit: 1d03f99 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5 · Effort: high · Cost: $13.45

@github-actions

github-actions Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

Site preview

Preview: https://ff2deee8-site.fullsend-ai.workers.dev

Commit: fbfd01dbca1e5f0bf9dadd15eaf5c5b969725d06

@codecov

codecov Bot commented Sep 25, 2026 •

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 76.39752% with 38 lines in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
internal/inference/vertexauth/vertexauth.go 73.97% 21 Missing and 17 partials ⚠️

📢 Thoughts on this report? Let us know!

@fullsend-ai-review fullsend-ai-review Bot added the risk/elevated PR risk: elevated label Sep 25, 2026
@fullsend-ai-review

Copy link
Copy Markdown

Risk Assessment: elevated (3/5)

Details

Large diff spanning 47 files with modifications to 8 protected CI workflow paths and high historical churn in touched packages elevate risk, balanced by an opt-in configuration design and complete alignment with the linked issue.

@fullsend-ai-review

Copy link
Copy Markdown

Review

Findings

High

  • [regression] internal/repos/probe.go:138 — This PR's conflict resolution against current main dropped GitLab scaffold/probe/converge/uninstall behavior that already landed on main, and every test that pins it — and it introduces a guaranteed compile failure. Verified independently: current main's internal/repos/uninstall.go defines gitlabAuxiliaryScriptPaths(), the package-level var gitlabRetiredScaffoldPaths, and the constants gitlabInstallCLIScriptPath/gitlabPollJobScriptPath/gitlabAgentJobScriptPath. None of these exist anywhere in this PR's changed files. internal/repos/remote_scaffold_test.go:113 and internal/repos/uninstall_test.go:651-653 (both unchanged by this PR, same repos package) still reference those symbols — so the repos test package will fail to compile once this PR merges. Beyond the compile break, probe.go:138 still walks only the two script paths instead of the fuller GitLab helper set and no longer uses readWorkflowMarker; converge.go's orphan handling only warns instead of deleting known-retired GitLab paths. Several GitLab-specific tests present on current main are absent here. This contradicts the PR's own testing checklist, which claims go test ./internal/repos ... and make go-test passed.
    Remediation: Re-merge current main into probe.go, uninstall.go, converge.go, probe_test.go, converge_test.go (and check remote_scaffold_test.go/uninstall_test.go for compatibility) keeping both sides: restore readWorkflowMarker, gitlabAuxiliaryScriptPaths, gitlabRetiredScaffoldPaths, the additional GitLab helper-script constants, and every GitLab probe/converge test from main — then re-apply only the OpenAI route changes on top.

Medium

  • [logic-error] internal/repos/converge.go:258 — secretsPresent still hard-codes the GCP secret pair as the probed component list, so on the openai route hasSecrets is always false. An already-installed openai repo's re-converge path (convergeSecrets) only writes secrets that appear as missing probe components, so batch --inference-project/--inference-wif-provider flags are silently ignored for existing openai repos on re-converge — contradicting this PR's own documented dual-route behavior. The fresh-install path is unaffected and is the only path the new tests cover.
    Remediation: Make secretsPresent/convergeSecrets route-aware so GCP flags can still be applied to an existing openai repo on re-converge.

  • [stale-guide-instructions] docs/guides/infrastructure/openai-workload-identity.md:274 — Route C step 1 tells users to re-run fullsend github setup <owner/repo> (without --inference-provider openai) before step 3 sets FULLSEND_OPENAI_API_KEY, which fails on a repository with no GCP credentials. Neither Route B nor Route C states GCP is not required, even though Section D of the same doc gives the correct order — a reader following Route C's numbered steps can get stuck. This falls short of issue Make the GCP/Vertex inference settings optional when a repository runs inference on OpenAI (WIF or static key) #7481's documentation goal.
    Remediation: Point Route C's openai-only path at Section D's secret-first ordering and state explicitly in Route B/C that GCP is not required when --inference-provider openai is used.

  • [missing-documentation] docs/guides/infrastructure/infrastructure-reference.md:415 — The GitLab CI/CD variables table still lists the GCP pair as required and OPENAI_API_KEY as merely optional, without reflecting the new openai route (the equivalent GitHub section a few dozen lines above was updated).
    Remediation: Note that the GCP pair is not required and OPENAI_API_KEY is required on GitLab when inference_provider is openai.

  • [contradictory-prerequisites] docs/guides/getting-started/repo-management.md:28 — Prerequisites/installation notes still assert GCP WIF provisioning is unconditionally required before repos install, contradicting a new section added later in the same file stating openai-route repos need neither GCP secret.
    Remediation: Qualify the prerequisites to state GCP WIF provisioning is required only for the default Vertex route.

  • [cross-repo-schema-coordination] internal/harness/harness.go:88 — host_files[].optional_for_openai is the sole mechanism the companion fullsend-ai/agents#1502 PR is expected to consume. Harness YAML parsing uses plain yaml.Unmarshal (pre-existing, not introduced here), which silently drops unrecognized keys — a field-name/casing mismatch on the agents-repo side would parse as a no-op and fail OpenAI-only sandbox bootstrap with no warning. This review has no access to the agents repo to confirm alignment.
    Remediation: Confirm agents#1502's harness YAML uses the exact key optional_for_openai at the host_files[] item level before/alongside merging.

  • [protected-path] .github/actions/setup-gcp/action.yml and 7 .github/workflows/reusable-*.yml files — This PR modifies protected-path files. It links an authorizing issue (Make the GCP/Vertex inference settings optional when a repository runs inference on OpenAI (WIF or static key) #7481) and explains its rationale, and the author's own description acknowledges these changes "require the repository's normal human review." Human approval is always required for protected-path changes regardless of stated justification.

Low

  • [logic-error] internal/cli/github.go:766 — resolveInferenceReuse skips existing-secret inspection on the openai shortcut, so a leftover partial GCP pair is only caught later at workflow runtime rather than at setup/probe time.
    Remediation: Fail setup/probe on a detected partial GCP pair even when using the openai shortcut.

  • [architectural-coherence] internal/harness/harness.go:88 — OptionalForOpenAI bakes a provider name into an otherwise provider-neutral HostFile schema field; a one-provider escape hatch that won't scale cleanly to a third provider.
    Remediation: Consider generalizing provider conditioning for host files, or record this as an accepted trade-off in an ADR.

  • [documentation-intent] docs/guides/infrastructure/openai-workload-identity.md:327 — The "Known gap, harness side" callout doesn't mention that host_files[].optional_for_openai (added by this PR) is the mechanism a custom/local harness can use today.
    Remediation: Update the callout to mention optional_for_openai: true as the available mechanism alongside the fleet-harness gap note.

  • [incomplete-cli-reference] docs/cli/repos.md:94 — The repos install "Manifest add" prose omits --inference-provider from its list of per-repo overrides, though the flags table below documents it.
    Remediation: Add --inference-provider to the parenthetical list at line 94.

  • [incomplete-cli-reference] docs/cli/repos.md:114 — The flags table for --inference-project/--inference-wif-provider doesn't note they're optional for the openai route, unlike the equivalent docs/cli/github.md wording.
    Remediation: Note these flags are required only for the vertex route.

  • [incomplete-manifest-reference] docs/guides/getting-started/repo-management.md:137 — The none-sentinel field list omits runtime and the newly added inference_provider.
    Remediation: Add both fields to the list.

  • [contradictory-prerequisites] docs/guides/getting-started/configuring-github.md:13 — This unedited guide lists GCP WIF as an unconditional prerequisite with no pointer to the new openai-only path.
    Remediation: Add a cross-reference to openai-workload-identity.md for OpenAI-only setups.

  • [cross-repo-reference-accuracy] docs/guides/infrastructure/openai-workload-identity.md:331 — Cites agents#1383 while the PR body cites agents#1502/agents#1382, with no cross-reference between the numbers; could not verify from this repo whether they're distinct.
    Remediation: Confirm the issue numbers are intentionally distinct or correct a typo.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR

@fullsend-ai-review fullsend-ai-review Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note: The following review comments could not be posted on the diff (GitHub returned 422) and are included here instead:

  • internal/repos/probe.go (file-level): Line 138 · [high] regression

This PR's conflict resolution against current main dropped GitLab scaffold/probe/converge/uninstall behavior that already landed on main, and every test that pins it — and it introduces a guaranteed compile failure. Verified independently: target-repo/internal/repos/uninstall.go (current main) defines gitlabAuxiliaryScriptPaths(), the package-level var gitlabRetiredScaffoldPaths, and the constants gitlabInstallCLIScriptPath/gitlabPollJobScriptPath/gitlabAgentJobScriptPath. None of these exist anywhere in this PR's changed files (grep -rn across /sandbox/workspace/pr-head/ returns zero hits). internal/repos/remote_scaffold_test.go:113 (unchanged by this PR, package repos) references gitlabRetiredScaffoldPaths, and internal/repos/uninstall_test.go:651-653 (also unchanged) references gitlabInstallCLIScriptPath/gitlabPollJobScriptPath/gitlabAgentJobScriptPath. Since those two test files are not part of this diff, they retain their current-main content when this PR merges — and the repos test package will fail to compile. Beyond the compile break, probe.go (line 138) still walks only gitlabTrustScriptPath/gitlabRoleTokenScriptPath instead of the fuller gitlabAuxiliaryScriptPaths() set, and no longer uses readWorkflowMarker; converge.go's orphan handling only warns instead of deleting known-retired GitLab paths. GitLab-specific tests present on current main (TestProbeComponents_GitLab_MissingExtractedJobScripts, TestProbeComponents_GitLab_ReadsPipelineMarker, TestProbeComponents_GitLab_FallsBackToLegacyDispatchMarker, TestProbeComponents_GitLab_PrefersPipelineMarkerOverLegacy, TestProbeComponents_GitLab_LegacyDispatchOnly, TestConverge_GitLab_RepairsMissingExtractedJobScripts, TestConverge_GitLab_RemovesObsoleteDispatchFile, TestConverge_GitLab_RemovesObsoleteDispatchFile_DryRun, TestConverge_GitLab_KeepsDispatchStubWhenPinnedRefWrapperStillIncludesIt) are absent from this PR's probe_test.go/converge_test.go. This contradicts the PR's own testing checklist, which claims go test ./internal/repos ... and make go-test passed.

Suggested fix: Re-merge current main into probe.go, uninstall.go, converge.go, probe_test.go, converge_test.go, and check whether remote_scaffold_test.go/uninstall_test.go need matching updates. Keep both sides: restore readWorkflowMarker, gitlabAuxiliaryScriptPaths, gitlabRetiredScaffoldPaths (and the converge delete path that consumes it), the additional GitLab helper-script constants/paths, and every GitLab probe/converge test from main — then re-apply only the OpenAI route changes (requiredSecretsForRoute, WithInferenceRoute, and the new openai-route tests) on top.

  • internal/repos/converge.go (file-level): Line 258 · [medium] logic-error

secretsPresent still hard-codes both FULLSEND_GCP_PROJECT_ID and FULLSEND_GCP_WIF_PROVIDER as the probed component list. On the openai route those GCP secrets are not probed, so hasSecrets is always false for an openai repo. An already-installed openai repo takes the component-converge path, where convergeSecrets only writes secrets that appear as missing probe components — so batch --inference-project/--inference-wif-provider flags passed during a re-converge are silently ignored for existing openai repos. This contradicts the dual-route behavior this PR's own doc update (docs/guides/getting-started/repo-management.md) describes: passing the GCP flags for the batch still applies them to such repositories. The fresh-install path (Install) is unaffected and is what the added test (TestConverge_OpenAIRoute_GCPFlagsStillApplyWhenGiven) covers, so the re-converge gap is untested.

Suggested fix: Make secretsPresent (and convergeSecrets) route-aware: treat GCP flags as an independent write set from the openai required-secret probe list, or probe leftover GCP secrets when the flags are supplied so an existing openai repo can still receive the Vertex pair on re-converge.

  • docs/guides/infrastructure/openai-workload-identity.md (file-level): Line 274 · [medium] stale-guide-instructions

Route C step 1 tells users to re-run fullsend github setup <owner/repo> (without --inference-provider openai) before step 3 sets FULLSEND_OPENAI_API_KEY. On a repository with no GCP credentials, that command fails today: setup requires the GCP pair unless --inference-provider openai is passed, and that flag in turn requires the OpenAI secret (or a complete inference.openai trio) to already exist. Neither Route B nor Route C states that GCP is not required — Section D in the same document does give the correct secret-first-then---inference-provider openai order, but a reader following Route C's numbered steps can easily miss it and get stuck in a chicken-and-egg sequence. This falls short of issue #7481's stated documentation goal that "the OpenAI guide's Route B/C sections say GCP is not required."

Suggested fix: In Route C, distinguish resyncing an existing (Vertex-provisioned) repo from openai-only enrollment; for the latter, point to Section D's order (set FULLSEND_OPENAI_API_KEY first, then run fullsend github setup <owner/repo> --inference-provider openai), and state explicitly in Route B/C that GCP is not required when --inference-provider openai is used.

  • docs/guides/infrastructure/infrastructure-reference.md (file-level): Line 415 · [medium] missing-documentation

Under the GitLab CI/CD variables table, FULLSEND_GCP_PROJECT_ID/FULLSEND_GCP_WIF_PROVIDER are still listed as required when inference is configured, with no note that they are omitted when inference_provider is openai, and OPENAI_API_KEY is described only as optional rather than as the required credential for GitLab repos on the openai route (per requiredSecretsForRoute). This PR updated the equivalent GitHub per-repo secrets section a few dozen lines above, but the GitLab section on the same page was missed.

Suggested fix: Update the GitLab CI/CD variables section to note that the GCP pair is not required when inference_provider is openai, and that OPENAI_API_KEY is required on GitLab in that case.

  • docs/cli/repos.md (file-level): Line 94 · [low] incomplete-cli-reference

The prose description of repos install's "Manifest add" behavior lists the per-repo flag overrides written to the manifest entry (--inference-region, --fullsend-ref, --mint-url, --allowed-remote-resources, --runtime) but omits --inference-provider, even though this PR adds that flag and it is documented in the flags table further down the same page.

Suggested fix: Add --inference-provider to the parenthetical list of per-repo overrides at line 94.

  • docs/guides/getting-started/repo-management.md (file-level): Line 28 · [medium] contradictory-prerequisites

The prerequisites section and installation notes still assert unconditionally that GCP WIF provisioning must be completed before running repos install, while a new section added later in the same file by this PR states that repositories using inference_provider: openai need neither GCP secret. The getting-started path now contradicts the openai-only path documented lower on the same page.

Suggested fix: Qualify the prerequisites/installation-notes language to state that GCP WIF provisioning is required only for repositories using the default Vertex inference route.

  • docs/cli/repos.md (file-level): Line 114 · [low] incomplete-cli-reference

The repos install flags table documents --inference-project/--inference-wif-provider without noting they are optional when the provider is openai, unlike docs/cli/github.md, which this PR updated with "(required unless the provider is openai)" wording for the equivalent flags.

Suggested fix: Update the --inference-project/--inference-wif-provider descriptions in the repos install flags table to note they are required only for the vertex route.

  • docs/guides/getting-started/repo-management.md (file-level): Line 137 · [low] incomplete-manifest-reference

The prose describing the none sentinel in repos.yaml lists fullsend_ref, mint_url, mint_mode, config_base.source, config_base.sha256 but omits runtime and the newly added inference_provider, both of which accept the none sentinel per manifest.go/manifest_test.go.

Suggested fix: Add runtime and inference_provider to the list of string fields supporting the none sentinel.

@fullsend-ai-review fullsend-ai-review Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See the review comment for full details.

Comment thread internal/cli/github.go Resolved
Comment thread internal/harness/harness.go Resolved
Comment thread internal/harness/harness.go Resolved

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review with three models (Codex, Claude, Grok via pi) plus a local run. Only MEDIUM+ findings not already raised by qodo or the review bot are posted inline.

Local verification at 1d03f99 (agents#1502 harnesses, every GCP variable unset): fullsend run triage --runtime pi --model openai/gpt-5.6-luna created the sandbox, the agent exited 0 and validation passed. So the no-GCP OpenAI path works end to end in the runner, which covers part of the unchecked "real agent run" box. The same env with --runtime claude fails late; see the harness.go thread.

Not inline (the file is outside the diff), MEDIUM: fullsend agent new (internal/agentnew/render.go buildHarness) still emits {Src: "${GOOGLE_APPLICATION_CREDENTIALS}", Dest: "/tmp/.gcp-credentials.json"} without OptionalForOpenAI. So a harness scaffolded for an openai-only repo fails the same way the fleet harnesses did before agents#1502. Consider setting the flag there, plus a render test.

qodo's leftover partial GCP pair (#1/#2), the batch-flag bypass of the OpenAI credential check (#3) and the preset-route timing (#4) still reproduce at this head, so they are not re-posted. qodo's drift-revert finding (#5) no longer applies, because CheckFileContentDrift skips both config layers.

Comment thread internal/repos/converge.go Resolved
Comment thread internal/repos/converge.go Resolved
Comment thread internal/cli/github.go Resolved
Comment thread internal/harness/harness.go Resolved
Comment thread .github/actions/setup-gcp/action.yml Resolved
Comment thread .github/actions/setup-gcp/action.yml Resolved
Comment thread .github/workflows/reusable-code.yml Resolved
Comment thread .github/workflows/reusable-fix.yml Resolved
@waynesun09 waynesun09 added the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Sep 30, 2026
…ps audience

Workload identity providers provisioned by fullsend allow the audiences
https://iam.googleapis.com/<provider> and fullsend-mint. The runner asked
GitHub for a token with the //iam.googleapis.com/ form, which STS rejects.
Request the https form, keep the scheme-relative form for the STS
exchange and the credential file, and carry the https form in
FULLSEND_GCP_OIDC_URL for the refresh loop.

Add an optional OnSubjectToken callback so the caller can mask the
fetched token in its log.

The test's fake OIDC endpoint now refuses audiences outside the
provider's allowed list, so the old request form fails the test.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
On GitHub Actions, fullsend run now picks Vertex credentials in this order:
both FULLSEND_GCP_PROJECT_ID and FULLSEND_GCP_WIF_PROVIDER set prepares
WIF and overrides GOOGLE_APPLICATION_CREDENTIALS; neither set keeps an
existing credential file from an earlier step; one set is an error. A
kept file must be non-empty JSON with a type, and an external_account
file must read its token from credential_source.file: a URL source is
unreachable from the sandbox and its headers would copy the runner's
request token into it. The run prints which source it used.

An OpenAI parent can dispatch Vertex sub-agents, so it also prepares WIF
when both inputs are set; a failure there is a warning. The dummy
runtimes resolve to a "none" provider and need no GCP inputs. The
Vertex-parent path stays fail-closed.

Runtime resolution and credential preparation now run before harness
env validation and expansion. A harness whose host_files require
${GOOGLE_APPLICATION_CREDENTIALS} failed validation on every GitHub
Actions Vertex run, because the workflow no longer sets it before
fullsend run starts.

The fetched and refreshed GitHub OIDC tokens are masked in the Actions
log. validateVertexGCPCredentials rejects an empty file.
prepareGitHubWIF is a package variable so runAgent tests can stub the
exchange.

Functional tests no longer receive the GCP inputs, so they keep using
their prepared credential file.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
Describe how fullsend run picks Vertex credentials on GitHub Actions,
what local and GitLab runs need, and that an OpenAI run with the GCP
inputs still gets credentials for Vertex sub-agents. Align the OpenAI
workload identity guide and the workflow contract note with that.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
… credentials

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
A run whose parent does not use Vertex now clears
GOOGLE_APPLICATION_CREDENTIALS when the file fails validation, so an optional
host file cannot copy a URL-source credential into the sandbox. The validator
caps the file size and checks impersonation sources. Dummy runtimes skip WIF.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
…file

A non-Vertex run whose required ${GOOGLE_APPLICATION_CREDENTIALS} mount has
no file now fails before the pre-script. The credential check follows
source_credentials only for impersonation files.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
@github-actions github-actions Bot removed the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Sep 30, 2026
@waynesun09 waynesun09 added the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Sep 30, 2026
…e set

Behaviour tests run dummy agents on fleet harnesses whose GCP credential
mount is required. Skipping WIF for dummy runs left that mount empty and
failed the triage scenarios. Dummy runs now follow the OpenAI-parent rule.

Assisted-by: Claude (fix, review)
Signed-off-by: Wayne Sun <gsun@redhat.com>
@github-actions github-actions Bot removed the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Sep 30, 2026
@waynesun09 waynesun09 added the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Sep 30, 2026
@waynesun09

Copy link
Copy Markdown
Member

@amastbau I'm done; the branch is yours again. Head is fbfd01d: the review fixes, a merge of current main (the branch still pinned OpenShell 0.0.116), and one follow-up so behaviour-test dummy runs keep WIF. The PR description lists the live runs on Vertex and OpenAI, with both claude and pi. Behaviour, e2e and CI are green. functional-tests is failing in its "Create ephemeral repo and fixture" hook before any case runs; other PRs hit the same failure right now.

agents#1502 also needs current agents main merged before it can land. Its providers/vertex-ai.yaml still has _NOOP_VERTEX_AI, which OpenShell 0.1.2 rejects. Until it lands, OpenAI-only repos on the fleet harnesses stop before the pre-script with GOOGLE_APPLICATION_CREDENTIALS is empty; mark the mount optional or provide a credential file.

This branch was successfully deployed

1 active and 1 inactive deployments
site-preview — fbfd01db Deployed Sep 30, 2026 by github-actions[bot]
dev — fbfd01db Deployed Sep 30, 2026 by waynesun09 via run / behaviour #3577
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) risk/elevated PR risk: elevated

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants