Skip to content

refactor(#8065)!: remove org-specific config models and loading - #8088

Merged
ggallen merged 5 commits into
mainfrom
agent/8065-remove-org-config
Oct 6, 2026
Merged

ggallen merged 5 commits into
mainfrom
agent/8065-remove-org-config

Conversation

@fullsend-ai-coder

Copy link
Copy Markdown
Contributor

Summary

This removes the per-org configuration model and everything that read it: the config structs, parsers, interfaces and mode detection, plus their consumers in runtime, run, lock/eval, harness and repos code. Per-org installation is deprecated (ADR 0044). A config.yaml in the old per-org format now fails with a clear error. Before this change, its org-only keys would have been silently dropped.

Related Issue

Closes #8065 (part of #8063)

Changes

  • internal/config
    • Removed orgConfig, NewOrgConfig, ParseOrgConfig, DispatchConfig, RepoDefaults, RepoConfig, InferenceConfig, NewPerRepoConfigFromOrg, IsOrgMode and OrgConfigReader/OrgConfigWriter.
    • LoadConfig/LoadConfigWriter now reject a config.yaml with top-level dispatch, repos or defaults keys (errPerOrgConfig). YAML that isn't a mapping still reports a parsing config.yaml error.
    • Organization-valued fields that are still valid per-repo, such as create_issues.allow_targets.orgs, are kept.
  • internal/runtime: removed ResolveFromConfig.
  • internal/cli
    • fullsend run no longer has an org path for runtime or per-agent settings (resolveBackendFromConfigData, isOrgConfigData and runConfig.orgData are gone).
    • Comments were updated in orgconfig.go and reconcilestatus.go.
  • internal/repos
  • Tests
    • Tests of validators shared with per-repo config (agent entries, status notifications, create_issues) now run through per-repo config instead of being dropped.
    • Org-only tests were removed.
    • Org loading tests now check that the file is rejected.
    • Agent and run tests that wrote org-shaped fixtures now write per-repo ones.
  • Docs: removed OrgConfig and ResolveFromConfig references from docs/architecture.md and docs/contributing/runtime-implementation.md.

Some code is now unreachable but left in place, for example !ok branches after a PerRepoConfigReader assertion and the tryLoadOrgConfig/requireOrgConfig aliases. Per the issue's sequencing, the final sweep belongs to #8070.

Breaking change: a per-org .fullsend/config.yaml is now rejected with an error instead of being loaded.

Testing

  • make lint passes (stage changes first, then run). The pre-commit hook-repo fetch failed in the sandbox because of TLS, so the hooks were run directly: gofmt, go vet ./..., trailing-whitespace/EOF, lint-docs-links, lint-broken-symlinks, lychee offline and gitlint all passed.
  • Tests added/updated for new or modified logic. New tests cover rejection of per-org files in LoadConfig/LoadConfigWriter (each key, and alongside a base layer), backendFromConfigFile, tryLoadFullsendConfig/requireFullsendConfig, loadAgentConfig, runAgentSet and KillSwitchActive.
  • go build ./... and go test pass for internal/config, internal/cli, internal/runtime, internal/repos, internal/harness and internal/harnessdispatch. The exception is TestStartFetchService_* in internal/cli, which fails in the sandbox because local dials are denied. fetch, sandbox, security and telemetry also fail in the sandbox for network reasons, and none of them are touched here.

Checklist

  • PR title follows Conventional Commits (correct type, ! for breaking changes)
  • Commits are signed off (DCO) — human and human-directed agent sessions only (autonomous agent commit; exempt)
  • I wrote this contribution myself and can explain all changes in it (autonomous agent contribution)

🤖 Generated with Claude Code


Closes #8065

Post-script verification

  • Branch is not main/master (agent/8065-remove-org-config)
  • Secret scan passed (gitleaks — 6cbd29593d396828305081b72bc2bdc76c4a79eb..HEAD)
  • PR body secret scan passed (gitleaks — no-git)

@fullsend-ai-coder fullsend-ai-coder Bot added the ready-for-review Triggers review agent dispatch label Oct 5, 2026
@fullsend-ai-coder
fullsend-ai-coder Bot requested a review from a team October 5, 2026 19:23
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 7:31 PM UTC · Completed 7:54 PM UTC

Commit: 1d2ab0b · View workflow run →

Runtime: pi · Model: openai/gpt-6.1-sol → gpt-6.1-sol · Effort: high · Cost: $4.31

@github-actions

github-actions Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

Site preview

Preview: https://49c642c3-site.fullsend-ai.workers.dev

Commit: e6f4d8318c36da91f3678f690db00e44a3260373

@codecov

codecov Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 95.00000% with 2 lines in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
internal/repos/manifest.go 71.42% 1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

@fullsend-ai-review fullsend-ai-review Bot added the risk/moderate PR risk: moderate label Oct 5, 2026
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

Risk Assessment: moderate (2/5)

Details

Extensive removal of deprecated org configuration across 42 files with strong test coverage and zero protected path edits balances elevated git history churn and coupling in core CLI and config modules, maintaining moderate risk.

Previous run

Risk Assessment: moderate (2/5)

Details

Extensive removal of deprecated org configuration across 42 files with strong test coverage and zero protected path edits balances elevated git history churn and coupling in core CLI and config modules, maintaining moderate risk.

Previous run (2)

Risk Assessment: moderate (2/5)

Details

Extensive removal of deprecated org configuration across 42 files with strong test coverage and zero protected path edits balances elevated git history churn and coupling in core CLI and config modules, maintaining moderate risk.

Previous run (3)

Risk Assessment: moderate (2/5)

Details

Extensive removal of deprecated org configuration across 39 files with strong test coverage and zero protected path edits balances elevated git history churn and coupling in core CLI and config modules, maintaining moderate risk.

Previous run (4)

Risk Assessment: moderate (2/5)

Details

Extensive removal of deprecated org configuration across 37 files with strong test coverage and zero protected path edits balances elevated git history churn and coupling in core CLI and config modules, maintaining moderate risk.

Previous run (5)

Risk Assessment: moderate (2/5)

Details

The large diff primarily removes deprecated org configuration and includes extensive test changes with no protected-path edits; elevated churn and regression history in core modules balance the clearly authorized scope, yielding moderate risk.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

Looks good to me

Previous run

Review

Findings

High

  • [fail-open] internal/cli/run.go:507 — The fail-closed guard checks only config.yaml. tryLoadOrgConfig returns nil when that overlay is absent, without reading config.base.yaml. A supported base-only allowed_remote_resources: [] policy therefore falls back to the default allowlist during agent-source resolution, permitting an agents-repository harness download. A malformed base-only configuration likewise permits fetching before loadRunConfig reports its error.
    Remediation: Load both configuration layers strictly before agent-source resolution, using the same layered loading approach as loadLockConfig. Treat configuration as absent only when neither layer exists, preserve inherited deny-all lists, and propagate load failures before any fetch. Add base-only deny-all and malformed-base regression tests.

  • [fail-open] internal/cli/evalmeasure.go:234 — The fail-closed fallback distinguishes load failure from absence using only config.yaml. With that overlay absent, tryLoadOrgConfig does not read config.base.yaml, so a base-only allowed_remote_resources: [] policy is replaced by the default allowlist passed to the agents-repository measurement fetch. Malformed base-only configurations are also ignored.
    Remediation: Read the layered configuration without requiring config.yaml to exist first. Preserve the resolved base-layer allowlist and deny remote fetching on any configuration load failure; use defaults only when both layers are absent. Add tests for base-only deny-all and malformed-base configurations.


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
Previous run (2)

Review

Findings

High

  • [fail-open] internal/cli/lock.go:159 — loadLockConfig returns nil when config.yaml is absent without examining config.base.yaml, even though the layered loader supports base-only configurations. This ignores a base-layer allowed_remote_resources: [] and base-layer read/parse errors. For a local harness whose local base inherits a remote ancestor, locking passes the default allowlist into composition; the immediate-URL-base check does not reload configuration for this indirect case. Remote ancestors under the default prefixes can therefore be fetched and locked despite the base-layer deny-all policy.
    Remediation: Use the layered loader before deciding configuration is absent, returning nil only when neither layer exists. Preserve inherited empty allowlists and propagate base-layer read/parse errors. Add lock tests for base-only deny-all and malformed configurations, including a local base with a remote ancestor.

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
Previous run (3)

Review

Findings

Medium

  • [logic-error] internal/cli/lock.go:442 — Locking treats a rejected configuration as absent. tryLoadOrgConfig converts the new per-org rejection into nil, so fullsend lock --all discovers only local harnesses. With one valid local-only harness and an existing lock entry for a config-only agent, processing can succeed without loading the config again; pruning then deletes the config-only entry, saves the reduced lock file, and returns success. An unsupported configuration must cause an error rather than be interpreted as removal of its registered agents.
    Remediation: In runLockAll and lockOneAgent, distinguish missing config from an existing config that fails to load, and propagate the rejection before mutating the lock file. Add a regression test with an org-shaped config, a valid local-only harness, and an existing config-only agent lock entry; assert failure and unchanged lock contents.

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
Previous run (4)

Review

Findings

Medium

  • [fail-open] internal/cli/orgconfig.go:26 — The newly rejected org-shaped configuration is converted to nil by tryLoadFullsendConfig. For a previously loadable config containing allowed_remote_resources: [], runAgent consequently substitutes DefaultAllowedRemoteResources and can fetch the first-party harness before loadRunConfig reports the configuration error. The later error prevents agent execution, but the explicit deny-all policy has already been bypassed for remote fetching.
    Remediation: Distinguish absent configuration from loading failures and propagate rejection errors to runAgent before agent-source resolution or remote composition. Add a regression test asserting that rejected configuration with an empty allowlist causes no harness fetch.

Low

  • [stale-doc] docs/guides/user/bugfix-workflow.md:132 — The guide still describes a configurable CI-failure retry cap using defaults.max_implementation_retries in config.yaml. This PR removes the org configuration model defining that setting, and the loader now rejects top-level defaults.
    Remediation: Remove the reference to the deleted setting, or document the currently supported retry-control mechanism if one exists.

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
Previous run (5)

Review

Findings

Medium

  • [fail-open] internal/cli/orgconfig.go:26 — Rejecting an existing org-shaped configuration makes tryLoadFullsendConfig return nil. In evalMeasureFetchContext, nil leaves DefaultAllowedRemoteResources in effect rather than preserving an explicit allowed_remote_resources: []. The measurement-manifest fallback can consequently fetch from the agents repository despite the previous deny-all setting. This is a bounded policy replacement, not arbitrary remote access.
    Remediation: Distinguish absent configuration from existing-file load errors in evalMeasureFetchContext. Propagate rejection errors or disable remote fetching on those errors. Add a regression test for an org-shaped config with allowed_remote_resources: [].

  • [stale-doc] docs/guides/infrastructure/private-repositories.md:62 — The private-repositories guide still provides deprecated per-org configuration instructions using defaults and repos. Those keys are now explicitly rejected by the configuration loaders. The supported per-repo roles example immediately above remains valid.
    Remediation: Remove or clearly retire the per-org bullet and YAML snippet, retaining the supported per-repo roles instructions. Coordinate broader per-org guide cleanup with the documentation-removal stage.

Low

  • [code-organization] internal/repos/discover.go:12 — After removal of the discovery implementation, discover.go contains only MarshalWithHeader, while discover_test.go contains serialization and readWorkflowRef tests. The discovery filenames no longer describe their contents.
    Remediation: Consider moving MarshalWithHeader and its tests alongside manifest serialization, and readWorkflowRef tests alongside status tests.

  • [naming-convention] internal/cli/run_test.go:1279 — The migrated TestResolveAgentSource fixtures still use the variable name orgCfg for values constructed with config.NewPerRepoConfig.
    Remediation: Rename these fixture variables to cfg or perRepoCfg.

  • [stale-doc] docs/guides/user/customizing-agents.md:534 — The disabling-agents section says the role can remain in defaults.roles. That is a removed per-org configuration field; the supported per-repo field is roles.
    Remediation: Replace defaults.roles with roles in this sentence.

  • [stale-doc] docs/guides/dev/cli-internals.md:743 — The implementation table describes internal/config/config.go as containing Org/repo config structures, although this PR removes the org-specific configuration structures.
    Remediation: Change the table description to Per-repo config structures or Configuration structures and validation.

  • [stale-doc] docs/reference/config-reference.md:196 — The per-repo configuration reference calls allowed_remote_resources an org-level list. With per-org configuration loading removed, repo-level or config-level more accurately identifies this fallback policy.
    Remediation: Replace org-level list with config-level list in this paragraph without changing the documented allowlist semantics.


Labels: This removes deprecated installation configuration and its consumers, matching the repository’s maintenance/install labeling conventions.


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[bot]

This comment was marked as outdated.

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review fullsend-ai-review Bot added type/chore Maintenance and housekeeping tasks component/install CLI install and app setup labels Oct 5, 2026
@ggallen

ggallen commented Oct 5, 2026

Copy link
Copy Markdown
Member

/fs-fix

Rebase. Squash commits. Fix all review issues. Reply to and resolve all threads.

1 similar comment
@ggallen

ggallen commented Oct 5, 2026

Copy link
Copy Markdown
Member

/fs-fix

Rebase. Squash commits. Fix all review issues. Reply to and resolve all threads.

@fullsend-ai-coder

fullsend-ai-coder Bot commented Oct 5, 2026 •

Copy link
Copy Markdown
Contributor Author

🤖 Finished Fix · ✅ Success · Started 8:52 PM UTC · Completed 9:02 PM UTC

Commit: 1d2ab0b · View workflow run →

Runtime: claude · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.96

Per-org installation mode is deprecated (ADR 0044). Remove the org
config model and every reader of it:

- config: drop orgConfig, NewOrgConfig, ParseOrgConfig, the
  DispatchConfig/RepoDefaults/RepoConfig/InferenceConfig types,
  NewPerRepoConfigFromOrg, IsOrgMode and the OrgConfigReader/Writer
  interfaces. LoadConfig and LoadConfigWriter now reject a config.yaml
  with top-level dispatch, repos or defaults keys instead of loading it.
- runtime: drop ResolveFromConfig.
- cli: drop the org branch of `fullsend run` runtime and agent-settings
  resolution (resolveBackendFromConfigData, isOrgConfigData).
- repos: delete the org-to-per-repo Migrate helper (no callers since
  #8064) and the org-only discovery code; move MarshalWithHeader into
  manifest.go and the readWorkflowRef tests into status_test.go, and
  remove the now-misnamed discover.go/discover_test.go.

Validators shared with per-repo config (agent entries, status
notifications, create_issues incl. AllowTargets.Orgs) keep their tests,
now exercised through per-repo config.

Changes from review feedback on PR #8088:
- eval-measure fails closed (denies the agents-repo fetch) when an
  existing config.yaml cannot be loaded, so a rejected org-shaped config
  with allowed_remote_resources: [] no longer reverts to the default
  allowlist; regression test added.
- TestResolveAgentSource fixtures renamed from orgCfg to cfg.
- Docs: drop the per-org roles snippet from the private-repositories
  guide and update stale per-org wording in customizing-agents,
  cli-internals and config-reference.

BREAKING CHANGE: a .fullsend config.yaml in the per-org format (top-level
dispatch, repos or defaults) is now rejected with an error instead of
being loaded. Use a per-repo .fullsend/config.yaml.

Note: pre-commit could not fetch hook repositories (sandbox TLS
failure); gofmt, go vet ./... and whitespace checks were run directly.
Unrelated TestStartFetchService_* and TestPostScriptEnv_NoSchema* tests
fail in the sandbox with and without this change.

Closes #8065

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@fullsend-ai-coder

Copy link
Copy Markdown
Contributor Author

🔧 Fix agent — iteration 1 (human-triggered)

Addressed all 7 review findings (1 medium fail-open fix with regression test, 1 medium doc fix, 5 low items). The branch was already up to date with main, so the rebase was a no-op, and the PR is squashed to a single commit. Thread replies and resolution are left to the post-script. Pre-commit could not fetch hook repos in the sandbox, so gofmt, go vet and whitespace checks were run directly. Some internal/cli tests (TestStartFetchService_, TestPostScriptEnv_NoSchema) fail in the sandbox both with and without this change.

Fixed (9):

  1. Rebase onto the target branch: No-op: origin/main is already an ancestor of the branch, so no rebase ran. rebased_onto_target is set per the no-op procedure.
  2. Squash commits: Rewrite base was origin/main (merge-base 6cbd295). The original commit plus the review-fix work were combined into exactly one commit via soft reset, with a message covering the whole PR including the review changes.
  3. fail-open: rejected or unreadable config reverts to default remote allowlist in eval-measure (internal/cli/evalmeasure.go): evalMeasureFetchContext now sets an empty allowlist (deny remote fetch) when a config.yaml exists but fails to load, instead of falling back to defaults. Added a regression test covering an org-shaped config with allowed_remote_resources: [] and a malformed config.
  4. stale-doc: private-repositories guide still documents per-org defaults/repos config (docs/guides/infrastructure/private-repositories.md): Removed the per-org bullet and YAML snippet; the supported per-repo roles example remains.
  5. code-organization: discover.go and discover_test.go names no longer match contents (internal/repos/discover.go): Moved MarshalWithHeader into manifest.go next to Marshal, moved its test into manifest_test.go and the readWorkflowRef tests into status_test.go, and deleted discover.go and discover_test.go.
  6. naming-convention: orgCfg fixture variables in TestResolveAgentSource (internal/cli/run_test.go): Renamed the orgCfg fixtures built with NewPerRepoConfig to cfg.
  7. stale-doc: defaults.roles mentioned in disabling-agents section (docs/guides/user/customizing-agents.md): Replaced defaults.roles with roles.
  8. stale-doc: cli-internals table describes config.go as Org/repo config structures (docs/guides/dev/cli-internals.md): Changed the description to Per-repo config structures.
  9. stale-doc: allowed_remote_resources described as an org-level list (docs/reference/config-reference.md): Replaced org-level with config-level in the paragraph, leaving the allowlist semantics unchanged.

Disagreed (1):

  1. Reply to and resolve all review threads: The fix agent cannot post comments or resolve threads; the post-script handles PR mutations. The two open inline threads (discover.go organization and the orgCfg rename) are addressed by this commit and can be resolved.

Tests: passed

CI inspections (8):

  1. test (passing, success)
  2. build (passing, success)
  3. codecov/patch (passing, success)
  4. playback (transient-infra, failure): The end-to-end playback scenario failed waiting on an issue-open workflow run in the test repository, which concluded with failure. It does not touch code changed in this PR. Recommend rerunning the E2E Tests workflow on the new head.
  5. behaviour (transient-infra, cancelled): Cancelled after the playback job failed. Rerun with E2E Tests.
  6. functional-tests (transient-infra, cancelled): Run was cancelled, not failed. Recommend a rerun.
  7. commit-lint (transient-infra, cancelled): Run was cancelled, not failed. Recommend a rerun.
  8. test-sandbox-darwin (transient-infra, cancelled): Run was cancelled, not failed. Recommend a rerun.
Decision points
  • Handle the squash when the PR was already a single commit (alternatives: Leave the original commit and add a separate fix commit, Fold the fixes into one squashed commit; rationale: The goal of the squash request is a single-commit PR, so the fixes were folded into one commit with a message covering the whole PR.)

Next steps:

  • /fs-review — request a re-review of the changes
  • /fs-fix <your instruction> — run another fix pass with specific guidance
  • Push commits directly — review re-runs automatically on push
    Updated by fullsend fix agent

@fullsend-ai-coder
fullsend-ai-coder Bot force-pushed the agent/8065-remove-org-config branch from 1d2ab0b to a6704c9 Compare October 5, 2026 21:02
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 9:15 PM UTC · Completed 9:33 PM UTC

Commit: a6704c9 · View workflow run →

Runtime: pi · Model: openai/gpt-6.1-sol → gpt-6.1-sol · Effort: high · Cost: $3.44

fullsend-ai-review[bot]

This comment was marked as outdated.

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-coder

fullsend-ai-coder Bot commented Oct 5, 2026 •

Copy link
Copy Markdown
Contributor Author

🤖 Finished Fix · ✅ Success · Started 9:41 PM UTC · Completed 9:51 PM UTC

Commit: a6704c9 · View workflow run →

Runtime: claude · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.62

Address review feedback on PR #8088:
- runAgent no longer treats an existing but unloadable config.yaml as
  absent; the load error is returned before agent-source resolution so an
  explicit allowed_remote_resources deny-all cannot be replaced by the
  default allowlist. Adds a regression test and updates three tests that
  asserted the old fallback error.
- Remove the reference to the deleted defaults.max_implementation_retries
  setting from the bugfix workflow guide.

Note: pre-commit could not fetch hook repos (TLS); gofmt, go vet and
git diff --check were run directly. Unrelated pre-existing failures in
TestStartFetchService_* and TestPostScriptEnv_NoSchema* also fail on the base.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@fullsend-ai-coder

Copy link
Copy Markdown
Contributor Author

🔧 Fix agent — iteration 2 (bot-triggered)

Addressed both findings: runAgent now fails closed on an existing-but-rejected config.yaml before any remote resolution (with regression test), and the stale max_implementation_retries doc reference was removed. The new test passes. Other cli test failures (fetch service and post-script env tests) also fail on the base commit and are unrelated sandbox issues. Pre-commit could not fetch hook repos, so gofmt, go vet and diff checks were run directly.

Fixed (2):

  1. fail-open: rejected org-shaped config is treated as absent, so the default remote allowlist replaces an explicit deny-all before the load error surfaces (internal/cli/run.go): runAgent now returns the config load error when config.yaml exists but cannot be loaded, before agent-source resolution or any remote fetch. Added a regression test (offline, unregistered agent, deny-all config) that fails without the fix, and updated three existing malformed-config tests that asserted the old fallback error message.
  2. stale-doc: bugfix workflow guide references the removed defaults.max_implementation_retries setting (docs/guides/user/bugfix-workflow.md): Removed the reference to the deleted setting; the text now says the retry is bounded by a fixed cap. No other supported retry-control mechanism exists in the repo.

Tests: passed

CI inspections (3):

  1. test (passing, success): Passed on the previous head.
  2. build (transient-infra, cancelled): Cancelled, likely superseded by a newer run; no failure evidence. Recommend rerunning if it does not restart on the new push.
  3. gate (transient-infra, cancelled): Cancelled, likely superseded by a newer run. Recommend rerunning if it does not restart on the new push.

Conflict update: skipped; strategy none; forge MERGEABLE; target main@6cbd29593d396828305081b72bc2bdc76c4a79eb

Decision points
  • Scope of the fail-closed change (alternatives: Also change lock.go callers of the best-effort loader, Limit to runAgent as cited by the review; rationale: The finding names runAgent only; the eval-measure path was already fail-closed, and lock commands were left untouched to keep the diff minimal.)

Next steps:

  • /fs-review — request a re-review of the changes
  • /fs-fix <your instruction> — run another fix pass with specific guidance
  • Push commits directly — review re-runs automatically on push
    Updated by fullsend fix agent

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 10:46 PM UTC · Completed 10:55 PM UTC

Commit: 13e6076 · View workflow run →

Runtime: pi · Model: openai/gpt-6.1-sol → gpt-6.1-sol · Effort: high · Cost: $1.76

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-coder

fullsend-ai-coder Bot commented Oct 5, 2026 •

Copy link
Copy Markdown
Contributor Author

🤖 Finished Fix · ✅ Success · Started 10:57 PM UTC · Completed 11:07 PM UTC

Commit: 13e6076 · View workflow run →

Runtime: claude · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.75

runAgent and evalMeasureFetchContext checked only config.yaml when
deciding whether config was absent, so a base-only
allowed_remote_resources: [] fell back to the default allowlist and a
malformed config.base.yaml was ignored before fetching. Both now load the
layered config strictly via loadLockConfig (absent only when neither
layer exists) and deny remote fetching on any load failure.

Pre-commit could not fetch hook repos (TLS failure); ran gofmt, go vet
and git diff --check directly. Unrelated pre-existing failures in
TestStartFetchService_* and TestPostScriptEnv_NoSchema* also fail on the
base branch in this sandbox.

Addresses #8088

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@fullsend-ai-coder

Copy link
Copy Markdown
Contributor Author

🔧 Fix agent — iteration 5 (bot-triggered)

Both findings shared one root cause (config absence decided from config.yaml alone). Run and eval-measure now reuse the layered strict loader already used for locking. Tests in internal/cli pass except six pre-existing sandbox-environment failures (TestStartFetchService_, TestPostScriptEnv_NoSchema) that also fail on the unmodified branch. pre-commit could not fetch hook repos; gofmt, go vet and diff checks were run directly.

Fixed (2):

  1. run.go: fail-closed guard checks only config.yaml, so a base-only deny-all or malformed config.base.yaml falls back to the default allowlist (internal/cli/run.go): runAgent now loads the layered config strictly via loadLockConfig (absent only when neither layer exists) and returns any load error before agent-source resolution. Added a malformed base-only regression test and updated existing error-message assertions; base-only deny-all is covered by TestLoadLockConfig_BaseLayer for the shared loader.
  2. evalmeasure.go: fail-closed fallback checks only config.yaml, so a base-only deny-all is replaced by the default allowlist and malformed base config is ignored (internal/cli/evalmeasure.go): evalMeasureFetchContext now uses the same strict layered loader; the allowlist from either layer is preserved and any load failure yields an empty allowlist (remote fetch denied). Extended the fetch-context test with base-only deny-all and malformed base-only cases.

Tests: passed

CI inspections (1):

  1. test (passing, success): Project CI checks (test, build, functional, behaviour, playback, commit-lint, DCO, codecov/patch) all passing on the prior head.

Conflict update: skipped; strategy none; forge MERGEABLE; target main@n/a

Next steps:

  • /fs-review — request a re-review of the changes
  • /fs-fix <your instruction> — run another fix pass with specific guidance
  • Push commits directly — review re-runs automatically on push
    Updated by fullsend fix agent

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 5, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:09 PM UTC · Completed 11:18 PM UTC

Commit: e6f4d83 · View workflow run →

Runtime: pi · Model: openai/gpt-6.1-sol → gpt-6.1-sol · Effort: high · Cost: $1.47

@fullsend-ai-review
fullsend-ai-review Bot dismissed stale reviews from themself October 5, 2026 23:18

Superseded by updated review

@fullsend-ai-review fullsend-ai-review Bot added the ready-for-merge All reviewers approved — ready to merge label Oct 5, 2026
@ggallen
ggallen enabled auto-merge October 5, 2026 23:19
@ggallen
ggallen added this pull request to the merge queue Oct 5, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Oct 5, 2026
@ggallen
ggallen added this pull request to the merge queue Oct 6, 2026
Merged via the queue into main with commit 5160548 Oct 6, 2026
54 checks passed
@ggallen
ggallen deleted the agent/8065-remove-org-config branch October 6, 2026 00:41

This branch was successfully deployed

1 active and 1 inactive deployments
site-preview — e6f4d831 Deployed Oct 5, 2026 by github-actions[bot]
dev — e6f4d831 Deployed Oct 5, 2026 by fullsend-ai-coder[bot] via behaviour #15422
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

component/install CLI install and app setup needs-human Agent loop needs human intervention ready-for-merge All reviewers approved — ready to merge ready-for-review Triggers review agent dispatch risk/moderate PR risk: moderate type/chore Maintenance and housekeeping tasks

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Remove org-specific configuration models and loading

1 participant