feat(#4976): store mint and inference settings in per-repo config by fullsend-ai-coder[bot] · Pull Request #5976 · fullsend-ai/fullsend

fullsend-ai-coder · 2026-08-06T12:11:55Z

Summary

Add mint_url, inference_provider, inference_project, inference_region, and inference_wif_provider fields to per-repo config schema (ADR 0069 Decision 1)
Implement accessor methods (ConfigMintURL(), ConfigInferenceProvider(), etc.) on PerRepoConfigReader with layered fallback (overlay → base → code defaults) and corresponding setters on PerRepoConfigWriter
Wire fullsend github setup to populate mint/inference values in config: flag values go into config.yaml (no preset) or overlay (with preset), while the base layer stays identical to the fetched preset

Design

Schema fields: Flat YAML keys on perRepoConfig (mint_url, inference_provider, inference_project, inference_region, inference_wif_provider). Field names are prefixed to avoid colliding with the org-mode detection keys in IsPerRepoYAML (dispatch, inference).

Accessor pattern: Follows the existing scalar fallback convention (e.g., ConfigRuntime(), ConfigForge()): local value → parent → empty string. Code defaults in perRepoDefaults return "" for all mint/inference fields — callers that need a fallback (e.g., the CLI) use DefaultMintURL.

Installer wiring: When --config (preset) is provided, flag-specified values go into the overlay; unspecified flags inherit from the base layer via the accessor chain. The changedFlags map records which CLI flags were explicitly set.

Backward compatibility: Existing repo variable/secret writes (FULLSEND_MINT_URL, FULLSEND_GCP_REGION, FULLSEND_GCP_PROJECT_ID, FULLSEND_GCP_WIF_PROVIDER) are preserved so existing workflow templates continue to read from vars/secrets when config fields are absent.

Testing

Config accessor fallback: overlay → base → defaults for all 5 new fields
Marshal roundtrip: YAML serialize → parse → accessor match
Empty overlay omits inherited fields (no leak from base to overlay marshal)
IsPerRepoYAML correctly identifies config with new fields as per-repo (not org)
buildPresetOverlay correctly writes only flag-changed values
Validation rejects invalid inference_provider values
Existing tests continue to pass unchanged

Closes #4976

Post-script verification

Branch is not main/master (agent/4976-mint-inference-config)
Secret scan passed (gitleaks — fac93d227b4e30eaa553b16d059eac6624ec9fc2..HEAD)
PR body secret scan passed (gitleaks — no-git)

fullsend-ai-review · 2026-08-06T12:13:06Z

🤖 Finished Review · ✅ Success · Started 12:13 PM UTC · Completed 12:28 PM UTC
Commit: 9492ea9 · View workflow run →

codecov · 2026-08-06T12:18:09Z

Codecov Report

❌ Patch coverage is 90.24390% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/cli/github.go	89.39%	4 Missing and 3 partials ⚠️
internal/config/interfaces.go	88.09%	5 Missing ⚠️

📢 Thoughts on this report? Let us know!

fullsend-ai-review · 2026-08-06T12:28:33Z

Review

Findings

Low

[data-completeness] internal/cli/github.go:302 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because changedFlags won't include those flag names. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[detection-robustness] internal/config/interfaces.go:646 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant now relies on org configs always emitting dispatch, repos, or defaults (all have no omitempty tag). Sound today with test coverage confirming both directions.

Previous run

Review

Findings

Medium

[stale reference to removed identifier] internal/cli/run_test.go:3575 — Comment still lists "inference" as an org-only detection key: // No "defaults"/"dispatch"/"repos"/"inference" key, so this parses as per-repo config. The PR removes "inference" from the IsPerRepoYAML detection list (it is now a shared key between org and per-repo configs). This comment is now stale and misleading.
Remediation: Update the comment to remove "inference" from the list of org-only keys, matching the updated IsPerRepoYAML logic.

Low

[data-completeness] internal/cli/github.go:301 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because changedFlags won't include those flag names. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[detection-robustness] internal/config/interfaces.go:645 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant now relies on org configs always emitting dispatch, repos, or defaults (all have no omitempty tag). Sound today with test coverage confirming both directions.

Previous run (2)

Review

Findings

Low

[data-completeness] internal/cli/github.go:301 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because changedFlags won't include those flag names. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[detection-robustness] internal/config/interfaces.go:641 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant relies on org configs always emitting dispatch (which has no omitempty tag). Sound today with test coverage confirming both directions.
[Input Validation] internal/config/config.go:799 — Validate() checks inference provider against the ValidProviders() whitelist but does not validate mint_url (HTTPS scheme) or wif_provider (format). CLI-level checks exist (validateMintURLHTTPS, validateWIFProvider) but configs loaded from YAML bypass them. No runtime consumers exist today.
[spelling-consistency] internal/config/interfaces.go:499 — Comment uses British spelling initialises; the codebase convention is American English initializes.

Info

[Dual-Write Consistency] internal/cli/github.go:383 — Mint URL and inference region are stored in both config.yaml and repo variables for backward compatibility. The effective-value resolution logic correctly falls back to code defaults. Subsequent updates via fullsend github set only update repo variables, not config.yaml. Tracked for migration in Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977.

Previous run (3)

Review

Findings

Medium

[edge-case] internal/cli/github.go:746 — In the no-preset path, SetMintURL, SetInferenceProvider, and SetInferenceRegion are called unconditionally with CLI flag defaults (DefaultMintURL, "vertex", "global"). By contrast, SetInferenceProject and SetInferenceWIFProvider are guarded by if != "" checks. This asymmetry means the generated config.yaml always contains locally-set values for mint URL, provider, and region, which would shadow a subsequently added base layer. In the preset path (buildPresetOverlay), the changedFlags pattern correctly handles this distinction.

Low

[data-completeness] internal/cli/github.go:301 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because the empty flag values are skipped. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[detection-robustness] internal/config/interfaces.go:641 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant relies on org configs always emitting dispatch (which has no omitempty tag). Sound today with test coverage confirming both directions.
[Input Validation] internal/config/config.go:796 — Validate() checks inference provider against the ValidProviders() whitelist but does not validate mint_url (HTTPS scheme) or wif_provider (format). CLI-level checks exist (validateMintURLHTTPS, validateWIFProvider) but configs loaded from YAML bypass them. No runtime consumers exist today.
[naming-alignment] internal/cli/github.go:148 — The --inference-provider flag defaults to "vertex" while perRepoDefaults.ConfigInferenceProvider() returns "". This is a deliberate separation between CLI-flag defaults (user convenience) and config-layer defaults (layered fallthrough), but may cause confusion about the effective default.
[documentation-accuracy] docs/guides/infrastructure/layered-config-reference.md:88 — Code defaults table correctly shows inference.provider as "" (empty), but does not note that fullsend github setup writes "vertex" (the CLI flag default) into config.yaml for non-preset installs, making the effective default differ from the code default.

Previous run (4)

Review

Findings

Low

[data-completeness] internal/cli/github.go:301 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because the empty flag values are skipped. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[detection-robustness] internal/config/interfaces.go:657 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant relies on org configs always emitting dispatch (which has no omitempty tag). Sound today with test coverage confirming both directions.
[naming-coherence] internal/config/config.go:110 — Org-mode uses InferenceConfig (single Provider field) while per-repo uses PerRepoInferenceConfig (with Provider, Project, Region, WIFProvider). The PerRepo prefix disambiguates clearly, and the structural difference is justified by the richer per-repo configuration requirements.

Previous run (5)

Review

Findings

Medium

[missing-doc] docs/guides/infrastructure/layered-config-reference.md:72 — Per-field merge rules table and code defaults table do not document mint_url or the inference block (inference.provider, inference.project, inference.region, inference.wif_provider). These new fields follow scalar-override merge semantics and have empty-string code defaults, but are missing from both tables.
Remediation: Add rows for mint_url and inference subfields to both the per-field merge rules table and the code defaults reference table.
[stale-doc] docs/cli/github.md:45 — CLI flags table for github setup is missing the new --inference-provider flag (default: vertex).
Remediation: Add a row for --inference-provider to the flags table.

Low

[data-completeness] internal/cli/github.go:300 — When re-running setup with reused secrets (reuseProject=true / reuseWIF=true), the generated config.yaml will not contain inference.project or inference.wif_provider because the empty flag values are skipped. This mirrors the existing reuse behavior for secrets and is benign today (no production code reads from config.yaml), but creates a latent gap for the config-only migration (Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977).
[data-exposure] internal/cli/github.go:296 — GCP project ID and WIF provider resource name are now written to plaintext config.yaml (mode 100644) alongside encrypted GitHub repo secrets. These are identifiers (not credentials) and are already visible through workflow files, so the exposure increase is minimal.
[naming-coherence] internal/config/config.go:106 — Org-mode uses InferenceConfig (single Provider field) while per-repo uses PerRepoInferenceConfig (with Provider, Project, Region, WIFProvider). The PerRepo prefix disambiguates clearly, and the structural difference is justified by the richer per-repo configuration requirements.
[detection-robustness] internal/config/interfaces.go:620 — IsPerRepoYAML removes inference from org-only detection keys. The safety invariant relies on org configs always emitting dispatch (which has no omitempty tag). Sound today with test coverage confirming both directions.

Previous run (6)

Review

Findings

Critical

[logic-error] internal/cli/github.go:363 — Shim workflow templates (shim-per-repo.yaml) reference ${{ vars.FULLSEND_MINT_URL }}, ${{ vars.FULLSEND_GCP_REGION }}, ${{ secrets.FULLSEND_GCP_WIF_PROVIDER }}, and ${{ secrets.FULLSEND_GCP_PROJECT_ID }}. The reusable dispatch workflow declares mint_url and gcp_region as required: true inputs. The setup function no longer writes these repo variables/secrets — for new installs, workflows will receive empty strings for critical auth parameters, breaking inference authentication. No code in the dispatch workflow reads config.yaml for mint/inference values.
Remediation: Either (a) continue writing repo variables/secrets alongside config.yaml, or (b) update shim/scaffold workflow templates to read from config.yaml, or (c) wire a config-reader step in reusable workflows.
[scope-violation] internal/cli/github.go:363 — Issue feat(config): store mint and inference settings in layered config by default (ADR 0069) #4976 explicitly requires: "Keep existing repo variable/secret writes and workflow reads functional as fallback — no breaking change." The "Not in this issue" section explicitly excludes "removing or deprecating repo variables/secrets." The prior review noted the dual-write as intentional. This commit removes all writes, directly contradicting the issue's authorization.
Remediation: Restore dual-write: continue writing FULLSEND_MINT_URL, FULLSEND_GCP_REGION as repo variables and FULLSEND_GCP_PROJECT_ID, FULLSEND_GCP_WIF_PROVIDER as repo secrets alongside config.yaml fields.

Medium

[data-exposure] internal/cli/github.go:296 — GCP project ID and WIF provider resource name move from encrypted GitHub secrets to plaintext config.yaml committed with mode 100644. While these are identifiers (not credentials), they reveal the target GCP project and exact WIF identity pool path. Shim workflows still reference secrets, creating a split-brain state.
Remediation: Evaluate whether GCP project ID and WIF provider can remain secrets-only. If moved to config, update shim templates and document the visibility change.
[dead-code] internal/cli/github.go:210 — reuseProject and reuseWIF check for existing repo secrets, but secrets are no longer written. When both flags are omitted on re-run, the reuse check sets reuseProject/reuseWIF=true but these only control log messages. The config.yaml path unconditionally skips empty inferenceProject/inferenceWIFProvider, so reused secret values are NOT populated into config.yaml.
Remediation: When reusing existing secrets, read the values and populate them into config.yaml, or require flags to always be provided.
[missing-doc] docs/guides/infrastructure/layered-config-reference.md:72 — Per-field merge rules table and code defaults table do not document mint_url or inference.* fields.
Remediation: Add rows for mint_url and inference.provider, inference.project, inference.region, inference.wif_provider to both tables.
[stale-doc] docs/cli/github.md:45 — CLI flags table missing --inference-provider. Does not mention config.yaml persistence.
Remediation: Add --inference-provider to the flags table with default vertex.
[stale-doc] docs/guides/infrastructure/infrastructure-reference.md:235 — Per-Repo Mode section lists FULLSEND_GCP_PROJECT_ID, FULLSEND_GCP_WIF_PROVIDER as secrets and FULLSEND_MINT_URL, FULLSEND_GCP_REGION as variables. These are no longer written by the installer.
Remediation: Update to reflect config.yaml storage. Only FULLSEND_PER_REPO_INSTALL remains.
[stale-doc] docs/guides/getting-started/operations.md:18 — Shows updating FULLSEND_GCP_PROJECT_ID and FULLSEND_GCP_REGION as repo secrets/variables. Now stored in config.yaml.
Remediation: Update to reflect config.yaml storage for new installations.
[stale-doc] docs/guides/dev/cli-internals.md:249 — Phase 6 diagram shows secrets/variables no longer written by the installer.
Remediation: Update Phase 6 diagram to remove the four entries for per-repo mode.

Low

[missing-validation] internal/config/config.go:779 — Validate() validates inference Provider but not mint_url (HTTPS scheme) or wif_provider (format) when loaded from YAML. Pre-existing pattern but worth noting for defense-in-depth.
[detection-weakening] internal/config/interfaces.go:623 — IsPerRepoYAML no longer treats inference as org-mode indicator. Safe: org configs always contain dispatch/repos/defaults (non-omitempty). Test coverage validates both directions.
[config-schema-change] internal/config/config.go:531 — Purely additive schema change with omitempty. Backward compatible.
[config-file-permission] internal/cli/github.go:354 — config.yaml committed with mode 100644. Previous model stored some values as GitHub secrets.
[incomplete-doc] docs/ADRs/0069-ready-made-configuration-presets.md:102 — Decision 2 gives MintURL() and InferenceProvider() as example accessor names, but implementation uses ConfigMintURL() and ConfigInferenceProvider().
[incomplete-doc] docs/guides/getting-started/configuring-github.md:72 — States "creates files, secrets and variables" but now only creates files and guard variable.

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 (7)

Review

Findings

Medium

[error-handling-gap] internal/cli/github.go:318 — When a preset base layer is provided via --config, the overlay config built by buildPresetOverlay is marshaled and written without calling Validate(). In the non-preset path (line 306), Validate() is explicitly called before Marshal(). User-provided flag values like --inference-type openai are written to config.yaml without validation in the preset path.
Remediation: Add a Validate() call on the overlay config before marshaling, matching the non-preset path.
[missing-doc] docs/guides/infrastructure/layered-config-reference.md:71 — The per-field merge rules table and code defaults reference table do not document the new config fields (mint_url, inference block with type, project, region, wif_provider). This is the canonical layered-config reference.
Remediation: Add rows for mint_url and inference.* fields to both tables.
[stale-doc] docs/cli/github.md:45 — CLI flags table is missing the new --inference-type flag and doesn't mention that mint/inference flag values are now persisted to .fullsend/config.yaml.
Remediation: Add --inference-type to the flags table and note about config file persistence.

Low

[naming-coherence] internal/config/config.go:114 — Per-repo inference uses Type (YAML: type) while org-mode uses Provider (YAML: provider) for the same concept (inference backend name). Naming divergence is a deliberate structural choice given the richer nested struct shape.
[scope-alignment] internal/cli/github.go:357 — Mint/inference values are written to both config files and repo variables/secrets. Dual-write is intentional per issue feat(config): store mint and inference settings in layered config by default (ADR 0069) #4976 for backward compatibility.
[architectural-coherence] internal/config/interfaces.go:619 — IsPerRepoYAML removed inference from org-detection keys, now relying on dispatch/repos/defaults always being present in org configs (non-omitempty). Invariant holds today with test coverage.
[comment-style] internal/config/defaults.go:41 — Default getter comments add (empty) suffix, slightly more verbose than established concise pattern.
[setter-style] internal/config/interfaces.go:462 — New inference setters use single-letter parameter names (t, p, r, w); the codebase convention is mixed (some setters use single letters, some use full words).
[incomplete-doc] docs/ADRs/0069-ready-made-configuration-presets.md:102 — Decision 2 gives InferenceProvider() as example accessor name, but implementation uses ConfigInferenceType() etc.
[incomplete-doc] docs/guides/getting-started/configuring-github.md:62 — Setup examples don't explain where config values are stored post-installation.
[yaml-schema-evolution] internal/config/config.go:530 — Per-repo config.yaml schema gains optional mint_url and inference block. Purely additive, backward-compatible change.
[detection-heuristic-change] internal/config/interfaces.go:619 — IsPerRepoYAML no longer treats inference as an org-mode indicator. Safe because org configs always contain dispatch/repos/defaults.

Labels: PR adds mint and inference settings to per-repo config schema

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 (8)

Review

Findings

Medium

[architectural-coherence] internal/cli/github.go:298 — SetInferenceProvider("vertex") is hard-coded in the non-preset setup path with no corresponding --inference-provider CLI flag. Unlike the other four mint/inference settings, users cannot specify this value via the command line. Functionally correct today (only one valid provider), but creates flag-set asymmetry.
Remediation: Add --inference-provider flag defaulting to "vertex", and include "inference-provider" in buildPresetOverlay's flagNames.
[missing-doc] docs/guides/infrastructure/layered-config-reference.md:71 — The per-field merge rules table and code defaults reference table do not document the five new config fields (mint_url, inference_provider, inference_project, inference_region, inference_wif_provider). This is the canonical layered-config reference.
Remediation: Add entries for all five fields as scalar-override fields with empty-string code defaults.
[stale-doc] docs/cli/github.md:45 — CLI flags table documents --mint-url, --inference-project, etc. but does not mention that values are now persisted to .fullsend/config.yaml fields.
Remediation: Note that flag values are written to the per-repo config file during setup.

Low

[data-exposure] internal/cli/github.go:300 — GCP project ID and WIF provider resource name are now in both .fullsend/config.yaml (plaintext) and GitHub repo secrets. These are resource identifiers (not credentials), and dual storage is likely transitional for backward compatibility.
[input-validation] internal/config/config.go:773 — Validate() checks inference_provider against ValidProviders() but does not validate mint_url (HTTPS), inference_wif_provider (format), or other new fields when loaded from disk. CLI validates at the flag level, but configs loaded via LoadConfig bypass those checks.
[design-smell] internal/config/config.go:206 — Comment claims field names are "prefixed to avoid colliding with org-mode detection keys" but names avoid collision by specificity (inference_provider vs inference), not by a shared prefix.
[incomplete-doc] docs/ADRs/0069-ready-made-configuration-presets.md:71 — Decision 1 doesn't specify the exact YAML field names implementing the decision.
[incomplete-doc] docs/guides/getting-started/configuring-github.md:62 — Setup examples don't explain where config values are stored post-installation.
[pattern-inconsistency] internal/config/interfaces.go:375 — Section comment block explaining the fallback pattern is more verbose than equivalent sections for existing scalar getters.
[comment-style] internal/config/interfaces.go:346 — ADR reference in interface getter doc comments inconsistent with other getters in the file.
[comment-style] internal/config/defaults.go:40 — Default getter comments more verbose than established concise pattern.
[incomplete-doc] docs/cli/github.md:56 — --config flag description doesn't explain how flag-specified mint/inference values interact with presets (overlay behavior).

Labels: PR modifies per-repo config schema and CLI setup for mint/inference settings

ifireball · 2026-08-09T08:18:43Z

/fs-fix lets try to group the inference setting in YAML under a single "inference" top level key and add a "type" filed that defaults to and may contain only "vertex" for now but may contain other types of inference providers in the future

also deal with the AI review comments

fullsend-ai-coder · 2026-08-09T08:19:59Z

🤖 Finished Fix · ✅ Success · Started 8:19 AM UTC · Completed 8:37 AM UTC

Commit: 9492ea9 · View workflow run →

Restructure per-repo inference config from flat YAML keys (inference_provider, inference_project, etc.) to a nested inference block (inference.type, inference.project, etc.). Add --inference-type CLI flag defaulting to "vertex" to replace the hard-coded provider value. Fix comment verbosity and update IsPerRepoYAML to allow shared inference key. Addresses human review feedback and bot review findings on #5976

fullsend-ai-coder · 2026-08-09T08:37:42Z

🔧 Fix agent — iteration 1 (human-triggered)

Restructured per-repo inference config from flat YAML keys to a nested inference: block with type/project/region/wif_provider fields. Added --inference-type CLI flag (default: vertex) replacing the hard-coded provider value. Fixed comment style issues. Disagreed with 5 doc/informational findings as out of scope.

Fixed (7):

group inference settings under nested YAML key (internal/config/config.go): Replaced flat inference_provider/project/region/wif_provider YAML keys with a nested inference: block containing type/project/region/wif_provider. Added PerRepoInferenceConfig struct, updated MarshalYAML, Validate, and IsPerRepoYAML.
add type field defaulting to vertex (internal/config/config.go): Added Type field to PerRepoInferenceConfig that maps to inference.type in YAML. Renamed ConfigInferenceProvider/SetInferenceProvider to ConfigInferenceType/SetInferenceType across interface, implementations, and defaults.
hard-coded SetInferenceProvider vertex without CLI flag (internal/cli/github.go): Added --inference-type CLI flag (default: vertex) and wired it through setupConfig, runGitHubSetupPerRepo, and buildPresetOverlay. Replaces the hard-coded SetInferenceProvider(vertex) call.
design-smell: inaccurate comment about field naming (internal/config/config.go): Replaced incorrect comment about field names being prefixed to avoid collisions. The nested inference struct now inherently avoids collision.
comment-style: verbose default getter comments (internal/config/defaults.go): Simplified default method comments to match established concise pattern.
pattern-inconsistency: verbose section comment (internal/config/interfaces.go): Removed verbose section comment block for mint/inference getters to match existing scalar getter sections.
comment-style: ADR reference in interface comments (internal/config/interfaces.go): Removed ADR reference from PerRepoConfigReader interface comments to match other getters in the file.

Disagreed (5):

missing-doc: layered-config-reference.md not updated: Documentation updates for the new config fields are out of scope for this code restructuring commit.
stale-doc: docs/cli/github.md not updated: Documentation updates are out of scope for this code restructuring.
incomplete-doc: ADR 0069 and getting-started guide: Documentation updates for ADRs and guides are out of scope for this config restructuring.
data-exposure: dual storage of project ID and WIF provider: Informational finding. Dual storage is intentional for backward compatibility with existing workflow templates.
input-validation: missing HTTPS/pattern validation for mint_url and wif_provider: CLI already validates at the flag level. Adding duplicate validation in Validate() is a separate concern.

Tests: passed

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 · 2026-08-09T08:38:46Z

🤖 Finished Review · ✅ Success · Started 8:38 AM UTC · Completed 8:54 AM UTC

Commit: 6d95a49 · View workflow run →

ifireball · 2026-08-09T09:10:08Z

/fs-fix this is a user-visible config change, docs are in scope! also deal with other new AI comments

ifireball · 2026-08-10T05:53:38Z

/fs-fix per review comments: use "provider" instead of "type" for inference; do not write values to variables but keep the reads in place for backwards compatibility;

Deal with other comments you did not deal with yet as well.

fullsend-ai-coder · 2026-08-10T05:54:59Z

🤖 Finished Fix · ✅ Success · Started 5:54 AM UTC · Completed 6:13 AM UTC

Commit: 6d95a49 · View workflow run →

fullsend-ai-coder · 2026-08-10T06:13:43Z

🔧 Fix agent — iteration 2 (human-triggered)

Addressed 4 of 8 review findings with code changes: renamed inference Type to Provider for naming coherence with org-mode config, removed variable/secret writes (values now in config.yaml only), added Validate() call for preset overlay path, and fixed setter parameter names. Disagreed with 4 informational/already-addressed findings.

Fixed (4):

naming-coherence: per-repo inference uses Type while org-mode uses Provider (internal/config/config.go): Renamed PerRepoInferenceConfig.Type to Provider (YAML tag: provider) to align with org-mode InferenceConfig naming. Updated all accessor methods (ConfigInferenceType to ConfigInferenceProvider), setters (SetInferenceType to SetInferenceProvider), and CLI flag (--inference-type to --inference-provider) across config, interfaces, defaults, CLI, and all test files.
scope-alignment: dual-write of mint/inference values to config and variables/secrets (internal/cli/github.go): Removed FULLSEND_MINT_URL and FULLSEND_GCP_REGION variable writes and FULLSEND_GCP_PROJECT_ID/FULLSEND_GCP_WIF_PROVIDER secret writes from per-repo setup. Values are now stored in config.yaml only (ADR 0069). Kept secret existence reads at top of function for backward compatibility with existing installations.
error-handling-gap: preset overlay path missing Validate() before Marshal() (internal/cli/github.go): Added Validate() call on the overlay config before Marshal() in the preset path, matching the non-preset code path that already validates before marshaling.
setter-style: single-letter parameter names in inference setters (internal/config/interfaces.go): Changed SetMintURL(u), SetInferenceProvider(t), SetInferenceProject(p), SetInferenceRegion(r), SetInferenceWIFProvider(w) to use full-word parameter names: mintURL, provider, project, region, wifProvider.

Disagreed (4):

architectural-coherence: IsPerRepoYAML detection heuristic change: Informational finding. The detection heuristic change is correct and has test coverage. No code change needed.
yaml-schema-evolution: per-repo config gains optional mint_url and inference block: Informational finding about backward-compatible schema evolution. No code change needed.
detection-heuristic-change: IsPerRepoYAML no longer treats inference as org indicator: Informational finding. Safe because org configs always contain dispatch/repos/defaults. Has test coverage.
comment-style: default getter comments add (empty) suffix: Already addressed in iteration 1. No further change needed.

Tests: passed

Decision points

Chose to keep the per-repo guard variable (FULLSEND_PER_REPO_INSTALL) write while removing other variable/secret writes (alternatives: Remove all variable writes including the guard var, Keep all writes for backward compat; rationale: The guard variable is a functional marker that other workflows check; removing it would break per-repo detection. Mint/inference values are now in config.yaml and do not need variable/secret storage.)
Kept backward-compat secret existence reads at top of runGitHubSetupPerRepo (alternatives: Remove the reads since values are in config, Add config.yaml fallback for the reads; rationale: Human explicitly instructed to keep reads for backward compatibility. Existing installations may have secrets from before this change, and the reads allow re-runs without re-providing flags.)

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}

Address review feedback on PR #5976: - Rename PerRepoInferenceConfig.Type to Provider (YAML: provider) to align with org-mode InferenceConfig naming (review: naming-coherence, human instruction: use "provider" instead of "type") - Remove FULLSEND_MINT_URL, FULLSEND_GCP_REGION variable writes and FULLSEND_GCP_PROJECT_ID, FULLSEND_GCP_WIF_PROVIDER secret writes from per-repo setup; values are now stored in config.yaml only (human instruction: do not write values to variables) - Keep secret/variable existence reads for backward compatibility - Add Validate() call on preset overlay before Marshal(), matching the non-preset code path (review: error-handling-gap) - Use full-word parameter names in inference setters instead of single letters (review: setter-style) - Rename --inference-type flag to --inference-provider Addresses review feedback on #5976

fullsend-ai-review · 2026-08-10T06:14:58Z

🤖 Finished Review · ✅ Success · Started 6:14 AM UTC · Completed 6:36 AM UTC

Commit: 3669336 · View workflow run →

ifireball · 2026-08-10T11:17:28Z

/fs-fix per review comments: restore dual-write of mint/inference repo variables/secrets alongside config.yaml (undo the attempt to stop writing them — workflows still require those vars until #5870/#4977). Keep reading existing secrets for backward compatibility. Address the other open AI review findings as well.

fullsend-ai-coder · 2026-08-10T11:18:45Z

🤖 Finished Fix · ✅ Success · Started 11:18 AM UTC · Completed 11:27 AM UTC

Commit: 3669336 · View workflow run →

Re-add writing FULLSEND_MINT_URL and FULLSEND_GCP_REGION as repo variables, and FULLSEND_GCP_PROJECT_ID / FULLSEND_GCP_WIF_PROVIDER as repo secrets during per-repo setup. Existing workflow templates still reference these vars/secrets until #5870 / #4977 migrate reads to config.yaml. Reading existing secrets for backward compatibility is preserved via the reuseProject / reuseWIF checks. Update tests to assert dual-write behavior: variables and secrets are set alongside config.yaml values, partial-reuse tests verify only the non-reused secret is written. Addresses review feedback on #5976

fullsend-ai-coder · 2026-08-10T11:27:09Z

🔧 Fix agent — iteration 3 (human-triggered)

Restored dual-write of mint/inference repo variables and secrets alongside config.yaml per human instruction. This fixes the critical logic-error and scope-violation findings - workflow templates that reference vars/secrets will continue to receive values. The reuseProject/reuseWIF dead-code finding is also resolved. Updated 4 test functions to assert dual-write behavior. Disagreed with 4 low-severity informational findings as pre-existing patterns or out-of-scope.

Fixed (4):

logic-error: workflow templates reference vars/secrets that are no longer written (internal/cli/github.go): Restored dual-write of FULLSEND_MINT_URL and FULLSEND_GCP_REGION as repo variables, and FULLSEND_GCP_PROJECT_ID / FULLSEND_GCP_WIF_PROVIDER as repo secrets (conditional on reuseProject/reuseWIF). Workflow templates continue to read from vars/secrets until Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870/chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977 migrate to config-only reads.
scope-violation: removing repo variable/secret writes contradicts issue feat(config): store mint and inference settings in layered config by default (ADR 0069) #4976 requirements (internal/cli/github.go): Restored the original dual-write behavior per issue feat(config): store mint and inference settings in layered config by default (ADR 0069) #4976 requirement to keep existing repo variable/secret writes functional as fallback.
dead-code: reuseProject/reuseWIF checks had no effect after secret writes were removed (internal/cli/github.go): With dual-write restored, reuseProject and reuseWIF are meaningful again - they gate whether FULLSEND_GCP_PROJECT_ID and FULLSEND_GCP_WIF_PROVIDER secrets are written.
test assertions incorrect after removing secret/variable writes (internal/cli/github_test.go): Updated TestRunGitHubSetupPerRepo, TestRunGitHubSetupPerRepo_ReusesExistingSecrets, TestRunGitHubSetupPerRepo_PartialReuse_ProjectOnly, and TestRunGitHubSetupPerRepo_PartialReuse_WIFOnly to assert dual-write behavior.

Disagreed (4):

data-exposure: GCP project ID and WIF provider move from secrets to plaintext config.yaml: These values are identifiers, not credentials. With dual-write restored, they continue to live in encrypted GitHub secrets for workflow consumption. The config.yaml copy enables the new config-based accessor path (ADR 0069). The visibility change is documented in the PR body.
missing-validation: Validate() does not check mint_url (HTTPS) or wif_provider format: Pre-existing pattern. mint_url is already validated by validateMintURLHTTPS() at the CLI layer before reaching config. wif_provider format is validated by validateWIFProvider(). Adding redundant validation to Validate() is out of scope for this PR.
detection-weakening: IsPerRepoYAML no longer treats inference as org-mode indicator: Review itself notes this is safe - org configs always contain dispatch/repos/defaults (non-omitempty), and test coverage validates both directions. No action needed.
config-file-permission: config.yaml committed with mode 100644: Config holds identifiers, not credentials. Credentials remain in GitHub secrets via dual-write. 100644 is the standard mode for config files committed via the GitHub API.

Tests: passed

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 · 2026-08-10T11:28:27Z

🤖 Finished Review · ✅ Success · Started 11:28 AM UTC · Completed 11:46 AM UTC

Commit: cace2d2 · View workflow run →

ifireball · 2026-08-11T08:01:39Z

/fs-fix Fix the spelling nit only: in internal/config/interfaces.go, change British initialises to American initializes to match codebase convention.

Disagree/close (already handled — do not reopen):

data-completeness (reuse secrets / config.yaml gap) — known until Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870/chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977
detection-robustness (IsPerRepoYAML / inference) — intentional, tested
Input Validation (Validate mint_url/wif) — CLI already validates; out of scope
Dual-Write Consistency (info) — tracked in Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870/chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977

Do not change behavior beyond the spelling fix.

fullsend-ai-coder · 2026-08-11T08:02:45Z

🤖 Finished Fix · ✅ Success · Started 8:02 AM UTC · Completed 8:08 AM UTC

Commit: 7f6bf6b · View workflow run →

Change British "initialises" to American "initializes" in the ensureInference doc comment to match codebase convention. Addresses review feedback on #5976

fullsend-ai-coder · 2026-08-11T08:08:44Z

🔧 Fix agent — iteration 6 (human-triggered)

Fixed spelling nit (initialises → initializes) per human instruction. Disagreed with 4 prior review findings: all are either known gaps tracked in existing issues, intentional design decisions, or out of scope for this config-plumbing PR.

Fixed (1):

British spelling initialises in ensureInference comment (internal/config/interfaces.go): Changed British "initialises" to American "initializes" in the ensureInference doc comment to match codebase convention

Disagreed (4):

data-completeness (reuse secrets / config.yaml gap): Known gap tracked in Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977 — out of scope for this PR, which only adds the config schema fields and accessors
detection-robustness (IsPerRepoYAML / inference key detection): Intentional design: org-mode always emits dispatch/repos/defaults (non-omitempty), so the remaining markers are sufficient. This is tested and documented in the IsPerRepoYAML godoc
Input Validation (Validate mint_url / wif): The CLI already validates mint_url via validateMintURLHTTPS when non-empty; further validation of WIF provider format is out of scope for this config-plumbing PR
Dual-Write Consistency (info): The dual-write of repo vars/secrets alongside config.yaml is intentional backward compatibility documented in code comments — migration to config-only reads is tracked in Absorb workflow-side agent setup into fullsend run (ADR 0072 Option C) #5870 / chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977

Tests: passed

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 · 2026-08-11T08:09:50Z

🤖 Finished Review · ✅ Success · Started 8:09 AM UTC · Completed 8:26 AM UTC

Commit: 9b3b0f4 · View workflow run →

ifireball · 2026-08-11T08:37:28Z

/fs-fix Fix the medium stale-reference finding only: in internal/cli/run_test.go (~line 3575), update the comment that still lists "inference" as an org-only IsPerRepoYAML detection key. Remove "inference" from that list so it matches current logic (inference is a shared key between org and per-repo configs).

Do not reopen or change code for previously closed findings (data-completeness, detection-robustness, Input Validation, spelling, dual-write).

fullsend-ai-coder · 2026-08-11T08:38:45Z

🤖 Finished Fix · ✅ Success · Started 8:38 AM UTC · Completed 8:45 AM UTC

Commit: 9b3b0f4 · View workflow run →

The comment in TestSetupStatusNotifier_PerRepoConfigYAML listed "inference" as an org-only IsPerRepoYAML detection key, but inference is now a shared key used in per-repo configs too. Update the comment to match the current detection logic which only checks "defaults", "dispatch", and "repos". Addresses review feedback on #5976

fullsend-ai-coder · 2026-08-11T08:45:43Z

🔧 Fix agent — iteration 7 (human-triggered)

Fixed the stale-reference finding: removed "inference" from the org-only detection key comment in run_test.go to match current IsPerRepoYAML logic.

Fixed (1):

stale comment lists inference as org-only IsPerRepoYAML detection key (internal/cli/run_test.go): Removed "inference" from the comment listing org-only detection keys in TestSetupStatusNotifier_PerRepoConfigYAML. The IsPerRepoYAML function only checks "defaults", "dispatch", and "repos" — inference is now a shared key used in per-repo configs.

Tests: passed

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 · 2026-08-11T08:46:58Z

🤖 Finished Review · ✅ Success · Started 8:46 AM UTC · Completed 9:00 AM UTC

Commit: 9cbf231 · View workflow run →

fullsend-ai-retro · 2026-08-11T10:47:01Z

🤖 Finished Retro · ✅ Success · Started 10:47 AM UTC · Completed 11:04 AM UTC

Commit: 9cbf231 · View workflow run →

fullsend-ai-retro · 2026-08-11T11:04:01Z

Retro: PR #5976 — store mint and inference settings in per-repo config

Workflow overview: Issue #4976 was triaged (5 triage runs), coded (1 code run producing the initial PR), then went through 8 review/fix cycles over 5 days (Aug 6–11) before ifireball merged it. The PR added +1054/−26 lines across 11 files implementing ADR 0069 Decision 1.

What went well

Review agent caught a critical breaking change. In Fix 3, the human instructed the fix agent to stop writing repo variables/secrets. The review agent correctly flagged this as [critical] — shim workflows still depend on those values, and the linked issue explicitly required preserving them. This caught a breaking change before it reached main.
High precision at medium+ severity. The review agent's critical findings had 100% precision (2/2 true positives), and medium findings had ~88% precision (7/8). The agent's most impactful contributions were at these severity tiers.
Fix agent executed reliably. All 8 fix iterations completed successfully, each producing correct code for the given instruction. The agent handled a complex rebase (100 commits behind main, 3 conflicts) without issues.

Rework analysis

Of the 9 fix iterations, the root causes break down as:

4 fixes due to design refinements not specified in the original issue (nested YAML structure, naming convention, defaults layer architecture, and undoing Fix 3)
2 fixes for valid review findings the initial code agent missed (docs update, stale test comment)
2 trivial fixes (spelling nit, rebase)
1 silent/no-op fix (docs instruction that produced no visible output)

The costliest rework was the Fix 3 → Fix 4 ping-pong: the human instructed the agent to remove dual-write of repo variables, contradicting the issue's explicit requirement to preserve them. The fix agent followed the instruction, the review agent caught the scope violation, and a second fix was needed to undo the change. This wasted one full cycle (~30 min of agent compute + human review time).

Existing issues covering observed patterns

Review agent re-raising dismissed findings: The same "data-completeness" and "detection-robustness" low findings were flagged in nearly every review iteration despite being dismissed. This is well-covered by existing issues: fullsend#3515, fullsend#1672, fullsend#5265, agents#721. This PR provides additional evidence — the human had to close the same findings 3+ times each.
Code agent not including documentation: The initial PR omitted doc updates for user-visible config schema changes, costing at least 1 fix cycle. Covered by fullsend#1733 and agents#452.
Low-severity noise: ~61% false positive rate at low severity (7/18 true positives). Covered by agents#370, agents#688, agents#150.

Autonomy assessment

The review agent demonstrated strong bug-catching capability (the critical dual-write finding alone justified its involvement). However, the human's primary value-add was design-level decisions — YAML structure, naming conventions, defaults layer architecture — that the review agent cannot replicate. The current collaborative model (agent reviews + human design oversight) is appropriate for this class of feature work.

Proposals filed

Fix agent should flag when /fs-fix instructions contradict linked issue requirements (in fullsend-ai/agents)

fullsend-ai-coder Bot requested a review from a team as a code owner August 6, 2026 12:11

fullsend-ai-coder Bot added the ready-for-review Triggers review agent dispatch label Aug 6, 2026

fullsend-ai-coder Bot assigned ifireball Aug 6, 2026

This comment was marked as outdated.

Sign in to view

fullsend-ai-review Bot added requires-manual-review Review requires human judgment component/install CLI install and app setup labels Aug 6, 2026

This comment was marked as outdated.

Sign in to view

fullsend-ai-review Bot added component/mint Token mint and cross-boundary credentials and removed requires-manual-review Review requires human judgment labels Aug 9, 2026

This comment was marked as outdated.

Sign in to view

ifireball mentioned this pull request Aug 10, 2026

chore(install): migrate existing installs from repo vars/secrets to config (ADR 0069, optional) #4977

Open

6 tasks

fullsend-ai-review Bot added ready-for-merge All reviewers approved — ready to merge and removed requires-manual-review Review requires human judgment labels Aug 11, 2026

fix: use American spelling "initializes" in config comment

9b3b0f4

Change British "initialises" to American "initializes" in the ensureInference doc comment to match codebase convention. Addresses review feedback on #5976

github-actions Bot deployed to site-preview August 11, 2026 08:10 View deployment

This comment was marked as outdated.

Sign in to view

fullsend-ai-review Bot added requires-manual-review Review requires human judgment and removed ready-for-merge All reviewers approved — ready to merge labels Aug 11, 2026

github-actions Bot deployed to site-preview August 11, 2026 08:47 View deployment

fullsend-ai-review Bot approved these changes Aug 11, 2026

View reviewed changes

Comment thread internal/cli/github.go

Comment thread internal/config/interfaces.go

fullsend-ai-review Bot added ready-for-merge All reviewers approved — ready to merge and removed requires-manual-review Review requires human judgment labels Aug 11, 2026

ifireball approved these changes Aug 11, 2026

View reviewed changes

ifireball enabled auto-merge August 11, 2026 10:33

ifireball added this pull request to the merge queue Aug 11, 2026

Merged via the queue into main with commit 99b2723 Aug 11, 2026
29 of 31 checks passed

ifireball deleted the agent/4976-mint-inference-config branch August 11, 2026 10:44

fullsend-ai-retro Bot mentioned this pull request Aug 11, 2026

Fix agent should flag when /fs-fix instructions contradict linked issue requirements fullsend-ai/agents#740

Open

qodo-code-review Bot mentioned this pull request Sep 1, 2026

feat(harness): add a max_cost_usd budget cap for agent runs #6588

Open

Uh oh!

Conversation

fullsend-ai-coder Bot commented Aug 6, 2026

Summary

Design

Testing

Post-script verification

Uh oh!

fullsend-ai-review Bot commented Aug 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Aug 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

fullsend-ai-review Bot commented Aug 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review

Findings

Low

Review

Findings

Medium

Low

Review

Findings

Low

Info

Review

Findings

Medium

Low

Review

Findings

Low

Review

Findings

Medium

Low

Review

Findings

Critical

Medium

Low

Review

Findings

Medium

Low

Review

Findings

Medium

Low

Uh oh!

This comment was marked as outdated.

Uh oh!

ifireball commented Aug 9, 2026

Uh oh!

fullsend-ai-coder Bot commented Aug 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Aug 9, 2026

🔧 Fix agent — iteration 1 (human-triggered)

Tests: passed

Uh oh!

fullsend-ai-review Bot commented Aug 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as outdated.

Uh oh!

ifireball commented Aug 9, 2026

Uh oh!

ifireball commented Aug 10, 2026

Uh oh!

fullsend-ai-coder Bot commented Aug 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Aug 10, 2026

🔧 Fix agent — iteration 2 (human-triggered)

Uh oh!

fullsend-ai-review Bot commented Aug 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

fullsend-ai-review Bot commented Aug 6, 2026 •

edited

Loading

codecov Bot commented Aug 6, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 6, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Aug 9, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 9, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Aug 10, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 10, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Aug 10, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 10, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Aug 11, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 11, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Aug 11, 2026 •

edited

Loading

fullsend-ai-review Bot commented Aug 11, 2026 •

edited

Loading

fullsend-ai-retro Bot commented Aug 11, 2026 •

edited

Loading