feat: implement Lane 1 Review/SDLC core (W1-W7)

[briansrls]

W1: gunbc-review CLI binary — entry point that builds the review DAG, resolves credentials from env/policy, and executes in Real mode. Input: git diff via --repo-path/-r and --base-ref/-b. Output: structured findings JSON to stdout.

W2: Credential smoke test — dry-run mode validates the full credential chain (GCP WIF → Secret Manager → LLM API key) with mock specs.

W3: Multi-provider support — --provider flag supports openai and anthropic. Credential policy override via GUNBC_CREDENTIAL_POLICY_JSON.

W4: Abstract 4-dimension review model — domain-agnostic dimensions: Coherence (bugs, contradictions), Quality (standards), Requirements (goal accomplishment), Aspirational (must-fix/defer/accept). Each dimension has typed input ports (artifact, criteria, depth, context). Dimensions with no criteria are skipped (opt-in). Aspirational runs last, seeing merged findings from the other three.

W5: Coding review profile — ReviewProfile maps dimensions to domain-specific criteria. coding_review_profile loads criteria from AGENT.md + clippy.toml (quality), issue body (requirements), invariant docs (coherence), refactoring heuristics (aspirational). Follows I6 (no escape hatches) — file contents are passed as parameters.

W6: CI status as review context — gunbc-pipeline --pr N queries CI via gh run list and injects failure context into the requirements dimension. --depth XS|S|M|L|XL flag controls review thoroughness.

W7: gunbc-pipeline command — orchestrates the full daily flow: fetch CI status → build coding profile → run review → output summary with actionable items categorized as must-fix / defer / accept.

https://claude.ai/code/session_01WSV1TUHvgpw9JnG6gB9xV7

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 69606fbef6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use requirements criteria when requirements context is present

Phase 1 builds combined_context from CI/PR/issue data and coding_review_profile_with_requirements stores that context on the requirements dimension, but this selection always pulls ReviewDimension::Quality into config.criteria. In runs with --pr or --issue, the gathered requirements/CI context is therefore dropped before prompt generation, so the review cannot actually validate requirement fulfillment. Select the requirements criteria when context is available (or wire all active dimensions) so this pipeline stage has effect.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Populate repo-based quality context from repo_path

coding_review_profile_from_repo ignores repo_path and always passes ProjectContext::default(), which means AGENT.md/clippy.toml content is never loaded into the quality criteria. Any caller expecting repository-specific standards gets generic checks only, reducing review accuracy across all repos. This helper should read the scoped files under repo_path and fill ProjectContext before building the profile.

Useful? React with 👍 / 👎.