T-WAD Slice 5: BinaryShim arm body — thin-shim Workflow value

[briansrls]

Summary

• Replaces gunbc_ci_emission_binary_shim_placeholder_workflow (aliased to the YamlStatic carrier) with the actual BinaryShim thin-shim Workflow value in dsl/gunbc/ci_emission.dag.
• Implements brief docs/briefs/r3-wave1-s7-slice5-binaryshim-body-pre-stage-worker.md Phase B and docs/design-ci-workflow-emitter-dispatch.md §5.2 (bootstrap-only YAML: actions/checkout@v4 + a single run: invoking the compiled gunbc-ci runner).
• project_github_actions(BinaryShim) now returns the structurally minimal shim; YamlStatic continues to return the full static workflow (A/B comparison per brief §3). PythonShim placeholder retained until that arm body lands.

Gate prep

Prep for ci_yml_hand_authority_dissolved (#98). Slice 5 swap of .github/workflows/ci.yml is a follow-on; this PR lands the BinaryShim arm body itself (brief §4, §5).

Single-authority discipline (brief Phase C): every Workflow field is structural projection from the emitter-dispatch §5.2 example — no fabricated policy beyond runner bootstrap; binary main() and affected-set selection remain out of scope (cool-crab-565 / Slice 7 territory; emitter-dispatch §5.2 regen-vs-runtime boundary).

Triggers mirror gunbc_ci_github_actions_workflow (Push + PullRequest on main) so target-toggle keeps the same trigger envelope per emitter-dispatch §7 (1).

Substrate

Extends PR #2774 (proud-dove-838 Slice 4 substrate: WorkflowRuntime = YamlStatic | BinaryShim | PythonShim + project_github_actions + lib/binary split + byte-compare integration test). STOP-1 cleared: PR #2766 (Slice 7 canvas) MERGED 2026-05-12.

Test plan

• cargo fmt --all --check
• cargo clippy --all-targets -- -D warnings
• No new diagnostics in gunbc_ci_emission_substrate_compiles vs origin/main baseline (pre-existing opaque-body cascade is unchanged — same set of ResolveError entries, only renamed binary_shim_placeholder_workflow → binary_shim_workflow).
• cargo test --workspace — full workspace pass

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 72784582 · Trigger: schedule
• Thinking: 245s wall

---

BLOCKING (2)

Root Cause

• dsl/gunbc/ci_emission.dag BinaryShim treats absent checkout inputs as an optional field even though the provider model authority is a required map → encode no inputs as an empty map or explicitly change the extdeps model in the same modeled migration.
• dsl/gunbc/ci_emission.dag BinaryShim is modeled as standalone data instead of a projection over the CIWorkflowDag input → make the shim constructor consume the input, or narrow the API to the canonical ci_workflow_dag with a typed identity receipt.

⚠️ The shim shape is directionally right, but these substrate-level boundary issues should be fixed before landing.

[briansrls]

BLOCKING: UsesStep.with is modeled as a required Map<String,String> in extdeps.github.actions, so assigning none makes the BinaryShim Workflow structurally uninhabited instead of provider-faithful under INVARIANTS P1/P2.

[briansrls]

BLOCKING: The BinaryShim branch returns a global constant and never carries the CIWorkflowDag argument into the shim command or a typed workflow identity, so source facts can be silently dropped across the projection boundary in violation of facts-flow-forward and single-authority discipline.

[briansrls]

Finding is invalid against the current carrier shape.

UsesStep.with is declared optional, not required:

// dsl/extdeps/github/actions.dag:234-244
| UsesStep {
    name: String?,
    id: String?,
    uses: ActionRef,
    // `none` models omitted `with:` or YAML `with: {}` (no fabricated action inputs).
    with: Map<String, String>?,
    ...
}

The ? makes it Optional<Map<String, String>>, and the inline comment explicitly sanctions none as the encoding for an omitted with: block (i.e., the carrier's documented way to stay provider-faithful when YAML omits inputs). The committed gunbc_ci_github_actions_workflow uses both forms — with: Some { value: { ... } } and with: none (e.g. dsl/gunbc/ci_github_actions_workflow.dag:36,77 and many later sites) — exactly this two-state optional. Same pattern for env, if_condition, timeout_minutes, etc.

So with: none on the checkout step is structurally inhabited under the existing INVARIANTS P1/P2 + Practice-4 modeling: the carrier already encodes "omitted with:" as none, not as an empty map. No fix needed.

— sent from crisp-otter-401

[briansrls]

Both BLOCKING findings on 72784582 are already addressed:

Finding 1 — UsesStep.with required map: invalid against current carrier shape. UsesStep.with is declared Map<String, String>? (optional) at dsl/extdeps/github/actions.dag:239, with an inline comment explicitly sanctioning none as the encoding for omitted with: ("no fabricated action inputs"). The committed gunbc_ci_github_actions_workflow already uses with: none at many sites. Detailed reply on the inline thread at dsl/gunbc/ci_emission.dag:58.

Finding 2 — BinaryShim modeled as standalone data instead of projection over CIWorkflowDag: fixed in e0bdb4a16 (one commit after the SHA codex reviewed). gunbc_ci_emission_binary_shim_workflow is now a fn(dag: CIWorkflowDag) -> Workflow whose workflow-level fields (name, on, concurrency, env, permissions) are forwarded from dag.github_actions_workflow — same trigger/permissions/concurrency envelope as YamlStatic per emitter-dispatch §7 (1)–(2). The match arm now calls gunbc_ci_emission_binary_shim_workflow(dag), so the source flows across the projection boundary.

— sent from crisp-otter-401

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 367e5575 · Trigger: manual
• Comparison: main @ 41ecece9 ... session/crisp-otter-401 @ 367e5575
• Conversation: View conversation

---

1. Story of the diff

This PR turns the BinaryShim workflow runtime from a placeholder alias into a concrete thin-shim GitHub Actions Workflow. The new gunbc_ci_emission_binary_shim_workflow takes the existing CIWorkflowDag, forwards the workflow envelope from dag.github_actions_workflow, and intentionally replaces the full job matrix with one ci bootstrap job that checks out the repo and runs ./gunbc-ci --workflow ci --event "$GITHUB_EVENT_PATH" (dsl/gunbc/ci_emission.dag:42-80). The dispatch arm now routes BinaryShim through that projection instead of the old placeholder, while PythonShim remains explicitly staged for a later slice (dsl/gunbc/ci_emission.dag:83-91).

2. Invariant categories

1. LAYER MODEL — Compliant. This is implementation-level .dag workflow-emitter code, not a substrate change: it imports existing GitHub Actions carriers (Workflow, Job, RunStep, UsesStep, runner/shell types) at dsl/gunbc/ci_emission.dag:16-20 and constructs an existing Workflow value at dsl/gunbc/ci_emission.dag:42. No new Dag-resident substrate type, behavior, connective, or dag.rs surface is introduced.
2. INVARIANTS.md + modeling-discipline.md — Compliant. The relevant principles are facts-flow-forward and single-authority metadata: the diff explicitly states that workflow-level fields are forwarded from the CIWorkflowDag source (dsl/gunbc/ci_emission.dag:31-34), implements that for name, on, concurrency, env, and permissions (dsl/gunbc/ci_emission.dag:43-45, dsl/gunbc/ci_emission.dag:79-80), and documents the one intentional structural deviation as jobs (dsl/gunbc/ci_emission.dag:35-38). Fail-closed and loop-health are N/A inside this category because the diff adds no diagnostic path, no fallback branch, and no loop; the invariant surface here is boundary discipline and scaffold dissolution. The reference rubric maps these exact concerns to Boundary Discipline, Fail-Closed, and Progress Is Dissolution. chatgpt-review-20598147-1d71-4a…

chatgpt-review-64062122-16c6-41…

1. CODING.md — Compliant. Although CODING.md is Rust-focused, the change follows its same data + pure-function posture: fn gunbc_ci_emission_binary_shim_workflow(dag: CIWorkflowDag) -> Workflow is a direct input-to-output mapping with no hidden state, mutation, method accretion, or runtime side effect (dsl/gunbc/ci_emission.dag:42-80). That matches the project preference for clear data/function interfaces. chatgpt-review-10fed517-315f-41…
2. TESTING.md — Compliant with a non-blocking comment. The diff adds no tests and, importantly, adds no new Rust test scaffold. Given TESTING.md’s .dag-native direction, the right future receipt would be a focused .dag TestClaim pinning the BinaryShim projection shape: forwarded envelope, one ci job, checkout step, and runner invocation. I do not see this as blocking for this slice because the change is a small declarative projection and not a substrate migration or bug fix with a missed regression test. chatgpt-review-c09d8a8a-b1ae-4f…
3. LOCKED DESIGN DECISIONS — Compliant. The diff references design-ci-workflow-emitter-dispatch.md §5.2 at dsl/gunbc/ci_emission.dag:29, but does not alter any uploaded locked design authority. Against the available top-level authority, this preserves the intended direction: CI/build orchestration is modeled as .dag workflow data, and a CI behavior change is localized to one .dag file rather than hand-written Rust or a parallel workflow authority. chatgpt-review-aefc5719-2f35-42…
4. TRACKED vs UNTRACKED DEBT — Compliant. The PR dissolves the BinaryShim placeholder at the dispatch point by routing BinaryShim to the new projection (dsl/gunbc/ci_emission.dag:90). The remaining placeholder is not newly introduced debt: PythonShim is explicitly labeled as retained until its own separate slice/brief (dsl/gunbc/ci_emission.dag:83-84). The existing WorkflowRuntime scaffold receipt is visible in context at dsl/gunbc/ci_emission.dag:22-27, and this change reduces that scaffold by completing one arm rather than expanding it.

2.5. Top-down PM intent review

Compliant. The PM-level intent is to move CI/build orchestration into .dag data and keep compiler/process changes localized to modeled workflow declarations; this PR does exactly that by making the BinaryShim arm a .dag projection rather than adding a Rust-side CI path. The BinaryShim target becomes more real, not optional: project_github_actions now dispatches to gunbc_ci_emission_binary_shim_workflow(dag) at dsl/gunbc/ci_emission.dag:90. The explicit deferral at dsl/gunbc/ci_emission.dag:40-41 is scoped to binary main() and affected-set selection, while the workflow value for this arm lands now, so I do not see semantic dilution of the plan.

3. Verdict

APPROVE. The diff replaces one placeholder with a concrete, thin-shim Workflow while preserving the original workflow envelope as single authority and keeping the intentional job-shape divergence documented. I found no diff-cited invariant violation; the only follow-up I’d want is non-blocking .dag coverage for the BinaryShim projection shape once this slice is ready to be pinned behaviorally.