Harden proof boundaries and close architecture drift

.github/workflows/ci.yml

@@ -45,6 +45,11 @@ jobs:
           go-version-file: go.mod
           cache: true
 
+      - name: Setup Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
+        with:
+          python-version: '3.14.6'
+
       - name: Verify diff hygiene
         run: |
           set -euo pipefail
@@ -70,26 +75,11 @@ jobs:
       - name: Verify Go formatting
         run: npm run go:fmt
 
-      - name: Run Go tests for app
-        run: go test ./cmd/agentic-proofkit ./internal/app
-
-      - name: Run Go tests for kernel
-        run: go test ./internal/kernel/...
-
-      - name: Run Go tests for commands
-        run: go test ./internal/command/...
-
-      - name: Run Go tests for coverage and package tools
-        run: go test ./internal/tools/coveragemetrics ./internal/tools/packagebuild ./internal/tools/packagepack ./internal/tools/packageverify
-
-      - name: Run Go tests for registry and Python tools
-        run: go test ./internal/tools/pypiregistry ./internal/tools/pythonpackage
-
-      - name: Run Go tests for release tools
-        run: go test ./internal/tools/releasecloseoutinput ./internal/tools/releasemanifest ./internal/tools/releasepreflight ./internal/tools/releasesbom ./internal/tools/textpolicyinput
+      - name: Verify generated command family catalog
+        run: npm run command-family:check
 
-      - name: Run Go tests for scripts
-        run: go test ./scripts
+      - name: Run all Go tests
+        run: npm run go:test
 
       - name: Run Go vet
         run: npm run go:vet

docs/proofkit-contract-map.md

@@ -47,7 +47,7 @@ owner boundaries. It is not a second command-family inventory.
 | Receipts and producers | `proof-receipt-admission`, `receipt-producer-admission`, `receipt-currentness-scope`, `receipt-trust-class`, `producer-policy-self-proof` | receipt sets, producer policy, scope/currentness facts, trust classes | receipt shape, producer/receipt compatibility, self-proof diagnostics | producer authentication, freshness policy, CI trust roots | receipt/provenance report |
 | Release and deployment | `release-authority`, `external-consumer`, `registry-consumer-proof-input-compose`, `registry-consumer`, `deployment-evidence-admission`, `completion-criteria`, `branch-authority`, `readiness-closeout` | package facts, tarball/registry facts, explicit primitive registry/install/smoke facts, deployment evidence, criteria, branch facts | artifact/channel boundary checks, registry-consumer input composition, release diagnostics, falsifiable criteria shape | package publication, registry fetch, package-manager execution, deployment, rollback, approval | composed input, release/deployment/readiness report |
 | Supply-chain and quality | `self-check`, release workflow, `npm run release:sbom`, `npm run self:coverage`, `npm run go:actionlint`, `npm run go:bench` | release artifacts, source workflows, specs, bindings, witness plans, explicit benchmark invocation | deterministic self-check report shape, SBOM candidate evidence, coverage metrics, workflow lint routing, benchmark entrypoints | public-source provenance, vulnerability triage, license approval, CI run admission, release approval | self-check report, SBOM, metrics report, CI signal, or benchmark output |
-| Repository structure | `repo-profile-admission`, `workspace-manifest-facts`, `workspace-registry`, `workspace-changed-package-plan`, `workspace-shard-partition`, `typescript-public-api-surfaces`, `text-policy`, `secret-scan`, `package-runtime-dependency-admission` | explicit repo/profile facts, caller-owned manifest records, caller-owned roots, caller-owned text file inventories, optional `environmentClassPolicies` tuples | structural admission, manifest-to-workspace fact projection, workspace graph projections, TypeScript package public API checks, text policy admission, explicit-inventory secret-like text detection, shard plans | repository freshness, git/file discovery, command policy, package manager truth, provider secret scanning | structural, fact, policy, or planning report |
+| Repository structure | `repo-profile-admission`, `workspace-manifest-facts`, `workspace-registry`, `workspace-changed-package-plan`, `workspace-shard-partition`, `typescript-public-api-surfaces`, `text-policy`, `secret-scan`, `package-runtime-dependency-admission` | explicit repo/profile facts, caller-owned manifest records, caller-owned roots, caller-owned text file inventories, explicit TypeScript package-manifest and per-condition source paths, optional `environmentClassPolicies` tuples | structural admission, manifest-to-workspace fact projection, workspace graph projections, bounded TypeScript package public API checks over referenced files, text policy admission, explicit-inventory secret-like text detection, shard plans | repository freshness, git/file discovery, compiler output provenance, command policy, package manager truth, provider secret scanning | structural, fact, policy, or planning report |
 | Custom and generated artifacts | `custom-rule-boundary`, `document-lifecycle-boundary`, `rendered-artifact-freshness`, `conformance-profile`, `json-report-cli-adapter-source`, `witness-plan`, `witness-scheduler-plan` | custom rule metadata, document lifecycle records, artifact digests, profile manifests, command metadata, adapter language | boundary checks, generated-view freshness shape, deterministic adapter source generation, scheduler metadata checks | rule execution, document meaning, cache contents, CI scheduling, committed generated-source freshness | boundary report, generated source artifact, or scheduler report |
 | CLI metadata | `help` | optional command name or help flag | built-in command catalog and help text routing | command selection, semantic proof, freshness, merge policy | text help only |
 
@@ -121,7 +121,7 @@ Decision tree:
 | Tests or proof evidence need inventory. | `test-evidence-inventory`; use `--projection discovery-draft` only for explicit discovered-test facts, then `requirement-coverage-input-compose` when an aggregate `coverage_compose_input` exists, then `requirement-coverage-view` when a `coverage_view_input` exists | Compose only from explicit caller-owned facts. Discovery drafts are candidate-only and cannot close coverage. Use `failureClassifications[]`, `warningClassifications[]`, and `agentActionPlan[]` for machine routing. Escalate when tests are route-only, weak-oracle, unbound, or outside the caller-owned coverage universe. |
 | A change set is known. | `changed-path-set`, optionally `requirement-impact-input-compose`, `impact`, then `selective-gate-plan --agent-envelope` | Raw `knownChangedPaths` in `agent-route` are diagnostic only. Materialize a caller-owned `changed_path_set`, compose a caller-owned `impact_input` before `impact`, and compose a caller-owned `selective_gate_plan_input` before `selective-gate-plan`. Use `scanObligation` to name whether a `text-policy`, `secret-scan`, or caller-owned external scanner is required. Fail closed on unknown scope, dynamic edges, missing owner routes, unbound proof-like paths, or full-gate escalation. |
 | Caller-owned file contents need secret-like text detection. | `secret-scan` | Provide explicit sorted file inventory with content. Stop before claiming repository-wide discovery, credential validity, provider ingestion, merge readiness, or replacement of GitHub secret scanning. |
-| Does a TypeScript package public API match a caller-owned manifest? | `agent-route` with `goal: "verify_typescript_public_api"` and explicit `typescript_public_api_manifest` plus `typescript_public_api_repo_root`, then `typescript-public-api-surfaces --repo-root <caller-selected-root>` | This is an explicit filesystem scanner. Stop before guessing `repo-root`, claiming checkout freshness, package-manager truth, or merge readiness. |
+| Does a TypeScript package public API match a caller-owned manifest? | `agent-route` with `goal: "verify_typescript_public_api"` and explicit `typescript_public_api_manifest` plus `typescript_public_api_repo_root`, then `typescript-public-api-surfaces --repo-root <caller-selected-root>` | The manifest must name each referenced `package.json`, sorted-unique export conditions, and a non-JSX `.ts`, `.mts`, or `.cts` `sourcePath` whose canonical target has the same admitted extension class. The bounded scanner accepts only the fail-closed export grammar in `proofkit/cli-contract.v1.json`; it does not parse unrestricted TypeScript or TSX, infer conventional layouts, or prove compiler output provenance, checkout freshness, package-manager truth, or merge readiness. |
 | Receipts are available for planned checks. | `selective-gate-evidence --agent-envelope`; then materialize a caller-owned `obligation_decision_input` from the evidence output plus command routes, currentness, and trust facts; then run `selective-gate-obligation-decision-input`; then materialize the resulting `obligation_decision` input and run `obligation-decision --agent-envelope` | Escalate on missing, stale, invalid, untrusted, blocked, unavailable, or unknown-scope evidence. |
 | Human inspection is needed. | `requirement-source-view`, `requirement-proof-view`, `requirement-coverage-view`, `requirement-spec-tree-view`, or `requirement-browser-server` | `agent-route` emits browser commands as plan-only by default. Use `browserMode: "serve_local_view"` for `--serve` and `openBrowser: true` for `--open`. Rendered HTML and Markdown are presentation only unless the consumer admits a tracked artifact freshness gate. |
 | Temporary external document lifecycle facts, generated views, or rendered views need authority classification. | `document-lifecycle-boundary` | Treat lifecycle records as caller-owned metadata. Temporary design docs and implementation plans are not retained repository authority unless rewritten into deterministic specs, proof bindings, tests, package-public docs, or backlog rows. |

docs/specs/proofkit-package-boundary/overview.md

@@ -13,22 +13,27 @@ denial, and package artifact behavior only.
 - `REQ-PROOFKIT-PACKAGE-002`: the CLI builds deterministic reports, plans,
   generated source artifacts, and policy-admission results from explicit
   caller-owned JSON, declared no-input command parameters, or declared explicit
-  scanner scope classes without executing native witnesses, scanning implicit
-  repository state, deciding proof freshness, or accepting broad caller-supplied
-  phrase suppressors that can hide readiness overclaims.
+  scanner scope classes. Explicit scanners consume only caller-named bounded
+  canonical files under documented fail-closed grammars rather than guessing
+  layouts or silently accepting unsupported syntax; no command executes native
+  witnesses, scans implicit repository state, decides proof freshness, or accepts
+  broad caller-supplied phrase suppressors that can hide readiness overclaims.
 - `REQ-PROOFKIT-PACKAGE-003`: the root package remains installable and
   executable by an outside consumer on the current native platform without
   claiming registry publication.
 - `REQ-PROOFKIT-PACKAGE-004`: CI package-gate receipts used as merge evidence
   are admitted through a declared producer policy and proof-receipt shape
   validator instead of current-build output alone.
-- `REQ-PROOFKIT-PACKAGE-005`: the Go source, static analysis, package gate,
-  and vulnerability gates remain the native merge-critical quality floor for
-  the current Proofkit source tree.
+- `REQ-PROOFKIT-PACKAGE-005`: tracked source hygiene covers each admitted text
+  language in staged blobs and the current worktree; Go formatting, static
+  analysis, test, package, and vulnerability gates remain the native
+  merge-critical quality floor and explicitly provision every runtime required
+  by a merge-critical oracle.
 - `REQ-PROOFKIT-PACKAGE-006`: Python/uv distribution is a platform wheel
   wrapper over the same Go CLI, with wheel-safe package metadata, wheel tags,
-  embedded binary identity, local install smoke proof, and explicit non-claims
-  until PyPI publication.
+  embedded binary identity, local install smoke proof, POSIX exec
+  process-identity and signal preservation, and explicit non-claims until PyPI
+  publication.
 - `REQ-PROOFKIT-PACKAGE-007`: package-public Markdown records release-channel
   state only and must not embed exact per-version provider facts that are owned
   by immutable registry, release, and manifest artifacts.

docs/specs/proofkit-package-boundary/requirements.v1.json

@@ -35,7 +35,7 @@
     {
       "requirementId": "REQ-PROOFKIT-PACKAGE-002",
       "ownerId": "proofkit.package-boundary",
-      "invariant": "The CLI builds deterministic reports, plans, generated source artifacts, and policy-admission results from explicit caller-owned JSON, declared no-input command parameters, or declared explicit scanner scope classes without executing native witnesses, scanning implicit repository state, deciding proof freshness, or accepting broad caller-supplied phrase suppressors that can hide readiness overclaims.",
+      "invariant": "The CLI builds deterministic reports, plans, generated source artifacts, and policy-admission results from explicit caller-owned JSON, declared no-input command parameters, or declared explicit scanner scope classes; explicit scanners consume only caller-named bounded canonical files under documented fail-closed grammars rather than guessing layouts or silently accepting unsupported syntax, and no command executes native witnesses, scans implicit repository state, decides proof freshness, or accepts broad caller-supplied phrase suppressors that can hide readiness overclaims.",
       "claimLevel": "blocking",
       "riskClass": "high",
       "proofBindingRefs": [
@@ -116,7 +116,7 @@
     {
       "requirementId": "REQ-PROOFKIT-PACKAGE-005",
       "ownerId": "proofkit.package-boundary",
-      "invariant": "The Go source, formatting, static analysis, vet, test, package artifact, and vulnerability gates remain the native merge-critical quality floor for the current Proofkit source tree.",
+      "invariant": "Tracked source hygiene covers every admitted project text language in both staged blobs and the current worktree, while Go formatting, static analysis, vet, test, package artifact, and vulnerability gates remain the native merge-critical quality floor and explicitly provision every runtime required by a merge-critical oracle.",
       "claimLevel": "blocking",
       "riskClass": "high",
       "proofBindingRefs": [
@@ -143,7 +143,7 @@
     {
       "requirementId": "REQ-PROOFKIT-PACKAGE-006",
       "ownerId": "proofkit.package-boundary",
-      "invariant": "Python/uv distribution is a platform wheel wrapper over the same Go CLI, with wheel-safe package metadata, wheel tags, embedded binary identity, local install smoke proof, and explicit non-claims until PyPI publication.",
+      "invariant": "Python/uv distribution is a platform wheel wrapper over the same Go CLI, with wheel-safe package metadata, wheel tags, embedded binary identity, local install smoke proof, POSIX exec process-identity and signal preservation, and explicit non-claims until PyPI publication.",
       "claimLevel": "blocking",
       "riskClass": "medium",
       "proofBindingRefs": [

docs/specs/proofkit-spec-proof-core/overview.md

@@ -25,16 +25,18 @@ execution receipts, and merge policy.
 - `REQ-PROOFKIT-SPEC-003`: witness planning accepts caller-owned structured
   command metadata, scheduler constraints, environment classes, and
   binding-derived command projections only through admitted witness vocabulary
-  and conservative safe argv policy, without executing commands or selecting
-  repository policy.
+  and conservative safe argv policy. Self-hosting bindings and witness plans
+  preserve command ids, argv, and environment classes without executing
+  commands or selecting repository policy.
 - `REQ-PROOFKIT-SPEC-004`: selective planning and selective evidence reports
-  keep changed-path facts, planned commands, receipts, evidence class,
-  producer-admission state, and obligation candidates explicit, keep merge
-  approval consumer-owned, and fail closed for unknown or unmatched proof
-  inputs.
+  keep canonical changed-path facts, resolvable agent references, planned
+  commands, receipts, evidence class, producer-admission state, and obligation
+  candidates explicit, keep merge approval consumer-owned, and fail closed for
+  unknown or unmatched proof inputs.
 - `REQ-PROOFKIT-SPEC-005`: rendered proof views and agent envelopes remain
-  bounded, derived presentations over structured source and never become
-  canonical proof or requirement authority.
+  bounded, reference-closed derived presentations over structured source,
+  count each semantic omission once, and never become canonical proof or
+  requirement authority.
 - `REQ-PROOFKIT-SPEC-006`: test evidence inventory, explicit test-discovery
   draft projections, proof-binding-derived inventory projections, normalized
   inventory projections, coverage-view input composition, and requirement

docs/specs/proofkit-spec-proof-core/requirements.v1.json

@@ -62,7 +62,7 @@
     {
       "requirementId": "REQ-PROOFKIT-SPEC-003",
       "ownerId": "proofkit.spec-proof-core",
-      "invariant": "Witness planning accepts caller-owned structured command metadata, scheduler constraints, environment classes, and binding-derived command projections only through admitted witness vocabulary and conservative safe argv policy, without executing commands or selecting repository policy.",
+      "invariant": "Witness planning accepts caller-owned structured command metadata, scheduler constraints, environment classes, and binding-derived command projections only through admitted witness vocabulary and conservative safe argv policy; self-hosting requirement bindings and witness plans preserve the same command ids, argv, and environment classes without executing commands or selecting repository policy.",
       "claimLevel": "blocking",
       "riskClass": "high",
       "proofBindingRefs": [
@@ -89,7 +89,7 @@
     {
       "requirementId": "REQ-PROOFKIT-SPEC-004",
       "ownerId": "proofkit.spec-proof-core",
-      "invariant": "Selective planning and selective evidence reports keep changed-path facts, planned commands, receipts, evidence class, producer-admission state, and obligation candidates explicit, keep merge approval consumer-owned, and fail closed for unknown or unmatched proof inputs.",
+      "invariant": "Selective planning and selective evidence reports keep canonical changed-path facts, resolvable agent references, planned commands, receipts, evidence class, producer-admission state, and obligation candidates explicit, keep merge approval consumer-owned, and fail closed for unknown or unmatched proof inputs.",
       "claimLevel": "blocking",
       "riskClass": "high",
       "proofBindingRefs": [
@@ -116,7 +116,7 @@
     {
       "requirementId": "REQ-PROOFKIT-SPEC-005",
       "ownerId": "proofkit.spec-proof-core",
-      "invariant": "Rendered proof views and agent envelopes remain bounded, derived presentations over structured source and never become canonical proof or requirement authority.",
+      "invariant": "Rendered proof views and agent envelopes remain bounded, reference-closed derived presentations over structured source, count each semantic omission once, and never become canonical proof or requirement authority.",
       "claimLevel": "blocking",
       "riskClass": "medium",
       "proofBindingRefs": [