feat: add FCAME private result transport

[andrei-hasna]

Outcome

Adds the package-owned FCAME-1 private transport needed for production Knowledge publication:

• process-private input, exact-title lookup, and result descriptors
• authenticated exact-title lookup bound to authority, tenant, scope, and parent
• metadata-only zero-or-one lookup proofs with ambiguity refusal
• private execute and readback result inspection without body or title serialization
• public capability discovery and generated declaration/runtime parity

Verification

• typecheck: exit 0
• guarded writer tests: 28 pass, 0 fail
• CLI suite: 87 pass, 0 fail
• full suite: 454 pass, 2 skipped, 0 fail
• package release tests: 6 pass, 0 fail
• build and contracts conformance: pass
• package validation: pass for @hasna/knowledge 0.2.100
• generated bundles rebuilt byte-identically
• gitleaks: 2 commits scanned, no leaks found

Review record

Initial implementation self-review returned NO_GO because the capability response implied package-wide body-output absence while the legacy public execute method still returns full readback. Commit af8fa95 scopes the claim to private transport only; affected-lane verification passes and the focused implementation self-review is GO.

The mandatory independent coding-agent npm release review remains outstanding. This PR does not authorize merge or publication until that reviewer binds its verdict to repo hasna/knowledge, exact commit af8fa95, package @hasna/knowledge@0.2.100, and the conventional registry release path.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[andrei-hasna]

[REVIEW] GO — #82 @ af8fa95 — lens: correctness+security+gates, reviewer Aruns (1 of 1)

Reviewed exact candidate af8fa95 against freshly fetched origin/main a1eb5bf.

What I read:

• git log --oneline origin/main..HEAD
• git diff origin/main...HEAD --stat
• the complete textual diff for all 17 changed text files, plus surrounding source for the private descriptor/result WeakMap lifecycle, guarded transport, authenticated server handler, tenant/authority binding, bounded body/response handling, parameterized title lookup, generated declarations/bundles, CLI capability surface, and tests
• bin/knowledge.js is a binary generated artifact, so it has no textual source diff; I exercised its changed command directly and verified the repository's generated-artifact/package-release tests

Commands and evidence:

• bun install — exit 0; setup only; 156 packages installed
• bun run test — an initial streamed invocation completed but its wrapper did not retain the terminal exit line, so I repeated the identical declared gate with stdout/stderr captured separately
• bun run test — repeat exit 0; 454 pass, 0 fail, 2 skip; 456 tests across 49 files; 3595 expect() calls
• ./bin/knowledge.js guarded capabilities --json — exit 0; reports FCAME-1 private input/result and exact-title lookup support with private_transport_body_output=false
• git status --short — exit 0; clean after removing the temporary test capture directory

The repository declares no typecheck script, so no typecheck gate was invented or run.

Blocking P0/P1 findings: none.

Security/correctness trace:

• The exact-title endpoint requires tenant-bound knowledge:read authentication.
• Authority, tenant, scope, parent, descriptor digest, title digest, expiry, request bounds, and response bounds are checked before results are accepted.
• The SQL path is parameterized, exact-binding scoped, limited to two rows for ambiguity detection, and fails closed on duplicates.
• Public descriptors serialize only metadata/digests; title/body material stays in process-private WeakMaps or the authenticated bounded request, and returned public result descriptors contain no title/body.
• The generated runtime, declarations, package exports, and source implementation agree under the declared test gate.

Non-blocking follow-ups: none.

[andrei-hasna]

GO — repo hasna/knowledge; PR 82; candidate commit af8fa95; package @hasna/knowledge@0.2.100; workflow target CI run 31307883975; npm tag target latest; registry target https://registry.npmjs.org.

P0 blockers: 0. P1 blockers: 0.

Evidence:

• Live PR head remained af8fa95. PR 82 was merged at 2026-08-09T10:35:06Z while this review was in progress; this reviewer did not merge, publish, tag, deploy, or install.
• All seven required branch-protection contexts were completed-success: test-matrix (windows-latest, bun), test-matrix (macos-latest, bun), test (macos-latest, bun), test-matrix (ubuntu-latest, bun), test (macos-latest, node), test (ubuntu-latest, bun), and test (ubuntu-latest, node). The only skipped check, [code]smith, is not required.
• CI evidence shows 454 pass, 2 skipped, 0 fail; the FCAME-1 private write/result and exact-title lookup tests pass; generated-artifact verification and public package release safety pass on the tested Bun/Node lanes.
• Independent Gitleaks 8.30.1 scan of a1eb5bf..af8fa95 scanned 2 commits and found 0 leaks.
• Source and generated surfaces keep payloads/titles/results behind WeakMaps and metadata-only toJSON/proof shapes; exact-title lookup authenticates, checks configured authority and tenant, filters authority/tenant/scope/parent, bounds to LIMIT 2, and refuses ambiguity. Client-side binding, expiry, digest, exactness, and bounded-result checks fail closed.
• Public root runtime/declaration exports include the intended descriptor/proof APIs but exclude materializeKnowledgePrivateInput, materializeKnowledgePrivateTitleLookup, and createKnowledgePrivateResultDescriptor; the capability claim is explicitly scoped to private transport.
• package.json is @hasna/knowledge 0.2.100 with publishConfig.registry https://registry.npmjs.org and public access. npm view confirms 0.2.100 is not yet published; no publish was attempted in this review.

Verdict: GO for the named candidate and conventional npm registry release path.

[andrei-hasna]

[REVIEW-RECEIPT] GO

Reviewer agent identity: Aruns
Publisher agent identity: quintilianus
Reviewer/publisher separation: distinct

Repository: hasna/knowledge
Merged publish commit: 5f03c6e
Reviewed candidate: af8fa95
Merged tree: a48f834e0e830f640d0a29a460f0ba64ea3055bf
Reviewed tree: a48f834e0e830f640d0a29a460f0ba64ea3055bf
Tree binding: exact equality

Package: @hasna/knowledge
Version: 0.2.100
Release tag: none
Registry target: https://registry.npmjs.org/

Publish workflow revision: local npm publish path at merged commit 5f03c6e. The package.json prepublishOnly gate is bun run contracts:conformance && contracts no-cloud-scan . && bun run build && node scripts/validate-public-package.mjs. Authentication uses a secure temporary npmrc containing the literal ${NODE_AUTH_TOKEN} placeholder, consumed through secrets exec. No GitHub Actions environment or manual human approval is used by this publish path. The repository's unused npm-release environment has zero protection rules and is not referenced by the sole workflow, .github/workflows/ci.yml.

Registry negative control: @hasna/knowledge=0.2.99; @hasna/knowledge@0.2.100 is absent, so publish remains blocked without this receipt.

Open reachable in-scope P0 findings: 0
Open reachable in-scope P1 findings: 0

Positive control: this exact GO binding permits the release gate.
Negative control: no GO, or any wrong commit, tree, package, or version binding, remains blocked.

Verdict: GO

Scope: focused release-binding metadata continuation only; unchanged code and evidence were not relitigated.