Compose trusted MI admission into server (#238) #241
No reviewers
Labels
No labels
accessibility
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
P1
P2
P3
question
wontfix
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Creatidy/scarcity-router!241
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "issue-238-mi-server"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Refs #238
Explicit
--mi-configenables the existing optional pinned MI consumer/admission in the authenticated composed server and gateway; default operation remains MI-free. Administrator-owned bounded trust configuration and an authenticated CSRF-protected reload validate exact publication references and bindings. Prior cuts and private decision receipts are retained in the existing protected, bounded server store (schema 3); exact utilization references are returned without exposing private routing context. Current store generation includes the full trusted configuration and prevents stale new decisions across server instances; captured in-flight snapshots remain immutable.Base
209f7d39f3ecebbd1b98b11d7c44c63192fe50a4; HEAD916a74350e61c7b262fe328f20f29d2d9db25966. This branch does not copy unmerged PR #240.Validation:
uv sync --only-dev; finalmake check3,196 tests passed / 18 skipped and basedpyright 0 errors/warnings; standaloneuv run basedpyright;make package-check;git diff --check; focused MI, store migration, authenticated reload, and offline installed-wheel composed-server acceptance with pinned MI extra. Earlier full runs and continuation readiness recovery are recorded on issue #238.Independent security/correctness review #2 APPROVE, findings [], at exact HEAD/base; #1's binding-currentness finding was remediated and independently verified. Installed acceptance uses synthetic evidence with no discovered eligible source: it verifies authenticated trusted-cut consumption, reference and retained receipt, not positive live-provider evidence or source-owner review. Z.ai positive applicability remains unproven. The bounded store fails closed at its documented retention cap. No merge or issue closure authorized.
Independent whole-PR security/correctness review #1/10 reserved for exact HEAD
91df2223b1f0aa7041584d1fc02162c115271a34, base/merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4. Fresh native read-only GPT-6.1 Sol High review covers the complete PR, trust configuration and publication replacement, private authorization, persistence/migration, data integrity, provenance, installed acceptance, standalone compatibility and UX. Reservation is not a verdict.Review #1 dispatched as fresh native user-visible read-only session
ses_e6b983884ffefkhphoqYN5gq4f, GPT-6.1 Sol via openai, high requested. Runtime backend identity unobserved; no verdict or approval is implied. The reviewed source remains frozen.Independent review #1/10: REQUEST_CHANGES, exact HEAD
91df2223b1f0aa7041584d1fc02162c115271a34, base/merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4; native read-only sessionses_e6b983884ffefkhphoqYN5gq4f. P2: activation digest excludes administrator trust bindings, so a second instance can continue new decisions under old binding restrictions after a bindings-only reload of identical publication bytes. Required: include complete admission-relevant trusted configuration in atomic activation identity and test two-instance stale detection while preserving captured snapshots. No approval; remediation in progress. Reviewer did not run runtime tests.Review #1 remediation pushed as
916a74350e61c7b262fe328f20f29d2d9db25966: the durable activation record includes a digest of the complete validated administrator trust configuration, so bindings-only changes with the same artifact invalidate stale instances. New regression verifies stale-instance refusal and unchanged already-captured snapshot. Focused MI/installed tests and finalmake check(3,196 tests, 18 skipped, basedpyright 0 diagnostics),make package-check, and whitespace checks pass. Review #1 did not approve this HEAD; fresh whole-PR review required.Independent whole-PR security/correctness review #2/10 reserved for exact HEAD
916a74350e61c7b262fe328f20f29d2d9db25966, base/merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4. Fresh native read-only GPT-6.1 Sol High review of the complete PR, including #1 binding/currentness remediation, publication trust and ownership, authorization, data integrity, restart, provenance, acceptance and UX. Reservation is not a verdict.Review #2 dispatched as fresh native user-visible read-only session
ses_e6b846465ffeGx09TxCFHNdq3c, GPT-6.1 Sol/openai/high requested; runtime backend identity unobserved. Source frozen; no verdict implied.Independent whole-PR security/correctness review #2/10: APPROVE, findings [], exact HEAD
916a74350e61c7b262fe328f20f29d2d9db25966, base/merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4. Native read-only sessionses_e6b846465ffeGx09TxCFHNdq3c, GPT-6.1 Sol/openai/high requested; runtime backend identity unobserved. Reviewer confirmed #1 binding-currentness remediation and assessed security, atomicity, history, standalone compatibility and UX. Reviewer did not run runtime tests. Explicit limitation: installed acceptance is synthetic with no eligible discovered source, not positive live-provider admission, source-owner review, current Z.ai applicability or production readiness. Approval is not integration; PR remains open.Owner-authorized integration update: #240 merged; current base
89aaba452283c5cf05d698201d0a7b7087712c71. Clean-cache reproduction afteruv sync --only-devshowed offlineuv pip installcould not resolvemcpbecause required registry resolution metadata was absent; the clean cache also lackedhatchlingfor an offline wheel build. Original CI assertion suppressed its uv stderr, so the exact original stderr remains unavailable. Repaircb7a155ea17938d6164d54f71995e97f5cd5bcb0explicitly prepares the wheel and locked dependencies through PyPI and canonical exact-pinned MI Git source, then verifies a separate offline install, MI VCS commit and installed path; failures report sanitized diagnostics. Fresh-empty-cache focused test passed. Actual combined merge tree94404947f45ef8bb947c9f9adaf37420de1d8207matched the independently prepared validation index exactly;make check3197 tests/18 skipped,uv run basedpyright,make package-check, and whitespace checks passed. Canonical CI and fresh exact-ref review remain pending.Independent whole-PR security/correctness review #3/10 reserved: current HEAD
cb7a155ea17938d6164d54f71995e97f5cd5bcb0, current base89aaba452283c5cf05d698201d0a7b7087712c71, merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4; expected combined tree94404947f45ef8bb947c9f9adaf37420de1d8207. Fresh read-only GPT-6.1 Sol High native review covers complete PR and new CI repair/combined architecture; #1 and #2 remain consumed. Reservation is not a verdict.Review #3 dispatched in fresh native read-only session
ses_e69f1baddffeHB87Zw77ipGGMp, GPT-6.1 Sol/openai/high requested; runtime backend identity unobserved. Source is frozen; no verdict implied.Independent whole-PR review #3/10: APPROVE, findings [], exact HEAD
cb7a155ea17938d6164d54f71995e97f5cd5bcb0, current base89aaba452283c5cf05d698201d0a7b7087712c71, merge-base209f7d39f3ecebbd1b98b11d7c44c63192fe50a4, combined tree94404947f45ef8bb947c9f9adaf37420de1d8207. Native read-only sessionses_e69f1baddffeHB87Zw77ipGGMp, GPT-6.1 Sol/openai/high requested; runtime backend identity unobserved. Reviewer assessed all 12 changed files, security, combined architecture, installed-package repair, provenance and UX; no runtime tests run in review. Synthetic installed acceptance does not establish positive live-provider or source-owner evidence. Canonical CI #3418 was still running at review completion and remains mandatory.