clio-relay 1.5.10

Patch release for the native FastMCP 4 / SEP-2663 relay task server.

• Constrains HTTPX below 1.0 so clean prerelease-enabled installs resolve a FastMCP-compatible client.
• Accepts authenticated, schema-validated native JARVIS progress in live release acceptance.
• Preserves relay as the sole task, scheduling, cancellation, and durability authority.

The full pre-release CI matrix passed on Linux and Windows with Python 3.12, 3.13, and 3.14. Exact released-wheel HTTP, stdio, reconnect, cancellation, and Ares application validation continues against these published assets.

clio-relay 1.5.9

Adds relay-backed native FastMCP 4 task protocol support for SEP-2663.

• Projects existing relay jobs as standard MCP tasks over authenticated streamable HTTP and stdio.
• Preserves relay task identity, lifecycle, replay, reconnect, terminal results, and explicit cancellation.
• Keeps the relay queue as the sole execution and durability authority; FastMCP's Docket executor remains inactive.
• Enables task semantics for the trusted virtual JARVIS MCP tools.
• Gates locked MCP launcher startup on verified cache lifecycle events.

Published release

• Release: https://github.com/iowarp/clio-relay/releases/tag/v1.5.9
• Product commit: ac0e7661f4b0bd390068cb2bbf22f2aaddc771d4
• Merged PR: #141
• Full pre-release CI: https://github.com/iowarp/clio-relay/actions/runs/30489159709
• Release-tag validation: https://github.com/iowarp/clio-relay/actions/runs/30493266384
• Full post-release tag matrix: https://github.com/iowarp/clio-relay/actions/runs/30493266349
• PyPI: https://pypi.org/project/clio-relay/1.5.9/
• Reviewed evidence bundle: https://github.com/JaimeCernuda/cc_dispatch/tree/39a9ce815e94d03424129a36a36bbc80ab34658d/mcp_ui_sciences/evidence/fastmcp-sep2663
• Published wheel asset ID: 494557250
• Published wheel SHA-256: 70f5e450935cb6cf9f17a3e6eed0bcfb5ee245fcbd45c678774a9e6f0dda1a5e

released-asset-attestation.json binds the downloaded wheel, source archive, and
checksum manifest to their public GitHub release assets and the exact product
commit. released-wheel-attestation.json proves the isolated runtime imported
version 1.5.9 from that downloaded wheel, exposed the
io.modelcontextprotocol/tasks extension, and had no Docket executor.
PyPI serves the same wheel and source archive sizes and SHA-256 digests as the
public GitHub Release assets.

Released-wheel live acceptance

The official FastMCP client-only package drove the downloaded public wheel over
real authenticated streamable HTTP and stdio. The recorded run passed:

• SEP-2663 discovery over HTTP and stdio, with anonymous HTTP rejected as 401.
• A real Ares-backed task, client detach, reconstruction by a new client, and
  successful completion under the same relay task identity.
• Six immutable relay-owned artifact records attached to the terminal result.
• A separate task submitted to an isolated relay queue, client detach,
  standards-based cancellation, and canonical convergence to cancelled under
  the same task identity.
• Relay as the only execution and durability authority throughout the run.

The machine-readable results are in released-live-report-v1.5.9.json and
recordings/released-v1.5.9/recording-report.json.

The reviewed recording is
recordings/released-v1.5.9/released-live-acceptance.webm:

• VP8 WebM, 1440 by 1000, 25 fps, 211.96 seconds.
• 16,460,518 bytes.
• SHA-256: a80d81d743f0a363bf176186daec16a659c268e71f4f0ef7b5c87c5d286dbe03.

All five captured interaction states were visually reviewed. They visibly show
the released-wheel identity, inactive Docket executor, transport/auth pass,
reconnected relay task ID and artifact evidence, cancellation pass, and final
campaign state passed.

clio-relay 1.5.8

Patch release for deterministic cross-platform release validation. The control-query overlap test now synchronizes workload release explicitly instead of converting a slow Windows runner into a synthetic exit-code-124 workload failure. Runtime transport, queue, session, bootstrap, and MCP behavior is unchanged from v1.5.7.

clio-relay 1.5.7

Patch release synchronizing the machine-readable release acceptance matrix and its test assertion after the v1.5.6 metadata update. Runtime behavior is unchanged from v1.5.6, whose published wheel passed live Ares JARVIS, WSS relay, direct XTCP, SSH, MCP routing, and detach/reconnect acceptance.

clio-relay 1.5.6

Fix bootstrap and JARVIS execution on symlinked home directories by preserving the stable lexical home spelling for relay-managed repositories.

Validated with an exact candidate wheel on Ares through bootstrap repair/no-op, a real JARVIS/Slurm execution, and user-profile MCP over authenticated SSH before and after detach/reconnect.

clio-relay 1.5.5

Highlights

• Complete the ParaView/JARVIS live-runtime path, including pinned artifact staging, execution-owned service binding, recovery, teardown, and acceptance coverage.
• Pin the published clio-kit 2.6.5 and JARVIS-CD 1.7.0 release artifacts by URL and SHA-256.
• Reconcile local file inputs automatically before JARVIS execution so submitted jobs receive the intended staged datasets.
• Repair Windows release validation and preserve one build across the complete Linux/Windows Python 3.12-3.14 matrix.
• Harden remote bootstrap preflight against inherited loader, Python, and shell-startup environment variables.

Included work

• #134 — Windows CI and v1.5.5 release preparation
• #136 — JARVIS run-input reconciliation
• #137 — ParaView live-runtime acceptance

The attached wheel and source distribution were built once from commit
e017ca7a2f90bb726cbe4e4ca8af79e26d6dd548 with embedded tag v1.5.5.

clio-relay 1.5.4

clio-relay 1.5.4 restores the complete Linux release gate and hardens the production boundaries it exposed.

• Fixed JARVIS input-lineage timestamp ordering, injected-queue seal fencing and lock cleanup, sealed lease-capacity recovery, Windows ACL sharing, and validated already-absent cgroup cleanup.
• Added bounded machine-readable pytest failure IDs and preserved failed validation reports.
• Aligned stale fixtures and acceptance node IDs with the current endpoint-owned MCP, cleanup, API, control-query, and clio-kit contracts.
• The final broad suite completed with 2,946 passing tests; the containment and sidecar acceptance subsets each passed 7/7. The only post-suite failure was a stale validator node name, repaired and verified by focused tests before release.

Build-once artifacts were produced from commit 43308ab and passed Twine plus isolated installed-wheel import and CLI smoke checks.

clio-relay 1.5.3

Repairs stale Linux CI fixtures for bootstrap recovery, JARVIS wrapper repository binding, and session ownership. Production safety behavior is unchanged.

clio-relay 1.5.2

clio-relay 1.5.2

This patch fixes the two relay defects exposed by the autonomous Ares LAMMPS demo:

• JARVIS bootstrap now binds JARVIS's stable builtin package repository and removes only stale builtin paths proven to come from relay-owned jarvis-cd wheel environments. Operator repositories and their precedence are preserved.
• Registered JARVIS jarvis_run receipts now accept relay's deterministic execution-ID injection while continuing to reject any other argument or contract mismatch.

Live candidate evidence: JARVIS execution jarvis_2bbdf4563f70683c7660957f970c874a, Slurm job 22240, completed at timestep 5000 with structured progress and nine durable artifacts. The desktop teardown preserved scheduler jobs by default.

Focused validation: 12 targeted pytest cases, Ruff, Pyright, package build, and Twine metadata checks passed locally. Full tag validation and trusted PyPI publication run asynchronously after release.

clio-relay 1.5.1

clio-relay 1.5.1

This patch completes the released remote-MCP path introduced in 1.5.0.

• Remote MCP submission now carries the operator-selected cluster route and server registration into the executor for that invocation only. The private snapshot is removed after submission and never changes persistent cluster configuration.
• The same authority boundary covers direct MCP calls, agent-facing virtual MCP tools, and contract refresh.
• Packaged validation retains Windows PROGRAMDATA, which Windows OpenSSH requires to start inside the scrubbed validation environment.

Focused local tests, Ruff, and Pyright pass. A wheel-installed candidate submitted Ares job job_784e7b7e4f544bb2a23fe4560a04a125 as a registered control query with no scheduler action; the real cluster-side Spack MCP returned the installed LAMMPS 20240829.1 package.