Wisp 0.2.0rc5 release decision and checklist

Status: release preparation, October 14, 2026. This PR prepares the candidate; it does not create v0.2.0rc5, publish packages, or change the stable 0.1.0 installation pin. At preparation time, RC4 (published October 14, 2026) is the latest candidate. The RC4 checklist records the decisions RC5 keeps.

Candidate decision

RC5 is a fix-only candidate on top of RC4. It keeps RC4’s platforms (native wheels for macOS arm64 and Linux glibc 2.28+ x86_64, plus a pure wheel), its live RPC contract, and its session format. The version pair is Python 0.2.0rc5 / Cargo 0.2.0-rc.5 and must match exactly.

It fixes two problems found by checking the published RC4 install:

• Stdio MCP servers failed to start in every public RC3 and RC4 install (HY-236). Public installs resolve mcp 2.3.0, where a private helper Wisp called became async; the HTTP transport’s shutdown deadline was also bypassed from mcp 2.2.0. CI did not see either, because uv.lock pins mcp 2.0.0. RC5 no longer depends on those helpers, and a new scheduled dependency-drift workflow tests the newest dependency versions that public installs receive.
• wisp update could drop the Rust TUI (HY-235). When uv’s default Python is older than 3.12, uv tool install selects the pure wheel. wisp update now passes the running Python version, and the documented install commands add --python 3.12.

No session, event, or protocol format changed since RC4, so RC4 and RC5 read each other’s sessions. RC3 still cannot open sessions in which RC4 or RC5 saved a response with a prompt-cache key; see the RC4 checklist.

Prepublication acceptance

• The exact PR head has terminal green CI and no actionable review threads.
• Python formatting, lint, configured mypy, full tests, and the live RPC schema check pass.
• The dependency-drift workflow passes on the newest allowed dependencies.
• Rust formatting, Clippy, workspace tests, build, and the Python/Rust handoff tests pass. The macOS handoff step still runs with continue-on-error (HY-29); check its failure annotations.
• mdbook build succeeds.
• The release workflow’s dry run verifies version parity, one sdist, one pure wheel, two native wheels, wheel content parity, checksums, SBOMs, and the live RPC schema bundle.

Publication and public-install acceptance

After this preparation PR merges, publishing requires a separate v0.2.0rc5 tag through the existing release workflow.

• The tag points to the accepted commit, and every release job succeeds, including attestation, PyPI publication, and the GitHub release.
• PyPI and the GitHub release have the four distributions, with matching SHA-256 hashes, and gh attestation verify succeeds for each.
• Outside the source checkout, uv tool install --force --python 3.12 "wisp-ai==0.2.0rc5" on each claimed native target installs the native wheel: wisp --version reports 0.2.0rc5 and wisp-tui --version reports 0.2.0-rc.5.
• With a stdio MCP server configured (for example npx -y exa-mcp-server), the published install starts it and registers its tools. Record the mcp version the install resolved.
• The pure wheel in a clean environment: print, JSON, SDK, and RPC work; interactive wisp reports the actionable missing-TUI error.

Rollback

If RC5 regresses, exit Wisp and install RC4 with uv tool install --force --python 3.12 --with "mcp==2.0.0" "wisp-ai==0.2.0rc4". The mcp pin keeps stdio MCP servers working in RC4. RC4 opens RC5 sessions. Check wisp --version and session resume before relying on the downgraded install, and never delete or hand-edit sessions as a rollback step.