v0.5.0 — CI Validation

CI validation (F5)

The MVP is now self-validating. Every push to main and every pull
request runs cargo fmt --check, cargo clippy --all-targets -- -D warnings, cargo test --all-targets, and cargo build --release
on both stable and beta toolchains.

CI runs

• Run #1 (commit d5dbd53): 1m 3s — both stable and beta ✅
• Run #2 (commit 1007398): 1m 4s — both stable and beta ✅
  • Added FORCE_JAVASCRIPT_ACTIONS_TO_NODE24=true to opt into Node
    24 early, before the 2026-06-16 forced default and 2026-09-16
    removal. Runner is on Node 24 now; actions/checkout@v4 works.

Workflow: https://github.com/LOUST-PRO/loust-llm-mempipe/actions/workflows/ci.yml

Workflow design

• Matrix: stable + beta, fail-fast: false (so a beta breakage
  is visible even if stable passes).
• Cache: Swatinem/rust-cache@v2 (registry + target/).
• Permissions: contents: read only. No secrets, no write tokens,
  no packages: scope.
• Security: zero ${{ github.event.* }} interpolations in run:
  blocks. The only ${{ }} reference is matrix.rust which resolves
  to static "stable" / "beta" values declared in the workflow
  file, not external input. No command-injection vectors.
• Supply chain: actions pinned by major tag (@v4, @v2,
  @master for dtolnay/rust-toolchain).

What CI catches

• Formatting drift (cargo fmt --check)
• Lint regressions, even in tests (cargo clippy --all-targets)
• Test failures on either toolchain (catches features that stable
  has but beta doesn't, or vice versa)
• Release build breakages (the path cargo install exercises)

Final MVP status

Phase	Scope	Status
F0.1	Pre-publish audit	✅ done
F0.2	Org hardening	✅ done
F1	Skeleton + contracts	✅ v0.1.0
F2	ChatGPT adapter MVP	✅ v0.2.0
F3	Pipeline core	✅ v0.3.0
F4	CLI ergonomics	✅ v0.4.0
F5	CI validation	✅ v0.5.0
F7	Reddit post	❌ cancelled

cargo install loust-llm-mempipe ships you the working MVP.