chore(release): cut v0.5.0 (#69)

chore(release): cut v0.4.1 (#66)

chore(release): cut v0.4.0 (#64)

* doc: mark merged plans complete with PR links

Four plan docs were left with `status: active` after their PRs merged.
Flip to `status: complete`, add `completed:` dates matching merge dates,
and link the `pr:` URL — matching the frontmatter shape used by other
completed plans that carry a PR link.

- coverage-check skill (#32, merged 2026-04-08)
- language-detection architecture (#50, merged 2026-05-14)
- track-2 observability (#59, merged 2026-05-16)
- mcp language arg (#63, merged 2026-05-19)

* chore(release): cut v0.4.0

chore(release): cut v0.3.1 (#56)

* chore(release): cut v0.3.1

Patch release with the symlink-canonicalisation fix from #55. Bumps
Cargo.toml to 0.3.1 and rotates the CHANGELOG `[Unreleased]` block.

* doc(release): bump README install snippet to v0.3.1

Mirrors the v0.3.1 cut. The parameterised install snippet uses a literal
`VERSION=...` example so users copy-pasting verbatim land on the latest
stable artifact. Until `release-prep.sh` learns to substitute this, it is
a per-release manual step (see the v0.1.0 bump for prior art).

chore(release): cut v0.3.0 (#54)

* doc: capture Rust edition-2024 env-var test pattern + `home_dir` gotcha

Documents two findings that surfaced during the etcetera XDG path
resolution refactor (#52):

  1. Rust 1.85+ `std::env::home_dir` falls back to `/etc/passwd` via
     `getpwuid_r` when `$HOME` is unset, instead of returning `None`.
     Crates that delegate (etcetera, dirs, directories, xdg) inherit
     this. Pre-check `$HOME` yourself if your CLI promises an explicit
     "HOME is unset" error path.

  2. Edition 2024 marks `std::env::set_var` as `unsafe`, and parallel
     tests race on shared env vars. Use the `temp-env` dev-dep
     (`temp_env::with_vars`) for safe, serialised env mutation in
     unit tests — keeps coverage off the `unsafe_code = "deny"` blast
     radius and avoids hiding invariants behind `#[ignore]`'d
     integration tests.

Lands in `docs/solutions/best-practices/` (knowledge track,
`problem_type: best_practice`) with cross-references to the
sibling insta-snapshot doc (same toolchain update, different
testing surface) and PR #52 for the originating context.

* chore(release): cut v0.3.0

chore(release): cut v0.2.0 (#49)

Rotates [Unreleased] into a dated [0.2.0] section in CHANGELOG and
bumps the Cargo.toml version from 0.1.0 to 0.2.0 (Cargo.lock follows).
Generated by `just release-prep 0.2.0`; `just ci` passed the same
500/4-sandbox-only-fail split that every recent run has produced.

Highlights vs v0.1.0:

- Universal-pattern predicate (`applies_when`) and engine/adapter
  split (Track 1)
- Edge-case-handling umbrella closed (NFC slug collisions, no-HEAD
  progress line, lossy-path warning, missing-git regression test)
- Effective-empty knowledge directory warning
- Knowledge database schema bumped to v3 (one-way ALTER TABLE
  migration; see Notes)

See CHANGELOG.md for the full set.

chore(release): cut v0.1.0 (#38)

* chore(release): cut v0.1.0

First stable release. Promoting `v0.1.0-alpha.1` to stable after end-to-end
validation of the release pipeline (build, publish-gate, deployment-policy,
verify). No user-facing changes from the prerelease.

- `Cargo.toml`: `0.1.0-alpha.1` → `0.1.0`
- `CHANGELOG.md`: `[Unreleased]` rotated to `[0.1.0] - 2026-05-01` with the
  promotion note pointing at `[0.1.0-alpha.1]` for the full feature list.

Generated by `just release-prep 0.1.0`. The script now invokes `just ci` as
its final step (per ci/release-followups), which passed against the bumped
tree — first real-world exercise of the structural fix that closes the
"stale CI before release-prep" gap from the alpha.1 cut.

* doc(release): bump README install snippet to v0.1.0

The parameterised install snippet uses a literal `VERSION=...` example. Bump it
to the version this release ships so users copy-pasting verbatim land on the
stable artifact rather than the alpha prerelease.

Long-term fix: bake `VERSION=...` substitution into `release-prep.sh` so the
script handles this automatically per release. Queued as a follow-up.

chore(release): cut v0.1.0-alpha.1 (#36)

* chore(release): cut v0.1.0-alpha.1

* test(release): redact version field in initialize_response snapshot

The `initialize` MCP response interpolates `env!("CARGO_PKG_VERSION")` into
`serverInfo.version`. The inline snapshot in `server::tests::initialize_response`
pinned the literal string `"0.1.0"`, so every release-prep version bump broke the
test until the snapshot was re-accepted — a recurring footgun in the release
workflow that would have fired on this very release.

Use insta's redaction support to substitute `[VERSION]` into the snapshotted
value at the `.result.serverInfo.version` path. The snapshot is now structural
(field exists, shape preserved) rather than literal. Enables the `redactions`
feature on `insta` (transitively pulls in pest + sha2; no new license categories
needed in deny.toml).

Also adds `*.pending-snap` to `.gitignore` so future failing snapshot tests
don't leave a dirty tree (insta writes these next to the source file when a
snapshot needs review).