mistral CLI v0.4.1

Added

• mistral apps capability (init, add, update, remove) now 3-way merges
  each generated file instead of overwriting it, so manual edits to Helm values,
  compose overlays, route aggregators, and apps/*/pyproject.toml survive the
  next wire operation. Unresolvable overlaps are written as conflict markers
  rather than silently discarded.
• mistral apps capability merge [--continue | --abort] reconciles the generated
  files left in conflict by a merge.

Fixed

• CLI reference docs now include the fully gated skills, prompts, and evals
  command groups. Root command groups are always registered and disabled ones are
  filtered at runtime, so the reference generator sees them.
• Feature-flag docs use the real mistral_cli_ gate-key prefix and cover the
  ai-registry and evals disciplines; the snapshot filename, refresh budget,
  identity-hash contents, and ungated-command list are corrected.

mistral CLI v0.4.0

Added

• mistral evals optimize <file> — turn an existing eval file into an optimization
  run that searches for better prompts and parameters (v0; optimizes .py eval
  files). Optional flags:
  • --params — comma-separated params to optimize (e.g. instruction,temperature:0..1);
    defaults to all System.params.
  • --steer — a free-text goal that steers the optimizer's reflective mutator
    (e.g. "reduce hallucinations when the context lacks the answer"), max 4000 chars.
  • --name / --description — human-friendly metadata for the optimization;
    auto-generated from the goal when omitted.
• Install via plain curl, with optional version pinning.

Fixed

• apps init checks that bun is on PATH before shelling out to it, failing with
  a clear error instead of crashing.

mistral CLI v0.3.0

Added

• mistral login, mistral logout, and mistral whoami with keychain-backed
  OIDC token storage and browser sign-in (RFC 8252 loopback + PKCE).
• mistral apps evals for running evaluation suites from the CLI.