v0.44.0

/scan goes from untracked work to a ready board in one command, /skills can create and improve a
skill end to end, and the whole tool now speaks English: the agent translates for the user.

Added

• /scan in one command (#738, #731). Scan-Project.ps1 makes a plan and writes nothing to
  GitHub. The user confirms it once, and /scan apply carries it out.

  • What the plan finds:

    • code debt markers;
    • unchecked checklists and pending-section bullets;
    • plan/spec documents.

    A plan document, or any file with 12+ open items, is ONE plan item: its own checklist tracks
    its steps. Skill, agent and template folders are skipped. A title found in several files is one
    item that lists every place. On a real repo this turned 536 noisy rows into 16.

  • What each item gets:

    • a preset type label (bug, feature, chore, docs, refactor, spike);
    • a proposed Priority with its reason;
    • its dependencies ("blocked by #n");
    • an order and a PR batch.

  • What apply does:

    • reuses or creates the board with the standard fields;
    • applies the label preset;
    • creates the issues with Priority, Status Blocked and Task Type Spike;
    • adds a plan epic with native sub-issues and the PR batches;
    • runs fill --auto.

    The plan file records each issue as it is created, so running apply again never duplicates one.

  • Board-Fill maps the new spike label to Task Type Spike.

• /skills create and /skills improve (#739). A pipeline that chains the existing tools:

  1. overlap with installed skills and public prior art;
  2. a scope decision, with the reason;
  3. skill-creator;
  4. a writing-skills RED/GREEN pressure test;
  5. a skill-improver loop that ends only when Skill-Pipeline.ps1 -Verify passes the audit gate
     (no high or medium finding);
  6. the trigger eval.

  A skill name must be kebab-case, so a name can never point outside skills/.

Changed

• English-only tool surface (#751). The conversation still follows the user's language.
  • Output and headings:
    • the 371 Spanish console messages in 32 scripts are translated, and the ratchet baseline is
      now empty;
    • errors, gh labels (Could not <what>) and prompts ([y/n], still accepting s/si) are
      English;
    • the closing-summary headings are English ("What I found / What I did / What is left / What I
      need from you") and the agent writes them in the user's language.
  • Menus: the /board, /skills, /expert and /tools menus are English, and every command
    menu tells the agent to show it translated.
  • Kept in Spanish on purpose: the vocabulary that reads Spanish boards and Spanish input.
  • Status phrases other code reads were renamed on both sides:
    • GATE SIN REVISAR → GATE UNREVIEWED;
    • CI NO SE EVALUO → CI NOT EVALUATED;
    • PASSED / BLOCKED / UNKNOWN.