0.5.11

[cleverhoods]

Actionability release: one headline Quality score, leverage-based triage, a
queryable rules CLI, variadic check targets, and the Imported/Referenced
link-reach split — plus Windows/npx hardening and a batch of correctness fixes.

Command surface

• ails --help reorganized into four intent panels — Get started (check),
  Explore (explain, rules), Account & setup (auth, config),
  Maintenance (install, update, version) — each command described in one
  imperative clause.
• Root-level --version / -V prints the version; ails version still gives the
  full install-method readout. -h works everywhere as a --help alias, and
  shell completion covers ails check <target>, --agent, and
  ails explain <id-or-slug> (enable via ails --install-completion).
• Per-capability checks: ails check takes variadic typed targets, so you can
  point it at a specific harness capability instead of the whole project —
  capability:name (ails check skill:backlog), a bare capability noun
  (ails check skills, ails check agents, ails check rules,
  ails check memories), or a path. Targets are mixable; no target = whole-project
  scan. The capability vocabulary is read from the detected agent's harness config
  (framework/rules/<agent>/config.yml), so each agent's real surfaces — skills,
  subagents, rules, memory — resolve per agent rather than from a fixed list.

Scoring & output

• Single Quality score (0–10) leads the summary — the analysis service's
  verdict on how well-formed your instructions are, shown verbatim. Structural
  completeness and content dropped past an agent's hard size cap fold in as a
  delivery factor, so a file that loses instructions can no longer read as clean.
• Per-finding leverage tier (gate_mover / conditional / cosmetic)
  drives inline triage: each file panel leads with what moves the score and
  collapses the rest into a single +N lower-priority row (-v to expand).
• Each finding shows a → remediation line with fix text written for that
  specific instruction.
• -f json gains top-level quality and level, a per-file regime block, and
  a per-finding leverage tier. Existing severity / score / violations
  fields are unchanged.

Rules CLI

• New ails rules — list (with repeatable --capability, --agent,
  --severity; text / md / json output), agents, and capabilities.
  The md form pipes straight into an authoring agent's prompt.
• ails explain <id-or-slug> accepts a rule ID or slug for rule-detail browsing.

Check targets & surfaces

• Variadic typed targets: capability:name (skill:backlog), bare capability
  nouns (skills), or paths — mixable; no args = whole-project. A leading
  Windows drive letter is treated as a path.
• Imported vs Referenced split: @-imported markdown counts toward Quality as an
  Imported surface; [text](path)-linked files surface as Referenced
  (findings only, kept out of the score). Requires generic_scanning: true.
• exclude_files config key / --exclude-files flag to drop individual files
  from discovery.
• --fix added as an alias for --heal; the standalone ails heal verb is
  folded into ails check --heal.

MCP / plugin

• Tool surface trimmed to validate, preflight, explain. validate now
  accepts single-file path targets and returns a structured needs_install
  payload when the framework is absent.

CI & GitHub Action

• The action's score output and min-score gate read the real quality
  verdict from the check JSON; an offline run leaves the score empty and the gate
  logs a warning and skips instead of failing on a fabricated number.

Platform & fixes

• Windows: forces UTF-8 stdout/stderr so the scorecard and -f md glyphs no
  longer crash under cp1252 (npx users).
• Resident embedding + spaCy models unload after an idle window so a background
  ails process no longer pins memory indefinitely.
• npx / MCP uvx invocations use --refresh-package reporails-cli instead of a
  blanket --refresh, ending the per-run dependency-graph re-resolution.
• Numerous discovery, classification, single-file-scan, and symlink fixes — see
  CHANGELOG.md for the full list.