v0.40.0

The headline is faster CI you can see and preview: ankra pipeline get and
the new ankra pipeline time-saved show the CI time path filters, the
dependency graph and promoted publishes saved; ankra pipeline validate
lists named path sets and, inside a checkout, answers affected: stages from
the same dependency graph the run reads; and --full on pipeline run and
rerun sets the path filters aside when you need everything. Every -o json
change is an added field; no existing key changes.

Added

• The agent skills steer assistants onto Ankra's recommended paths.
  ankra skills install ships all 25 skills reviewed against the current
  CLI and documentation. ankra-platform-principles names seven golden paths
  (ship code, GitOps as the source of truth, the estate, day 2, agent-driven
  work, fast CI, infrastructure as code) and routes every request to the skill
  that owns it. Every other skill opens with a short recommended path saying
  what to do first, which skill takes over next and what to avoid. Commands,
  flags and fields that had drifted are corrected: for example
  registry_url instead of the legacy repository_url, the teardown and
  preflight rules per provider, and backups being available to every
  organisation. Capabilities the skills did not cover are added: fast CI with
  promote_from and the merge train, hub-and-spoke observability, GitOps
  conflicts and repointing, Ankra Cloud Kubernetes, AI review repositories,
  service tokens and GitHub OIDC trust.

• The ankra-terraform skill covers Terraform provider 0.3.0. The
  provider now manages what the CLI can create, change or delete (97
  resources and 42 data sources: every cluster lane, managed Kubernetes,
  node groups, credentials, Helm registries, variables, cluster settings,
  alerts and DNS, applications and pipelines, stack profiles, backups, the
  image registry, members, tokens, and AI, cost and security settings), and
  ankra skills install now teaches assistants those resources, the
  project_id the provider needs in organisations with projects, and which
  changes replace, re-mint or refuse. On provider 0.2.x only the cluster
  resources exist, and the skill says so. It also says the provider's
  organisation override falls back to ANKRA_ORG, the CLI's variable, not
  the ANKRA_ORGANISATION_ID that Ankra Pipelines sets in every step.

• ankra pipeline validate lists the definition's named path sets. A
  definition that declares path_sets: (ankraio/cluster#4862) gets a Path sets block after the violations: each set with every set it includes
  expanded, the sets it includes, and every path list and set that uses it.
  -o json|yaml carries them as a new path_sets field, absent when the
  definition declares none. The ankra-ship and ankra-cicd skills document
  path_sets:.

• ankra pipeline run --full and ankra pipeline rerun --full run every
  stage. The pipeline's path filters (on.<event>.paths, when.paths and
  paths:) are set aside for that one run, and the run records who asked; a
  stage's if: still decides. The same flag is on ankra application pipeline run and rerun. pipeline get now prints a Full run: line
  saying why a run set its filters aside (a --full dispatch or re-run, the
  ankra:full-run pull request label, the organisation's default-branch
  policy or the definition's full_run: block), and -o json|yaml carries
  full_run_source, full_run_requested_by and full_run_message, absent
  for a run that applied its filters. A run skipped because its pushed commit
  carried a skip marker shows its skip_marker error class with what it
  means.

• The ankra-ship and ankra-cicd skills teach affected:. They say when
  to name the Go or pnpm workspace packages a stage tests instead of listing
  globs, that affected: beside a when path filter is refused as paths:
  is, and how a not-affected skip meets CI slots and promotion: it is settled
  without waiting for a slot, and a promoted push counts it when the push
  changes nothing the pull request run's graph was not asked about.

• ankra pipeline validate --changed-file reads the dependency graph for
  affected: stages. A stage that declares affected: (Go packages or pnpm
  workspace packages) is decided at run time by the run's checkout, from the
  dependency graph of the commit it checks out (ankraio/cluster#4866).
  Inside a git checkout, validate --changed-file now runs the same analysis
  against the working tree and shows, per stage, whether the change reaches it
  (graph: affected) or not (skip: not affected (graph), satisfied for the
  stages that need it). --no-local-graph leaves them to the run. The CLI
  carries its own copy of the analysis and says so when it differs from the
  platform's; it never runs a script the API sends. Needs sh, git, awk
  and grep (not on Windows). --changed-file names are now cleaned to the
  repository-relative form (./api/x.go is api/x.go), and one outside the
  repository is a usage error.

• ankra ai memories list and ankra ai memories clear show and delete
  what the assistant remembers about you. Ankra's chat stores short facts
  from your conversations and recalls them into later ones; until now they
  were only visible in the portal. list prints your memories newest first
  (--limit up to 200, and it says when there may be more), and clear
  deletes all of yours in the current organisation after a confirmation
  (--yes to skip it). Both act on your own memories only and take -o json|yaml. A platform without the token routes (ankraio/cluster#4859) is
  named as such instead of reading as "no memories".

• ankra pipeline get shows how much CI time a run saved, and ankra pipeline time-saved totals it. Under the run's steps, a Time saved
  line gives the platform's estimate of the CI time path filters and a
  promoted publish kept the run from spending, with one line per skipped
  stage saying why and what the estimate is based on. A stage with no
  history is shown as unknown and never counted. A stage the dependency
  graph skipped is named as such ("not affected: no changed file is in its
  dependency graph", reason not_affected, ankraio/cluster#4888), and
  time-saved totals it on its own line. -o json|yaml adds the
  record as a new time_saved field and leaves every existing field as it
  was. ankra pipeline time-saved totals the organisation, or one pipeline
  with --application or --repository, over the last 30 days (--window 7d for 7). A platform without the routes (ankraio/cluster#4863)
  prints nothing extra, and time_saved is null.