v1.0.2

kubetective 1.0.2

Keyless-signed with cosign (sigstore); SBOMs attached to the archives and
attested on the container image. Verification instructions:
SECURITY.md.
See the CHANGELOG for what shipped.

Changelog

• e8a98cd Formula: v1.0.1
• 5f9ad61 Merge main into feat/version-output-formats
• 7d9352e Merge pull request #19 from yunasora/feat/version-output-formats
• b07308a feat(calibration): five live hard-set scenarios, and the four bugs they found
• 1389b61 feat: add machine-readable version output
• 0db0745 feat: add machine-readable version output
• e606692 feat: add machine-readable version output
• 6f17852 feat: add machine-readable version output
• 973fe70 fix(docs): repair the brew install one-liner and gate it
• 8b79ee0 fix(release): correct the verify job's asset names and image tag
• 39a8f46 fix(server): reject path-like incident ids from REST and MCP
• 4ab6649 fix: live-cluster false positives, alert namespaces, opt-in calibration

v1.0.1

kubetective 1.0.1

Keyless-signed with cosign (sigstore); SBOMs attached to the archives and
attested on the container image. Verification instructions:
SECURITY.md.
See the CHANGELOG for what shipped.

Changelog

• c5881c6 Fix typos and improve clarity in README.md
• 35f500c Formula: v1.0.0
• fca2a18 Update demo description in README
• 2ec394c demo: cursor now tracks the real text cell width while typing
• 4535303 demo: draw blank separator lines instead of skipping them — no dead rows
• 742ea62 demo: hold the final screen ~4s so readers can see the verdict
• 08a81e5 demo: keep the verdict on screen ~4s via a long final frame; add done line
• 661c1d6 demo: smooth 1-char typing frames and a distinct white cursor block
• cf7db71 demo: tighten vertical height (remove dead bottom margin, 27px line height)
• 270c5e1 demo: zoomed terminal fonts, better palette, terminal-width wrapping
• a8ac6f2 feat(benchmark): add mutation, noise and redaction gates plus a hard set
• d016f9c feat(cli): kubetective scenario new — draft a benchmark scenario from a recording
• f99fdaf feat(redact): add kubetective sanitize for sharing incident records
• 2a36979 feat(release): keyless cosign signing, SBOMs, and a verification job
• 911cec3 fix(collect): anchor git evidence to the investigation window, not wall time
• 33d0725 fix(diag): stop a unit test overwriting the real user config
• 3fe6dc3 fix(memory): stop every investigation matching itself as a past incident
• 002f43e fix(redact): rewrite nested payloads and image references in free text
• 76fa2db fix(release): give goreleaser its own Dockerfile; drop krew
• 7a8cd8e fix(release): version gate must tolerate the formula gap on a tagged build
• 315cbd0 fix(score): refuse calibration fits the benchmark cannot justify
• 36b6c42 fix: retry the deployment investigation until the crashloop hypothesis is captured
• 1633327 fix: smoke gate waits for the crash-loop evidence before the actions E2E (flake)
• aa888f8 fix: smoke retries the pod investigation until the record shows the crash state
• 4268464 fix: version gate tolerates release-in-flight formula gap; changelog notes both gate fixes
• 255157e refactor(analyze): document the weight scale, move status card onto the interface
• 43c3bc6 scenario: add the first recording from a live cluster

KubeTective v1.0.0

v1.0.0 GA declaration. Three checklist items are waived by owner
decision for this release, tracked for the next release:

1. Calibration at scale: 15 ground-truth points ship today, not 30
   (the evaluation gate still passes: ECE within threshold, LOO-validated
   temperature).
2. Signed releases: cosign keys + CI signing wiring remain open; the
   goreleaser config is ready for them.
3. External security audit: a self-review is published instead
   (reports/security/review-v1.0.md); third-party review is the first
   stop on the next release.

Added

• v0.8 - Loki log collector (--loki-url / KUBETECTIVE_LOKI_URL):
  log evidence for the adaptive loop from Loki, silent degradation;
  incident memory v1 (kubetective incidents similar <id>, Jaccard
  overlap of failure shapes, "seen this before" note on every
  investigation); read-only web UI (kubetective serve: GET /,
  GET /incidents/{id}); self-telemetry (GET /metrics, expvar);
  GitHub Actions CI (build/vet/test/benchmark/evaluate gates);
  dns-events-only scenario (suite: 16 scenarios).
• kubetective incidents search: filter the incident store by target,
  cluster, analyzer, minimum severity, and time window
  (--since/--until), newest first (--limit); the linear-scan
  JSONL precursor to the indexed memory (roadmap v2).
• Container smoke test covers the full action-safety loop: preview plans
  the restart-pod action, --apply without --yes is rejected, and an
  approved apply deletes/recreates the pod and appends the audit record -
  all under the least-privilege RBAC (records persist across jobs via a
  PVC). Any gate failure fails the build.
• Record versioning contract: record.Load now rejects records written
  by a newer engine loudly (upgrade instead of silent mis-reads), with
  upgrade tests pinning older/no-meta/meta-version reads.
• Operational maturity: kubetective serve shuts down gracefully on
  SIGINT/SIGTERM and exposes pprof behind a --pprof flag.
• Dependency license report target (make license-report); CODEOWNERS
  and a pull request template.
• CLI target parsing accepts fully-qualified kind/namespace/name
  targets, the form incident records store.
• Completion webhook, opt-in and HMAC-secured (webhook_url +
  webhook_secret): a signed notification after every investigation in
  X-Kubetective-Signature; receivers must verify before parsing.
  Notification failure never fails the investigation.
• Structured logs (--log-format json|text or KUBETECTIVE_LOG_FORMAT)
  for investigations, HTTP requests, and action applies; byte-identical
  output when off.
• Evaluation report as a committed artifact (reports/evaluation/latest.md,
  make report): Action safety section + unsafe-action-rate = 0 gate.
• Alert integrations (inbound): kubetective alert pagerduty|grafana|slack
  investigates from a webhook payload (stdin or --file), with zero
  API keys. Grafana kubernetes_* alert labels (unified + legacy
  shapes), PagerDuty incident titles / impacted services / Events API v2
  details, and Slack slash-command text (deployment/checkout since=2h)
  name the target; the since= window is honored (precedence: flag >
  payload > config > 30m). Payloads without a Kubernetes target fail with
  a readable message instead of guessing. Runs the same recorded,
  webhook-notified pipeline as investigate.
• The uninstall test (roadmap v1.0 #10): hack/uninstall-test.sh
  (CI job uninstall-test) clones fresh, builds per the docs, verifies
  both binaries, runs the offline benchmark, and asserts doctor fails
  without a cluster with an actionable message - never a stack trace.
• Docs site: site/ (single-page static HTML, no framework)
  published to GitHub Pages by .github/workflows/pages.yml at
  https://gledilami.github.io/kubetective/ - install, quickstart,
  integrations, scoring, ops, security.
• krew submission script: hack/submit-krew.sh completes the
  plugin-index submission in one command - verifies release assets
  exist, fills the per-platform sha256 values, commits the manifest,
  and opens the krew-index PR. It refuses to run without real release
  assets (no placeholder hashes can ship).
• Security hardening (from the v1.0 self-review, the only finding):
  incident records are owner-only on disk (files 0600, store dir 0700;
  previously umask-dependent) with a regression test.

Fixed

• incidents similar --cluster scoping was bypassed by incidents that
  carry no cluster id; untagged (legacy) records no longer leak into
  scoped queries.
• Container smoke gate: --apply outside the approval gate no longer
  selects the wrong recommended action when a preview plans several.
• gofmt, go vet, and staticcheck clean across the tree.