Releases · Codesteward/codesteward

Release list

v1.5.0 Latest

Latest

github-actions released this 21 Jul 21:26

v1.5.0

c7e769e

Product session traces (ClickHouse), pull-only deploy stack, DeepAgents/workspace reliability, Graph MCP stdio
fixes, and unified-diff proposed fixes for SCM comments.

Added

Platform ClickHouse product traces — optional install-wide sink (Platform settings). When enabled, every
org dual-writes full session observations (agents, tools, LLM I/O, metadata; secret-redacted, no normal truncation)
for session deep-dive + analytics. Orgs cannot disable ingestion; they may only override TTL days
(GET/PUT /v1/org/trace-ttl). Read paths: GET /v1/org/trace-storage, GET /v1/org/traces/sessions,
GET /v1/sessions/:id/traces. Env: CLICKHOUSE_URL, CLICKHOUSE_USER, CLICKHOUSE_PASSWORD,
CLICKHOUSE_DATABASE, STEW_CLICKHOUSE_DEFAULT_TTL_DAYS.
Traces UI — org-scoped Traces nav (any tenant member when ClickHouse is enabled): list sessions with
stored observations, expand generations / tools / spans with readable I/O (raw JSON optional). Collapsible
help for event types and LangChain internals (RunnableSequence, RunnableLambda, graph spans).
ClickHouse compose overlay — deploy/compose/docker-compose.clickhouse.yml (+ clickhouse/config.d) for
local product-trace testing; configure the sink in Platform → ClickHouse (no preferred env wire file).
Pull-only stack — standalone deploy/compose/docker-compose.stack.yml (optional
docker-compose.stack.swarm.yml for Swarm): published GHCR images, no monorepo / no build.
Re-review GitHub thread resolve — when lifecycle marks prior findings fixed (fingerprint gone after a new
push/session), resolve matching PR review threads via GraphQL when scmCommentId / fingerprint / finding markers
match. Disable with STEW_RESOLVE_FIXED_THREADS=0.
Org Analytics — token totals, average tokens per session, estimated total cost, and average cost per session
(from session tokenUsage / list-price estimates).
Langfuse dual-write / sticky session IDs — full agent + subagent turns dual-written with product sessionId
(org + platform destinations); specialist trace ids scoped per unit/role for parallel runs.

Changed

Proposed fixes are always unified git diffs — specialists are prompted for unified-diff suggestedFix;
extract normalizes plain snippets → ---/+++/@@ hunks; PR comments, session reports, and Findings UI always
render Proposed fix as a ```diff fence (Context blocks still use path language, e.g. go / ts).
Fixes GH comment parsers that previously saw language-tagged fences around diff-shaped bodies.
PR finding Context blocks continue to use GFM language tags from the file path for non-fix snippets.
Finding status mix labels are humanized (false_positive → “False positive”).
DeepAgents review filesystem — FilesystemBackend rooted on the unit clone (virtualMode), path jail +
host/cross-repo rewrite, empty permissions (avoids path must be absolute kills), read-only write/edit refuse,
sandbox tools jailed to unit cwd, prompts no longer leak host repoPath.
Worker Graph MCP — default GRAPH_MCP_COMMAND=/opt/graph-venv/bin/codesteward-mcp; image installs uv CPython
under UV_PYTHON_INSTALL_DIR=/opt/uv-python (readable by uid 1001); entrypoint opens legacy /root/.local/share/uv
when needed; stdio protocol is NDJSON (not Content-Length) for codesteward-mcp.

Fixed

DeepAgents empty filesystem tools — built-in ls / read_file / glob / grep bound to the review clone
instead of an empty in-memory FS; sandbox_ls / smarter sandbox_read; cross-repo mistaken host paths
(workspace/local/ses_…/cross/…) rewritten or refused.
DeepAgents path must be absolute: "." — non-empty DeepAgents permissions forced absolute paths before
our normalizer and aborted specialists under STEW_REQUIRE_TOOL_AGENTS. Permissions cleared; jail/read-only
enforced in the review backend.
Graph MCP spawn … EACCES as steward — uv CPython lived under mode-700 /root; shebang could not load
libpython after privilege drop. Fixed install path + entrypoint chmod.
Worker not claiming jobs after Graph MCP hang — Content-Length framing broke codesteward-mcp NDJSON
init; worker never entered dequeue. NDJSON write/parse + short init timeout; job loop starts even if graph soft-fails.
Cross-repo specialist path probing — unit-rooted tools + prompt/metadata host-path redaction so agents do not
ls session tree prefixes relative to the linked clone.

Upgrade notes

Pull images / chart for 1.5.0:

helm upgrade --install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward \
  --version 1.5.0 \
  --set image.tag=1.5.0 \
  --set ui.image.tag=1.5.0

Optional product traces: deploy ClickHouse (or managed), set platform ClickHouse in UI, or compose
docker-compose.clickhouse.yml. Members see Traces when the sink is enabled.
Workers: rebuild so Graph MCP binary + NDJSON client ship; confirm logs show
[graph-mcp] stdio session ready then [worker] starting Codesteward review worker.
Suggested code fixes: existing DB rows may still store plain snippets; new extracts + PR publish always
emit unified ```diff for proposed fixes.

Container images

docker pull ghcr.io/codesteward/codesteward:1.5.0
docker pull ghcr.io/codesteward/codesteward/ui:1.5.0
docker pull ghcr.io/codesteward/codesteward/keycloak:26.7.0   # tag = upstream Keycloak, not product 1.5.0

App image runs API by default (SERVICE=api). Worker:

docker run --rm -e SERVICE=worker ghcr.io/codesteward/codesteward:1.5.0

Keycloak image tags match upstream Keycloak (theme + realm baked in; no git mounts):

docker run --rm -p 8083:8083 -e KEYCLOAK_ADMIN=admin -e KEYCLOAK_ADMIN_PASSWORD=admin \
  ghcr.io/codesteward/codesteward/keycloak:26.7.0 start-dev --import-realm --http-port=8083

Helm chart (OCI on GHCR)

helm install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.5.0
# or: helm pull oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.5.0

Chart package is also attached to this release as codesteward-1.5.0.tgz.
Docs: Kubernetes quick start in the product documentation site.

Images are signed with cosign keyless via GitHub OIDC.
Verify with:

cosign verify ghcr.io/codesteward/codesteward:1.5.0 \
  --certificate-identity-regexp 'https://github.com/Codesteward/codesteward/.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Assets 5

v1.4.0

github-actions released this 20 Jul 15:50

v1.4.0

aa89a63

Indirect eval / outcome loop, reliable GitHub App multi-install gate reviews, PR status + republish, and per-finding SCM comments.

Added

Automatic / indirect eval (outcome loop) — learn from what users and merges actually do:
- Merge webhooks (GitHub closed+merged, GitLab merge) enqueue jobKind=pr_outcome (no agent pipeline)
- Classifies findings: accepted / fixed / thumbs_up / false_positive / dismissed / unaddressed_at_merge
- Mines agent-miss candidates on sensitive paths changed with no finding
- Gate regret: approve+critical open, or request_changes+only noise
- Durable pr_outcomes + finding_outcomes (migration 015_review_outcomes.sql or file store)
- 👍 creates positive preference memories (mirrors 👎 suppress)
- Preference bag-of-words embeddings filter in noise stack (near 👎 → drop)
- Analytics: split fixAcceptRate / noiseRate / openRate + confidence calibration
  (GET /v1/analytics/address-rate, /v1/analytics/outcomes, /eval-export)
- Offline eval harness consumes production outcome fixtures (packages/evals)
- Outcome consolidator — promote frequent merge outcomes to memories with correct scope:
  repo-only common → repo memory; multi-repo or important (critical/high / gate regret) → org memory.
  POST /v1/analytics/outcomes/consolidate; also runs after each merge outcome job.
  Source outcome_aggregate (auditable on Learnings). No longer writes a repo memory on every single accept.
- Docs — GitHub App / GitLab webhook events for gate + merge outcomes (docs/docs/configure/connectors.md, README).
- GitHub pull_request_review_thread — resolved/unresolved maps comment ids → findings (scmCommentId), writes
  thread_resolved / thread_unresolved outcomes, soft status tags, consolidator feed.
- GitHub security_advisory (+ repository_advisory) — external GHSA → security_advisory outcome + soft pattern
  memory for coverage / FN; optional org promotion when severity is high.
- Manifest defaults include pull_request_review_thread and security_advisory.
PR status comments — webhook-triggered reviews post a progress comment on the PR (“Reviewing / Re-reviewing now…”),
then update it on prepare failure, crash (exhausted retries), or successful completion so GitHub-only readers are not stuck.
Republish to PR — post findings for a finished gate session without re-running agents:
- POST /v1/sessions/:id/publish (reviewer+)
- Sessions UI drawer action Republish to PR
- Diff-aware inline comments + conversation fallbacks with full finding body (suggestion / proposed fix)

Changed

SCM publish posts per-finding PR comments with UI-style bodies (severity, path, explanation, suggestion, proposed fix),
not only a summary review. Default cap raised to 40 findings (STEW_COMMENT_CAP).
Display brand in PR summary / gate check titles: Codesteward (not “CodeSteward”).
Gate review jobs carry headBranch for clone/checkout of the PR tip (not base).

Fixed

GitHub webhook redeliver — same X-GitHub-Delivery id is reprocessed after processed/failed (or stale
received), instead of always returning { duplicate: true } with no new session. Concurrent in-flight claims
still dedupe (~2 min, STEW_WEBHOOK_CLAIM_STALE_MS).
Multi-install GitHub App token pick — mint installation tokens for the repo owner (e.g. scigility) instead of a
stale connector accountLogin: local install. Applied to webhook SCM, clone auth, check runs, publish, and republish.
Fixes clone “Repository not found”, getPullRequest 404 on mentions, and postReview 404 after a successful review.
Workspace clone checkout — git checkout --force -- <sha> treated the SHA as a pathspec and failed with
pathspec '…' did not match any file(s) known to git. Checkout now uses switch --detach / checkout -f <ref>;
fetch fallbacks include pull/<n>/head, head branch, then object id.
PR review comments 422 “Line could not be resolved” — only attach inline comments to lines present in PR diff hunks;
other findings become conversation comments so they are not dropped when a batch fails.
Stuck “Re-reviewing now” comment — completion path now updates the status comment (and notes when SCM publish failed).
Webhook delivery logging — outcomes (ignored reason, session/job ids, publish errors) written to logs / delivery row
so redeliver forensics is not silent.

Container images

docker pull ghcr.io/codesteward/codesteward:1.4.0
docker pull ghcr.io/codesteward/codesteward/ui:1.4.0
docker pull ghcr.io/codesteward/codesteward/keycloak:26.7.0   # tag = upstream Keycloak, not product 1.4.0

App image runs API by default (SERVICE=api). Worker:

docker run --rm -e SERVICE=worker ghcr.io/codesteward/codesteward:1.4.0

Keycloak image tags match upstream Keycloak (theme + realm baked in; no git mounts):

docker run --rm -p 8083:8083 -e KEYCLOAK_ADMIN=admin -e KEYCLOAK_ADMIN_PASSWORD=admin \
  ghcr.io/codesteward/codesteward/keycloak:26.7.0 start-dev --import-realm --http-port=8083

Helm chart (OCI on GHCR)

helm install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.4.0
# or: helm pull oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.4.0

Chart package is also attached to this release as codesteward-1.4.0.tgz.
Docs: Kubernetes quick start in the product documentation site.

Images are signed with cosign keyless via GitHub OIDC.
Verify with:

cosign verify ghcr.io/codesteward/codesteward:1.4.0 \
  --certificate-identity-regexp 'https://github.com/Codesteward/codesteward/.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Assets 5

v1.3.0

github-actions released this 17 Jul 23:34

v1.3.0

1874028

Cloud one-click trial deploys, first-review product tour, and Scorecard publish fix.

Added

First-review product tour — guided spotlight walkthrough (driver.js) after first login:
Models → Connectors → Gate → Findings. Completion/skip dual-written to users.preferences
(productTour.firstReviewStatus) and browser localStorage (hard-refresh safe); replay from
Account → Replay product tour. Closing mid-tour (X) toasts how to restart.
Migration 014_user_preferences.sql; PATCH /v1/auth/me/preferences (all signed-in roles).
Popover follows light/dark theme tokens.
Cloud one-click trial deploys — shared single-VM stack under deploy/cloud/
(nginx edge HTTPS + Keycloak OIDC + API/worker/UI + Postgres). No LLM key at install (Models UI).
Self-signed TLS by default (PKCE / crypto.subtle); optional DOMAIN for cert CN.
- AWS CloudFormation Launch Stack (deploy/cloud/aws/)
- Azure Bicep/ARM Deploy (deploy/cloud/azure/)
- GCP Cloud Shell + gcloud (deploy/cloud/gcp/)
- DigitalOcean Marketplace 1-Click vendor assets (deploy/cloud/do/)

Fixed

Cloud trial deploys (deploy/cloud/): production-shaped single-VM path hardened after Azure pilot:
- nginx edge (HTTPS self-signed) instead of Traefik (Docker Engine 29 API incompatibility)
- SPA OIDC: ignore bake-time localhost issuer off-loopback; empty UI Dockerfile OIDC defaults
- nginx routes /auth/callback to UI (Keycloak is under /auth/*)
- Keycloak Admin issuer parse supports path-based /auth/realms/... (org create)
- first-boot: quoted .env, IaC placeholder sanitization, Azure IMDS, always-HTTPS for PKCE
- Azure cloud-init via Bicep format() + NSG on NIC
Product tour UX — sidebar nav scroll for Models/Connectors; step counter no longer resets on
SPA navigation; “Configure a provider” targets provider API keys (not the stage matrix);
dark-mode button hover contrast on popover controls.
CI — OpenSSF Scorecard job uses only uses: steps when publish_results: true (OpenSSF
workflow verification; shell retry steps caused 400 “scorecard job must only have steps with uses”).
Code scanning — pin Docker base images and Keycloak workflow actions by digest/SHA;
Confluence HTML strip handles </script…> end tags; git clone SHA fetch restricted to object ids +
checkout --; remove empty Keycloak password field from Helm values; drop unused locals flagged by CodeQL.

Container images

docker pull ghcr.io/codesteward/codesteward:1.3.0
docker pull ghcr.io/codesteward/codesteward/ui:1.3.0
docker pull ghcr.io/codesteward/codesteward/keycloak:26.7.0   # tag = upstream Keycloak, not product 1.3.0

App image runs API by default (SERVICE=api). Worker:

docker run --rm -e SERVICE=worker ghcr.io/codesteward/codesteward:1.3.0

Keycloak image tags match upstream Keycloak (theme + realm baked in; no git mounts):

docker run --rm -p 8083:8083 -e KEYCLOAK_ADMIN=admin -e KEYCLOAK_ADMIN_PASSWORD=admin \
  ghcr.io/codesteward/codesteward/keycloak:26.7.0 start-dev --import-realm --http-port=8083

Helm chart (OCI on GHCR)

helm install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.3.0
# or: helm pull oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.3.0

Chart package is also attached to this release as codesteward-1.3.0.tgz.
Docs: Kubernetes quick start in the product documentation site.

Images are signed with cosign keyless via GitHub OIDC.
Verify with:

cosign verify ghcr.io/codesteward/codesteward:1.3.0 \
  --certificate-identity-regexp 'https://github.com/Codesteward/codesteward/.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Assets 5

v1.2.0

github-actions released this 17 Jul 07:09

v1.2.0

29eb72c

Self-host hardening, operator docs site, multi-tenant workers, and release packaging (Helm OCI + SpaceXAI branding).

Added

Documentation site — Docusaurus handbook at top-level docs/ (replaces flat markdown-only docs folder).
Product handbook: why Codesteward, Compose + Kubernetes quick start, install (Compose/Helm),
configure, product UI, pipeline, integrations, ops, security, FAQ. Theme aligned with product UI
(official brand assets); Cloudflare Workers via docs/wrangler.toml.
Helm chart OCI publish on release — helm package + helm push to
oci://ghcr.io/codesteward/codesteward/charts/codesteward (version = product semver); .tgz
also attached to the GitHub Release. Docs: Kubernetes quick start + Helm install from GHCR.
Multi-tenant worker isolation — harden shared workers against cross-org clone reads
(path layout + tool jail + optional hard sandbox + claim affinity). See
docs/docs/ops/multi-tenant-workers.md.
- Workspace layout: {STEW_WORKSPACE_DIR}/{orgId}/{sessionId}
  (STEW_TENANT_ISOLATION=path default; off for legacy flat paths).
- Path jail on agent tools / packing (packages/agents/src/path-jail.ts).
- STEW_TENANT_ISOLATION=strict prefers Docker/k8s sandbox (no host shell for agent tools).
- Org-affine workers: STEW_WORKER_ORG_IDS filters Postgres job claim.
- Graph scope: graph tenant_id = product orgId; rebuild repo_path jailed to session workspace.
Platform queue republish — after optional broker (NATS/Rabbit/Pulsar) message loss,
platform operators rehydrate wake-up depth from Postgres via
GET/POST /v1/platform/queue (+ /republish) or Settings → Platform ops → Job queue recovery.
Embedded Graph MCP in workers — removed standalone graph-mcp service; workers spawn
codesteward-mcp --transport stdio. Shared Neo4j/Janus via GRAPH_BACKEND + NEO4J_URI
(tenant_id = product orgId). No shared clone PVC between graph and workers.

Changed

SpaceXAI rebrand — user-facing "xAI" renamed to SpaceXAI (docs, UI Models, CLI doctor).
Env: prefer SPACEXAI_API_KEY (legacy XAI_API_KEY still accepted). Provider id xai kept for
stored configs; alias spacexai accepted in the model router. API host remains api.x.ai until
upstream changes.
Default container images — Helm defaults use ghcr.io/codesteward/codesteward (+ /ui, /keycloak).
Models — removed unrealistic per-stage env apiKeyRef UI/API surface; org provider keys remain BYOK.

Fixed

CI — require-tools test imports after from node:test; zizmor template-injection in
Keycloak base-update workflow (inputs via env); serialize-javascript override for docs/Docusaurus
HIGH advisory; Hadolint on Dockerfile.node (DL3008 ignore placement + merge consecutive RUNs).

Upgrade notes

Pull images / chart for 1.2.0:

helm upgrade --install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward \
  --version 1.2.0 \
  --set image.tag=1.2.0 \
  --set ui.image.tag=1.2.0

Or from git: ./deploy/helm/codesteward with image tags 1.2.0.

Set SPACEXAI_API_KEY if you used Grok/XAI_API_KEY (legacy env still works).
Multi-tenant production: review STEW_TENANT_ISOLATION and optional STEW_WORKER_ORG_IDS
(docs/docs/ops/multi-tenant-workers.md).
Graph: no standalone graph-mcp Deployment — workers embed MCP; point GRAPH_BACKEND + Neo4j/Janus as needed.
Optional broker: after broker data loss, use Platform ops Republish pending or
POST /v1/platform/queue/republish.

Container images

docker pull ghcr.io/codesteward/codesteward:1.2.0
docker pull ghcr.io/codesteward/codesteward/ui:1.2.0
docker pull ghcr.io/codesteward/codesteward/keycloak:26.7.0   # tag = upstream Keycloak, not product 1.2.0

App image runs API by default (SERVICE=api). Worker:

docker run --rm -e SERVICE=worker ghcr.io/codesteward/codesteward:1.2.0

Keycloak image tags match upstream Keycloak (theme + realm baked in; no git mounts):

docker run --rm -p 8083:8083 -e KEYCLOAK_ADMIN=admin -e KEYCLOAK_ADMIN_PASSWORD=admin \
  ghcr.io/codesteward/codesteward/keycloak:26.7.0 start-dev --import-realm --http-port=8083

Helm chart (OCI on GHCR)

helm install codesteward oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.2.0
# or: helm pull oci://ghcr.io/codesteward/codesteward/charts/codesteward --version 1.2.0

Chart package is also attached to this release as codesteward-1.2.0.tgz.
Docs: Kubernetes quick start in the product documentation site.

Images are signed with cosign keyless via GitHub OIDC.
Verify with:

cosign verify ghcr.io/codesteward/codesteward:1.2.0 \
  --certificate-identity-regexp 'https://github.com/Codesteward/codesteward/.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Assets 5

v1.1.0

github-actions released this 16 Jul 16:20

v1.1.0

911b508

Post–1.0.0 release: parallel review pipeline hardening, install-wide ops visibility, findings quality, security triage, and product docs with UI screenshots.

Added

Parallel specialists + structured rationale → senior verifier — roles on a unit always run
concurrently (Promise.all) with a barrier before the next stage; findings may carry
reasoning (plus evidence.type=reasoning). Verifier is a principal-SWE batch pass
(keep/drop/severity) using rationale + packed context. Migration 013_finding_reasoning.sql.
Session timing ledger — session.audit.timings / metadata.timings: wall clock per pipeline
stage, per unit, rollups (longest stage/unit/specialist, byStageMs, tool time). Session report
Timing / bottlenecks section; worker logs stage=X done … + end-of-job summary.
Live specialist heartbeats — SSE specialist_run started / running (interval
STEW_SPECIALIST_HEARTBEAT_MS, default 15s) / completed / failed. Session blade Live specialists
banner with per-role elapsed timers.
Platform ops analytics — GET /v1/platform/analytics?days=N (platform operators) for
install-wide success rate, p50/p95 latency, stage averages, specialist role stats, worker queue,
tokens. UI: Settings → Platform ops (/settings/platform/ops). Distinct from tenant Analytics.
GitHub Code Scanning SARIF upload — PR gate publishes via code-scanning/sarifs (gzip+base64).
STEW_PUBLISH_SARIF (env → platform runtime → org → default On). Requires code scanning enabled
and security_events: write.
Three-level finding confidence — product confidence, specialist modelConfidence,
optional tokenConfidence (logprobs). UI/SARIF/suggested-fix gate use product confidence.
Migration 012_finding_confidence_layers.sql.
Suggested code fixes — suggestedFix / suggestion / existingCode on findings; Findings UI,
reports, SARIF, PR comments. Min confidence gate STEW_SUGGESTED_FIX_MIN_CONFIDENCE (default 0.75).
Migrations 011_finding_suggested_fix.sql.
Install-wide platform runtime store — GET/PUT /v1/platform/runtime-config; org may only
override suggested code fixes when platform policy is Unset.
Product docs — docs/UI_GUIDE.md (screenshot tour), docs/README.md,
docs/REVIEW_PIPELINE.md, session audit notes; screenshots under docs/screenshots/ (kebab-case).

Changed

Job queue is Postgres-only — removed file-backed FileJobQueue / jobs.json.
DATABASE_URL required for multi-replica safety; NATS/Rabbit/Pulsar remain optional wake-up brokers.
Specialist timeouts — STEW_SPECIALIST_TIMEOUT_MS (default 8m); truncated runs emit coverage-gap
findings (steward.specialist_timeout), audit coverageGaps, UI TIMEOUT ledger — never a silent
clean empty scan. LLM retries: STEW_LLM_MAX_RETRIES + STEW_LLM_REQUEST_TIMEOUT_MS.
Session stage pipeline UI — per-stage durations, live active step, skipped optional stages,
timing bars; audit JSON download only under Review audit (not duplicated on Review report).
Members UI — role capability help; Keycloak vs local create-user copy clarified.
Runtime UI — Unset / Off / On for booleans; platform vs org scope clearly labeled.
Codesteward Graph image — default ghcr.io/codesteward/codesteward-graph.
README — self-host focused; docs links to UI guide + pipeline.
Local sandbox defaults to in-place repo read (STEW_SANDBOX_COPY=1 for full tree copy).

Fixed

GitHub clone host — map api.github.com / GITHUB_API_URL to git host https://github.com
(resolveGithubGitHost); exact hostnames only (no substring SSRF). Hardened clone args
(assertSafeGitArg, -- on clone).
Keycloak first install user — first OIDC JIT user on empty store gets platformAdmin + product
admin (parity with local bootstrap).
Code scanning triage — crypto temp passwords; Confluence CQL/HTML strip; remove unused
vulnerable diff package (GHSA-73rr-hh4g-fpgx); root SECURITY.md; CodeQL quality cleanups.
Keycloak login path — no fallback to local password form when IdP is configured (break-glass
only /login?local=1).
Workspace GC — delete {STEW_WORKSPACE_DIR}/{sessionId} clones after terminal status
(STEW_WORKSPACE_KEEP=1 to retain).
Container permissions — entrypoint chowns data/workspace volumes for non-root steward.
LocalSandbox spawn — handle ENOENT; prefer /bin/bash or /bin/sh; no worker process crash.
Resume UI — failure branding only on terminal status; resume clears prior error.
CI / release — Trivy 0.72.0; Semgrep GCM authTagLength; zizmor cache-poisoning fixes;
multi-stage Docker; drop SaaS-billing image from public CI/release; CodeQL action v4.
GitHub connector icon visible on light theme (currentColor).
Plan-gate UI for audit log / SCIM; SCIM org entitlement; platform GitHub App enforce UX.

Migrations

Operators with Postgres should run migrations through 011–013 (suggested fix, confidence layers,
finding reasoning) if upgrading from 1.0.0:

pnpm migrate
# or: pnpm --filter @codesteward/db run migrate

Upgrade notes

Ensure DATABASE_URL is set (file job queue removed).
Run DB migrations 011–013.
Rebuild/redeploy API, worker, and UI images (or pnpm -r run build).
Optional: set STEW_PLATFORM_ADMIN_EMAILS for additional platform operators under Keycloak.
Helm: set image tag 1.1.0 (chart appVersion updated).

Container images

docker pull ghcr.io/codesteward/codesteward:1.1.0
docker pull ghcr.io/codesteward/codesteward/ui:1.1.0

App image runs API by default (SERVICE=api). Worker:

docker run --rm -e SERVICE=worker ghcr.io/codesteward/codesteward:1.1.0

Images are signed with cosign keyless via GitHub OIDC.
Verify with:

cosign verify ghcr.io/codesteward/codesteward:1.1.0 \
  --certificate-identity-regexp 'https://github.com/Codesteward/codesteward/.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Assets 4

Uh oh!

Releases: Codesteward/codesteward

Release list

v1.5.0

Added

Changed

Fixed

Upgrade notes

Container images

Helm chart (OCI on GHCR)

Uh oh!

v1.4.0

Added

Changed

Fixed

Container images

Helm chart (OCI on GHCR)

Uh oh!

v1.3.0

Added

Fixed

Container images

Helm chart (OCI on GHCR)

Uh oh!

v1.2.0

Added

Changed

Fixed

Upgrade notes

Container images

Helm chart (OCI on GHCR)

Uh oh!

v1.1.0

Added

Changed

Fixed

Migrations

Upgrade notes

Container images

Uh oh!