Skip to content

Releases

v0.6.0

Choose a tag to compare

@github-actions github-actions released this 11 Nov 23:19
a13c68c

Summary

ARTEX v0.6.0 makes a finding provable, hardens the agents against the content they read, and adds an opt-in approval gate for targets. The target gate ships OFF by default and through the API only, so existing deployments keep their current behavior after upgrading.

Full list: CHANGELOG.md · compare v0.5.1...v0.6.0

Highlights

  • Verified findings (S1): findings carry a POE level and a verification status. verified requires a bound proof or verification response, a declared POE level of 3 or more, and every evidence quote of 8 runes or more found verbatim in a bound body. A retest verdict can upgrade a finding, and the findings UI shows the badge, the filter and the note (#162, #163, #164, #169).
  • Target gate (S3b phase 1 and step 2.1): opt-in per task or conversation, API only, OFF by default. Approvals pin host:port and the resolved IPs. WebFetch and direct custom http calls connect only to pinned or exempt addresses on every redirect hop, with no direct retry in a gated scope (#177 to #186).
  • Agentic hardening (#124): code-owned untrusted-content rules for the planner and the compactor, a planner goal anchor, goal_met and prove_goal guards, truthful hint origins, and the [built-in] deny rules applied to the planner and main-agent Bash (#189, #190).
  • Intercept audit: approvals record decided_by, and the [model]/[Model] prefix bug no longer stores judge decisions as unknown (#159, #165).
  • Notifications: configurable summary and asset limits, plus Feishu digest overflow handling (#170 to #174).
  • Docker supply chain: recon tools verified against their release checksums, sqlmap pinned to tag 1.9.9 at a fixed commit, Node 22.23.4 verified against SHASUMS256.txt, and the Playwright packages pinned (#188).
  • Intent claim leases: a heartbeat, a TTL and an owner token stop a stale worker from overwriting a result (#176).
  • Smaller fixes: custom User-Agent per LLM profile, the custom-tool proxy scheme fix, and the list_task_credentials schema fix (#155, #187, #130).

Upgrade notes

  • No manual migration: schema changes are idempotent at startup. Existing findings stay unverified, and the new optional fields default to empty.
  • The target gate is off by default. With it off, behavior is unchanged. With it on, an uncovered target waits for an approval.
  • goal_met refuses while a goal is open, and prove_goal accepts a fact only with confidence=observed.
  • The planner and the main agent now run the [built-in] deny rules (destructive commands, delete-style endpoints). User rules and the judge still do not apply to them.
  • Findings gain counterevidence and confidence_rationale, optional and never blocking. Existing databases receive the schema property at startup.
  • 0.5.x is no longer supported. Fixes target the latest release line only; upgrade to 0.6.0.

Known limits

  • The target gate reads the targets a call declares. A subprocess that builds its target at run time is not seen, and a call through a proxy is not bound yet (recording-proxy binding is step 2.2).
  • URL rules match the URL as written, with no normalization; use a domain, IP or CIDR rule to keep an agent away from a host.
  • Fact provenance arrives in 0.7.0. Until then, prove_goal trusts the confidence the model declared.
  • The new add_hint and prove_goal tool descriptions reach fresh databases only; the server-side refusals apply everywhere.
  • The prompt rules are advisory to the model. The enforced parts are the goal_met/prove_goal handlers, the built-in deny rules and the target gate.
  • No Docker image is published for this release. Build from the checkout with docker compose up -d --build, or use a binary from the assets below.

Contributors

Thanks to @p091 for the detailed Anthropic thinking report (#181), fixed in Norma v0.4.7.

Download the release binaries below: five platform zips plus SHA256SUMS. On Docker, the image is built from source, and ./update.sh handles the checkout and the rebuild.

v0.5.1 (unsupported)

Choose a tag to compare

@github-actions github-actions released this 05 Nov 17:20
ce279d7

⚠️ Unsupported release. Upgrade to the latest release: https://github.com/BaaRai/ARTEX/releases/latest

See CHANGELOG.md for the list of changes in this release.

v0.5.0 (unsupported)

Choose a tag to compare

@github-actions github-actions released this 04 Nov 00:29
fc35ddd

⚠️ Unsupported release. Upgrade to the latest release: https://github.com/BaaRai/ARTEX/releases/latest

Summary

ARTEX v0.5.0 adds Ollama support, requires a one-time setup token before anyone can claim a fresh instance, removes startup secrets from the environment that agent shells and tools inherit, and makes the Docker image build from a fresh clone. It also moves to Go 1.26.10 and Node 22 and adds race-detector and shellcheck runs to CI. Full list: CHANGELOG.md. Entries marked "direct commit" landed on main without a pull request.

Upgrade note: first-run setup now asks for a setup token printed in the server log, and Docker installs publish port 8787 on 127.0.0.1 by default. Set ARTEX_BIND_ADDR=0.0.0.0 in .env, or use a reverse proxy or an SSH tunnel, for remote access.

New features

  • Ollama support: an ollama LLM profile format for a local or LAN Ollama server. No API key is needed, OPENAI_API_KEY is never inherited, the base URL defaults to http://127.0.0.1:11434/v1 and is normalized to the /v1 root, and the profile works in the failover pool, task LLM chains, model listing and connection tests (#128).
  • Setup token for first-run setup: while no admin password is set, the server generates a random 128-bit one-time token at startup and prints it to its log (docker compose logs artex | grep "Setup token"). The /setup page asks for it and it is cleared once the password is stored (#137).
  • ARTEX_BIND_ADDR sets the address docker-compose.yml publishes the web UI on; the default is 127.0.0.1 (#137).

Security fixes

  • First-run claim: anyone who reached port 8787 on a fresh install before the operator could set the admin password. POST /api/auth/init now requires the setup token (constant-time comparison, 403 when missing or wrong), and Docker installs bind 8787 to loopback by default (#137).
  • Startup secrets removed from the environment: ARTEX_PG_DSN, ANTHROPIC_API_KEY and OPENAI_API_KEY are unset inside server.New after they are read and before the MCP discovery goroutine and the task resume loop start, so agent shells, MCP stdio servers and custom tools no longer inherit them (#136). New did not apply its options, so the scrub from #136 did not run in the binary until #146 fixed it (#146).
  • Saved API keys go only to their own endpoint: POST /api/llm/test and POST /api/llm/models used a saved key for any base_url in the request body. A saved key is now sent only when the request's provider and normalized base URL match the ones it was saved for (#142).
  • ARTEX_LLM_PROVIDER is matched case-insensitively: a value such as Ollama fell through to the Anthropic default and sent ANTHROPIC_API_KEY to the configured base URL (#140).

Bug fixes

  • Ollama base URL on the environment path is normalized to the /v1 root like the saved-profile path, so an Ollama server configured through ARTEX_LLM_BASE_URL no longer answers 404 on chat (direct commit 1c35c9c).
  • A malformed config.json is an error with the file path, instead of being treated as missing (#141).
  • install.sh escapes the values it writes into config.json and .env, so passwords containing ", \, | or & no longer corrupt them; the port must be numeric (#141).
  • scripts/backup.sh --keep 0 is rejected instead of deleting every dump, including the one just written; values with a leading zero are rejected too (#139).
  • backups/ is ignored again in .gitignore, so database dumps no longer show up as untracked files (#138).

Build and CI

  • The image is built from the checkout instead of pulled from the defunct autumn27/artex namespace; install.sh and update.sh build it locally (direct commit 1ed154e).
  • A fresh clone builds the image: the Dockerfile is now a multi-stage build (Node 22 frontend export, Go 1.26 build.sh with the release job's flags, runtime image) and no longer needs a prebuilt dist/ binary (#147).
  • update.sh stops on a diverged clone with recovery commands instead of rebuilding the old tree (direct commit 1ed154e).
  • CI runs go test -race on every package with tests except ./server, and shellcheck -S warning on install.sh, update.sh, start.sh and scripts/backup.sh (#144). The race run has its own database (direct commit 0477751).

Upgrades

  • Go 1.26.10 (was 1.26.3) and golang.org/x/net v0.62.0 (was v0.58.0), with x/crypto 0.58.0, x/text 0.43.0, x/sys 0.49.0, x/sync 0.24.0, x/mod 0.41.0 and x/tools 0.51.0 (#145).
  • Node 22 in the runtime image instead of the end-of-life Node 20 (#143).

Known issues

  • The JWT signing key is not in the Docker volume. The server writes jwt.key next to the executable (/app/jwt.key in the image), not in /app/data as the Dockerfile and docker-compose.yml comments say. Recreating the container (for example docker compose up -d --build after an update) generates a new key and signs every user out. A key found at data/jwt.key is moved to /app/jwt.key on start and removed from the volume.
  • The secret scrub has limits. A process running as the same user can still read the server's /proc/<pid>/environ (and, in Docker, that of PID 1, start.sh). Other variables that may carry credentials, such as ARTEX_LLM_PROXY, are not scrubbed.
  • The LLM test and model-list endpoints still take proxy from the request, so a saved key can pass through a request-supplied proxy when the base URL is plain HTTP.
  • Postgres passwords with @, /, # or ? break the ARTEX_PG_DSN that docker-compose.yml builds by interpolation, and Compose interpolates $ in .env.
  • The Docker build context includes local secrets: .dockerignore does not exclude .env, config.json or jwt.key, so the build stage's COPY . . puts them in a build-stage layer of the local builder cache. They are not copied into the final image.
  • Image supply chain: recon tools are downloaded at pinned versions but not checksum-verified, sqlmap tracks its default branch, and the Playwright packages are installed at @latest. Draft #127 proposes checksum verification.
  • No Docker image is published for this release. The release workflow's docker job only runs when DOCKERHUB_ENABLED is set. Build the image from the checkout with docker compose up -d --build.
  • CI gaps: ./server does not run under -race; there is no govulncheck step and no docker build job. Four tests skip themselves (see #9).
  • Reported, not fixed in this release: list_task_credentials emits an invalid "required": [""] schema that strict LLM providers reject (#130), and the MCP Streamable HTTP client fails against servers that only support protocol version 2025-11-25 (#131, fix proposed in #132).

Contributors

  • @glauberportella: thank you for Ollama support (#128), a careful change across the provider, profile storage, failover pool, task chains and settings UI that keeps the profile keyless without ever falling back to a cloud key or endpoint.
  • @BaaRai: thank you for authoring the security, bug-fix, build and CI changes in #136 through #147, and for reviewing, verifying and merging #128.

v0.4.0 (unsupported)

Choose a tag to compare

@github-actions github-actions released this 23 Oct 12:43

⚠️ Unsupported release. Upgrade to the latest release: https://github.com/BaaRai/ARTEX/releases/latest

First release of this fork. The engine underneath is still upstream's, but this tree carries a long stretch of operational work on top of it.

Highlights

  • Recon toolbox baked into the image: nuclei, httpx, subfinder, katana, ffuf and dalfox at release-pinned versions, plus a sqlmap checkout that tracks the default branch. Each one is smoke tested at build time by running it once; none is checksum verified in this release, and #127 adds checksum verification.
  • Recursive recon skill: bbot v3.0.2 with a policy gate. Passive, zero-touch scans when a program bans scanners; active modules only when it allows them.
  • php2shell skill: upgrade a file-inclusion finding to verified RCE, lab first and gated on program policy.
  • Goal ambiguity gate at task creation, plus LLM chain failure escalation and per-intent profile rotation.
  • Stagnation warning in the graph overview when a worker repeats itself without producing findings.
  • Per-task credentials vault, cost ledger with per-call token accounting, backup script, MCP health reporting.
  • Findings discipline: proof-basis gates, measured impact, adversarial verification and dedup rules.

Changed

  • Norma upgraded to v0.4.4: per-task process-tree memory watchdog, UTF-8 aligned truncation, non-interactive subprocess environments, message_delta usage recovery, streaming fixes.
  • Pause preserves the task timeout budget; paused time is frozen in the task clock.
  • Prompt-injection hardening on seed material and MCP tool descriptions.

Removed

  • In-app self-update and the automatic version check. The update source was hard-coded to a repository that no longer exists and its checksums came from the same release. Upgrade with update.sh or a release archive. Thanks to @stevenAthompson for this one.

Full details in CHANGELOG.md.

Images are not published to a registry from this repository yet. Build locally with the Dockerfile, or run the archive for your platform.