{"full_name":"BaaRai/ARTEX","owner":"BaaRai","name":"ARTEX","shard":"1","url":"https://github.com/BaaRai/ARTEX","description":"ARTEX - AI autonomous penetration testing system (English translation; upstream: Autumn-27/ARTEX, AGPL-3.0)","homepage":null,"topics":[],"tag_pages":[],"language":"Go","license":"AGPL-3.0","avatar":"https://avatars.githubusercontent.com/u/130491490?u=2712c6194e5e7544d230db09e5d10e999704aa58&v=4","created_at":"2026-10-08T21:14:39Z","pushed_at":"2026-11-16T08:33:37Z","metrics":{"stars":4203,"forks":1881,"watchers":17,"issues":67,"prs":178,"releases":4,"issues_open":6,"issues_closed":61,"prs_open":1,"prs_closed":7,"prs_merged":170,"discussions":0},"growth":{},"history":{"dates":["2026-11-16"],"stars":[4203],"forks":[1881],"watchers":[17],"issues_open":[6],"prs_open":[1]},"boards":{"new":169,"debut":86},"same_owner":[],"issues":[{"number":154,"title":"[Feature Request] Support custom User-Agent and fixed full API endpoint URL","comments":5,"created_at":"2026-11-08T05:15:58Z"},{"number":129,"title":"DemonRoute as a provider option via ARTEX_LLM_BASE_URL","comments":4,"created_at":"2026-10-31T11:56:20Z"},{"number":242,"title":"UI concept: simulated operator console for autonomous pentest runs","comments":1,"created_at":"2026-11-15T16:34:31Z"},{"number":89,"title":"Benchmark: diversify public ground truth beyond Juice Shop (XBOW set is saturated; Cybench/CyberGym candidates)","comments":1,"created_at":"2026-10-12T19:48:55Z"},{"number":124,"title":"Adversarial audit: map the 2026 agentic attack classes (memory poisoning, plan hijacking, tool abuse) onto ARTEX itself","comments":0,"created_at":"2026-10-22T20:36:30Z"},{"number":5,"title":"OpenClaw-style chat bot: drive ARTEX from Telegram (inbound bridge + remote approvals)","comments":0,"created_at":"2026-10-09T12:34:51Z"}],"discussions":[],"latest_release":{"name":"v0.6.0","tag":"v0.6.0","published_at":"2026-11-11T23:19:53Z","body":"## Summary\n\nARTEX 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.\n\nFull list: [CHANGELOG.md](https://github.com/BaaRai/ARTEX/blob/v0.6.0/CHANGELOG.md) · [compare v0.5.1...v0.6.0](https://github.com/BaaRai/ARTEX/compare/v0.5.1...v0.6.0)\n\n## Highlights\n\n- **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).\n- **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).\n- **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).\n- **Intercept audit**: approvals record `decided_by`, and the `[model]`/`[Model]` prefix bug no longer stores judge decisions as `unknown` (#159, #165).\n- **Notifications**: configurable summary and asset limits, plus Feishu digest overflow handling (#170 to #174).\n- **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).\n- **Intent claim leases**: a heartbeat, a TTL and an owner token stop a stale worker from overwriting a result (#176).\n- **Smaller fixes**: custom User-Agent per LLM profile, the custom-tool proxy scheme fix, and the `list_task_credentials` schema fix (#155, #187, #130).\n\n## Upgrade notes\n\n- **No manual migration**: schema changes are idempotent at startup. Existing findings stay `unverified`, and the new optional fields default to empty.\n- **The target gate is off by default**. With it off, behavior is unchanged. With it on, an uncovered target waits for an approval.\n- **`goal_met` refuses while a goal is open**, and `prove_goal` accepts a fact only with `confidence=observed`.\n- **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.\n- **Findings gain `counterevidence` and `confidence_rationale`**, optional and never blocking. Existing databases receive the schema property at startup.\n- **0.5.x is no longer supported.** Fixes target the latest release line only; upgrade to 0.6.0.\n\n## Known limits\n\n- 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).\n- 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.\n- Fact provenance arrives in 0.7.0. Until then, `prove_goal` trusts the `confidence` the model declared.\n- The new `add_hint` and `prove_goal` tool descriptions reach fresh databases only; the server-side refusals apply everywhere.\n- 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.\n- 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.\n\n## Contributors\n\n- @kiserufetch: thank you for the `list_task_credentials` schema fix (#130).\n- ","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}