Skip to content

chore: stage ARTEX upstream hardening and operational readiness gate - #1

Draft
thebrierfox wants to merge 53 commits into
mainfrom
maintenance/upstream-2026-11-06-readiness
Draft

thebrierfox wants to merge 53 commits into
mainfrom
maintenance/upstream-2026-11-06-readiness

Conversation

@thebrierfox

Copy link
Copy Markdown
Owner

Scope

  • Stage the 52-commit, ancestry-verified fast-forward from BaaRai/ARTEX at a908e1ec170f973734b3258dd37d667d2562440a (includes v0.5.1) onto the fork's original main at b3717f7a5fe8d07fcd4fe7167603f314e9b64840.
  • Add docs/OPERATIONAL_READINESS.md, distinguishing source synchronization from a verified operational deployment.
  • No T3MP3ST source code or other external tool was imported. Its provenance, authorization, benchmark, and release validation practices are references for follow-up tests only.

Why

The previous docker-compose.yml referenced discontinued autumn27/artex and bound 8787 to every interface; a fresh Dockerfile build also needed pre-existing dist/ artifacts. Upstream transitions to locally built images, loopback defaults, and improved auth/secret handling. Recent ARTEX ecosystem trojanized-image reports make source/build provenance a priority.

Must verify before merge or deploy

Status

DRAFT — NOT CERTIFIED. Keep main and any deployment unchanged until evidence is recorded. Passing unit tests is necessary but does not establish operational readiness.

Sources: BaaRai/ARTEX and elder-plinius/T3MP3ST.

glauberportella and others added 30 commits October 23, 2026 18:12
Two review fixes on top of the Ollama provider:

FromEnv no longer inherits OPENAI_API_KEY for the ollama format. Ollama
ignores the bearer token, so inheriting the cloud key could only ship it to
the local or LAN endpoint, contradicting the comment in NewProvider and the
documented behavior.

ConfigFrom now normalizes the base URL to the /v1 root, so the natural
Ollama URL (host:11434 without /v1) works for chat as well. Previously the
provider appended /chat/completions to it and the request landed on a path
Ollama does not serve, while pgListModels already appended /v1.

Tests: bare-host normalization assertion, the /v1/chat/completions path
assertion, and a new test that the env config carries no inherited key.
The Streamable HTTP client always offered 2025-06-18 and kept sending
that MCP-Protocol-Version header after initialize, ignoring the
version the server answered with. A server that only supports
2025-11-25 negotiates its own version and then rejects every later
request with 400, so none of its tools load. Servers that validate
the header on initialize itself reject the handshake outright, and
the client never retried with another version.

- Offer 2025-11-25 first and fall back to 2025-06-18 when the server
  rejects the version: HTTP 400 "Unsupported protocol version",
  JSON-RPC -32602, or an UNSUPPORTED_PROTOCOL_VERSION code
- Send the protocolVersion from the initialize result on later
  requests
- Keep the legacy HTTP+SSE transport on 2024-11-05

Closes BaaRai#131

Tags: #mcp
Co-Authored-By: htjulia <htjulia1@gmail.com>
The toolbox entry claimed the tools were verified at build time. They are
version-pinned and smoke-tested by running them once; nothing is checksum
verified, sqlmap tracks the default branch, and the dalfox check cannot fail.
Say that plainly and point at BaaRai#127, which adds checksum verification.

Also fix two misattributions: BaaRai#65 is the deadline contract, not pause, and
BaaRai#108 adds documentation rather than reporting code.
docker-compose.yml referenced autumn27/artex:latest. That repository is gone
(404), but the Docker Hub account is alive, so whoever holds that name can
publish the tag and every compose pull would run their image. A local review
of the published artifacts found the same class of risk still present on the
default install path after BaaRai#125 removed the in-app updater.

Build from the checkout instead, and make update.sh stop when git pull cannot
fast-forward: it used to warn, rebuild the old tree and print 'Build complete',
which is what every clone made before the 2026-10-23 history rewrite now hits.
The saved-profile path normalized /v1 but FromEnv did not, so
ARTEX_LLM_BASE_URL=http://host:11434 sent chat to /chat/completions and got a
404 on a LAN Ollama server. Share one helper between both entry points, and add
an env-path test with a bare host and with a full endpoint URL.

Also soften the keyless comment: a local Ollama ignores the bearer token, but a
gateway in front of it can require one, which belongs in a saved profile.
The client sent MCP-Protocol-Version on the initialize request itself.
The spec requires the header on requests after initialization, and the
offered version already travels in the initialize body. A server that
validates the header rejected the handshake unless the client happened
to offer a version it supports, so connecting depended on matching the
error body and only worked for the two versions the client offers.

- Send the header only after initialize succeeds, with the version
  the server negotiated
- Keep the fallback loop, newest-first order and negotiated-version
  persistence for servers that still reject the initialize body version
- Rewrite both version tests around this, including a 2025-03-26-only
  server, and assert that initialize carries no version header

Tags: #mcp
Co-Authored-By: htjulia <htjulia1@gmail.com>
…ld processes spawn

ARTEX_PG_DSN, ANTHROPIC_API_KEY and OPENAI_API_KEY are read only during
startup (server.NewManager and agent.FromEnv in server.New), but stayed in
the process environment. Agent shells, custom tools and MCP stdio servers
copy os.Environ() when they start, so they could read the database password
and the LLM keys. main now unsets the three variables right after server.New
returns.
The scrub ran in main after server.New returned, but New resumes every
non-terminal task before returning, and those loops start workers that run
shells. Move the scrub into New, right after both environment readers, and
start MCP discovery after it as well: both paths spawn child processes.

New takes an option rather than scrubbing unconditionally. The test suite
builds several servers in one process and still needs ARTEX_PG_DSN on later
calls, so only the binary opts in.
fix(security): remove startup secrets from the environment before child processes spawn
…localhost

POST /api/auth/init accepted any caller while auth.password_hash was empty,
so whoever reached port 8787 first could set the admin password and take the
instance. When no password is set at startup, the server now generates a
random one-time setup token and writes it to the server log. authInit
compares it in constant time, rejects a missing or wrong token with 403, and
clears it once the password is stored.

The setup page has a field for the token and api.initPassword sends it.
docker-compose publishes 8787 on 127.0.0.1 only, like 8788. README says where
to find the token.
Publishing 127.0.0.1:8787 removed the only working access path for installs
reached through a host or WSL IP, without an obvious way back. Keep loopback as
the default and let ARTEX_BIND_ADDR override it (0.0.0.0 restores the old
behaviour; the setup token still protects the first-run claim).
fix(auth): require a setup token for /api/auth/init and bind 8787 to localhost
…nored

Commit 2c8359f appended backups/ to the /Dockerfile.bench.v2 line, which had no
trailing newline. The result, /Dockerfile.bench.v2backups/, matched neither file,
so dumps written by scripts/backup.sh showed up as untracked and could be
committed by accident.
With --keep 0, prune ran tail -n +1 over the newest-first listing and deleted
every dump, including the one written seconds earlier, then reported success.
The check also rejects leading zeros, which the $((KEEP + 1)) arithmetic would
parse as octal (08 and 09 abort the script after the dump).
FromEnv compared the raw value, so "Ollama" or "OpenAI" matched no case and
fell through to the anthropic default. With ARTEX_LLM_PROVIDER=Ollama and an
ANTHROPIC_API_KEY set, the Anthropic key was sent to the Ollama base URL.
…lues

config.Load discarded the json.Unmarshal error, so a config.json that did not
parse was treated as absent: startup reported "no database config found"
and skill_dir silently fell back to the default. Load now returns read and
parse errors (a missing file is still not an error); PostgresDSN and SkillDir
pass them up, and main exits with the message.

install.sh wrote the database values into config.json unescaped, so a password
containing " or \ produced invalid JSON, which the old Load then ignored. The
values are now JSON-escaped and the port must be numeric. The .env sed
replacements escape \, | and & so those characters no longer break the
substitution or get rewritten by it.
POST /api/llm/test and POST /api/llm/models take provider and base_url from
the form but fell back to the stored profile key (and, for the test, the
global key) when the form had none. Any base_url typed into the form, or
supplied by a crafted request, received the saved key.

The saved key is now used only when the request's provider and base URL
normalize to the ones it was saved with (sameLLMEndpoint, built on
agent.ConfigFrom). Otherwise the key must be entered in the form, and the
error says why.
The image installed Node 20 from NodeSource, which is end-of-life. CI and the
release workflow already build the frontend with Node 22; the runtime image
(Playwright MCP/CLI, npx-based MCP servers) now uses the same major.
go test -race runs on every package that passes under the race detector
locally (all but ./server, which needs a -race check against postgres
first). A new shell job runs shellcheck on install.sh, update.sh, start.sh and
scripts/backup.sh.
Raises the go directive from 1.26.3 to 1.26.10, the newest 1.26 patch release
(CI installs it from go.mod; release.yml's "1.26" already resolves to it; the
Dockerfile ships a prebuilt binary and has no Go image). golang.org/x/net goes
from v0.58.0 to v0.62.0, the latest; go mod tidy raises x/crypto, x/text,
x/sys, x/sync, x/mod and x/tools to the versions it requires.
The first run reported only info-level notes (SC2015 in update.sh, SC2012 and
SC2086 on the intentional glob in backup.sh), which made the job fail.
New accepted opts ...Option but never called them, so scrubSecrets stayed
false in the binary and ScrubSecretEnv was unreachable: the protection merged
in BaaRai#136 never engaged. The unit test passed because it called the scrub
function directly instead of going through New.

Apply the options in the constructor and add a test that builds a Server
through New with WithSecretScrub and asserts the secrets are gone. Verified to
fail without the loop.
fix(gitignore): ignore backups/ again
fix(agent): match ARTEX_LLM_PROVIDER case-insensitively
fix(docker): install Node 22 in the runtime image
BaaRai and others added 23 commits November 3, 2026 23:49
build: go 1.26.10 and golang.org/x/net v0.62.0
fix(config): fail on malformed config.json; escape values in install.sh
fix(llm): send a saved API key only to the endpoint it was saved for
fix(security): apply New options so the secret scrub actually runs
The -race step reused the job's ARTEX_PG_DSN, so it ran against the database the plain test step had just populated. The db tests are not idempotent across runs and failed on duplicate llm_profiles names, turning main red right after BaaRai#144 merged.
…Rai#147)

The Dockerfile copied a pre-built binary from dist/<arch>/artex into the
runtime image, but dist/ is git-ignored, so a fresh `git clone` has no such
file and `docker compose up -d --build` fails with
"/dist/amd64/artex: not found".

Build the binary from source in the image instead: a Node 22 stage runs the
frontend static export, a Go 1.26 stage runs build.sh with the same flags as
the release binaries job (CGO-free cross-compile, -trimpath, pinned version),
and the runtime stage copies the compiled binary. No build-context binary is
needed, so a clean clone builds.

Drop the now-obsolete dist/ warning in install.sh.
Move the Unreleased entries into a 0.5.0 section and add the changes merged since v0.4.0 (BaaRai#128, BaaRai#136-BaaRai#147), with known issues.
The server wrote jwt.key to the executable's directory, /app in the
image, so it lived in the container's writable layer and every container
rebuild generated a new key and signed all users out.

- ARTEX_JWT_KEY_DIR selects the key directory (default: config.BaseDir(),
  unchanged for non-Docker installs). docker-compose.yml sets it to
  /app/keys and bind-mounts ./keys there; the Dockerfile declares the
  volume. The key stays out of data/, which the file manager browses.
- loadOrCreateJWTKey creates the directory (0700). When keyDir/jwt.key is
  missing it migrates dataDir/jwt.key, then BaseDir()/jwt.key (mode 0600,
  old file removed), otherwise generates a new key. An existing key is
  never overwritten; a short one is now a startup error.
- .gitignore and .dockerignore exclude keys/.
- README and CHANGELOG updated; the 0.5.0 known issue is removed.
fix(auth): keep the JWT signing key in a dedicated volume
Added a Trendshift badge to the README.
…default

- POST /api/auth/login: 10 attempts per minute per TCP peer IP
  (X-Forwarded-For ignored), then 429 with Retry-After. The block starts
  at 1 minute and doubles on each attempt while blocked and on each new
  block, capped at 15 minutes. A successful login clears the IP. No
  per-account lockout. In-memory, idle entries swept after 15 minutes,
  at most 10,000 IPs.
- New passwords (init, change-password, /setup page) need 12 characters.
  Login does not check length, so existing 8-character passwords work.
- The binary's -addr default is 127.0.0.1:8787. The Docker image passes
  -addr :8787 explicitly and is unchanged.
- GET /api/health returns {"ok":true} without a valid token; with one it
  still returns service and version for the web UI top bar.
… reset-password.sh

- ARTEX_TRUSTED_PROXIES (comma-separated CIDRs, empty by default). When
  the TCP peer is listed, the login limit reads X-Forwarded-For from the
  right and keys on the first hop that is not a trusted proxy; otherwise
  the header is ignored. Without it every client behind a reverse proxy
  shares one bucket and an attacker can block the admin. Passed through
  by docker-compose.yml and listed in .env.example.
- README: nginx example sends X-Forwarded-For; document the variable;
  the binary's default listen address is 127.0.0.1:8787.
- reset-password.sh refuses passwords under 12 characters, counted in
  characters like the server.
- README: ARTEX_TRUSTED_PROXIES for nginx and ARTEX on the host
  (127.0.0.1) and for nginx on the host with ARTEX in Docker (the bridge
  gateway, found with docker network inspect).
- CI shell job: run shellcheck on reset-password.sh too.
fix(auth): rate-limit login per IP, trusted proxies, 12-character passwords
…lback

fix(mcp): negotiate protocol version with fallback
…E step

The build stage of the Dockerfile runs `COPY . .`, so anything left in the
checkout root is sent to the builder and copied into the build stage. Exclude
the local secrets and backups that the install and backup scripts create
there: .env, config.json, jwt.key and backups/. keys/ was already excluded.

The Docker Compose (manual) section of the README still told users to build
the single binary before `docker compose up -d --build`. Since the
multi-stage Dockerfile builds the frontend and the Go binary itself, that
step is no longer needed; remove it.
chore: keep secrets out of the Docker build context; drop stale README step
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants