Release/1.11.13 - #79
Merged
Merged
Conversation
… and source filtering Signal channel now has full parity with Telegram: Source filtering (signal.ts): - Private mode: only Note to Self messages pass through - Group mode: only Mercury group messages pass through - All other chats/DMs silently dropped before any processing - /pair only works in group mode after source filter - Private mode auto-pairs account owner as admin on first message Routing fixes (signal.ts): - Removed channelId sanitization — phone numbers and group IDs are no longer mangled, fixing all outgoing message routing - sendFile() group guard aligned with sendRaw() - resolveTarget() now correctly routes in both modes Permissions (index.ts): - Signal wired into onAsk permission handler - Signal wired into setOnPermissionMode callback - Signal wired into setChatCommandContext - Signal wired into setSendFileHandler with explicit channel param Task lifecycle (agent.ts): - Signal wired into beginTask/resetStepCounter at task start - Signal wired into sendToolFeedback/sendStepDone during tool execution - Signal wired into endTask/resetStepCounter on error - Signal wired into sendCompletion/popDeferredResponse on completion - Signal included in canStream for canned text streaming - SignalChannel imported alongside TelegramChannel send_file channel parameter (send-file.ts, registry.ts): - Added optional 'channel' param (signal|telegram) - When user says 'send via Signal', routes to Signal explicitly Unregister fixes (setup.ts, index.ts): - checkLocalAccountData() now checks registered field — treats unregistered accounts as not linked - deleteSignalCliAccountData() removes local account data after unregister so next setup triggers fresh device link - unregisterSignalNumber() uses findSignalCli() not ensureSignalCli() to avoid downloading signal-cli just to unregister - Both unregister flows (CLI and configure) call deleteSignalCliAccountData() - 'Keep config → No' no longer deletes signal-cli account data (device is still linked on server, only Mercury config is cleared) Setup flow fixes (index.ts): - 'Keep config → No' now re-asks mode (group/private) - Private mode auto-pairs account owner as admin during setup - Removed link URI display (QR code only) - startLinking timeout increased from 15s to 60s - startLinking now checks stderr for sgnl:// URI - sendSignalMessage uses findSignalCli() not ensureSignalCli()
…s, lifecycle hooks, and CLI commands
…ork with minimal caching
…ng, DM gating, and CLI commands
Channels should be open — anyone can talk to Mercury in a channel where it's been installed. Access requests (admin approval) only apply to DMs. Also adds token prefix hints (xoxb-/xapp-) and App Home step to setup.
- message event handler now only processes DMs - app_mention handler processes channel @mentions (with rate limiting) - Removed handleChannelMessage method (no longer needed)
… DM approval, no pairing codes - First user to DM /mercury start auto-becomes admin (no CLI pairing needed) - Subsequent users get 3-button approval DM to admins: Approve as Admin, Approve as Member, Reject - Removed completeInitialSlackPairing() from CLI (no longer needed) - Removed pairing code generation and approveSlackPendingRequestByPairingCode usage - Updated slack approve CLI command to only accept userId (no pairing code) - Cleaned up formatSlackUser to remove pairing code display
hotheadhacker
force-pushed
the
release/1.11.13
branch
from
June 18, 2026 07:24
23c95f0 to
8f80b29
Compare
…or android/x86_64 Rollup has no native binary for android/x86_64, causing tsup/vitest to crash with throwUnsupportedError. After npm ci, install @rollup/wasm-node and symlink it over node_modules/rollup so the WASM build is used instead. Also adds PATH and HOME exports needed for the Termux container environment.
- chown entire /home/builder (not just project dir) - mkdir+chown /home/builder/.npm before su - use 'su system' (no login flag) to preserve env exports - set npm_config_cache inside project dir (owned by system) - replace rollup with @rollup/wasm-node symlink (no native binary for android/x86_64)
- Web assets (UI + static) now resolve relative to the binary when running as a Bun --compile standalone, fixing the dashboard for binary users who previously got 'Run: cd ui && npm run build' - server.ts, auth.ts, brain-fallback.ts detect in __dirname and fall back to process.execPath-based paths - build-bin.cjs copies dist/web/ alongside each binary and creates web.tar.gz for GitHub release uploads; checksums include web assets - install.sh and install.ps1 download and extract web.tar.gz - mercury upgrade detects standalone binary via isStandaloneBinary() and re-runs the appropriate installer script instead of npm - Termux CI: fix ownership, cache, and rollup wasm fallback for android/x86_64
ash (Termux /bin/sh) reported 'ci: syntax error: unmatched {' for
the previous npm ci || { ... } block. Replace with plain
'npm ci || npm install' which ash accepts while keeping the same
fallback behavior when better-sqlite3 native build fails.
…r optional native deps npm ci aborts the entire install when better-sqlite3's gyp build fails, which is fine, but the previous || npm install fallback regenerated the lockfile and node_modules lost the tsup .bin symlink, causing 'tsup: not found' on the next step. Switch to plain 'npm install' for Termux — it tolerates optional native dep build failures (better-sqlite3) while keeping all devDeps installed. python + build-essential are still installed for any package that does need a native build.
…ectly npm 11's new allow-scripts policy blocks esbuild's postinstall, which prevents the esbuild platform binary from being downloaded. As a side effect, tsup's .bin symlink resolution can fail in some Termux/ash setups, causing 'sh: 1: tsup: not found' even though tsup itself is installed. Workaround: set npm_config_allow_scripts=true so esbuild's binary is fetched, and invoke tsup directly via 'node node_modules/tsup/dist/cli-default.js' to skip the .bin symlink lookup entirely. Then run the post-build script and tests as before.
… 11) npm 11 added a project-scoped install-script policy that blocks all postinstall scripts by default. The --allow-scripts flag and the npm_config_allow_scripts env var are both rejected in project installs with EALLOWSCRIPTS; the only allowed path is the allowScripts field in package.json (or .npmrc). Add 'allowScripts.esbuild: true' so the esbuild platform binary is downloaded during install. Drop the env var that caused the previous run to fail. This also fixes the tsup bin-link issue in Termux, because the postinstall chain that esbuild triggers is what makes tsup's .bin symlink resolve correctly.
Stop guessing. Add: - set -ex for full command tracing - node/npm/uname/pwd versions - ls -la of node_modules/.bin/tsup and node_modules/tsup/dist/ - file/readlink on the symlink - head -1 of cli-default.js to confirm the file is a real script - foreground-scripts on npm install so gyp output is visible - tail -200 to keep log size sane This will tell us, on the next run, exactly: 1. Which npm version is in the Termux container 2. Whether tsup is actually installed and linked 3. Whether the .bin symlink is broken or absent 4. Whether the better-sqlite3 gyp build is silently failing Once we have this data, we can replace this debug step with a real fix.
The previous 'su system -c \' ... \'' nested quoting was being mangled by bash — su received 'set' as the user and the script body as positional args, producing 'su: Unknown option -' followed by the script content. Fix: write the Termux CI commands to a file via heredoc on the host, mount it into the container read-only, copy to /home/builder, chown to system, and 'su system /home/builder/termux-ci.sh' — no inner quoting required. This is the same pattern the working d4028ba used.
The previous order printed node --version before pkg install nodejs-lts, so the script failed with 'node: not found' on the very first check — we never got any actual debug data. Reorder: env, then pkg install, then versions, then install, then inspection. Also add 'which node' / 'which npm' so we can see exactly where they live.
…uilt
The previous failures were all chasing the wrong target. The real
data from the diagnostic run proved:
- node 24.16.0 / npm 11.17.0 install fine in Termux
- node_modules/.bin/tsup symlink is present and correct
- 'node node_modules/tsup/dist/cli-default.js' runs and tsup
reports 'ESM Build success in 2882ms'
- The actual failure is scripts/post-build.cjs, which runs
'npm install' in ui/ and then 'npx vite build'. Vite pulls in
esbuild, and esbuild's postinstall tries to download a platform
binary for android|x64 — which esbuild does not ship. The
install.js script later tries 'spawnSync ui/node_modules/esbuild/
bin/esbuild' and fails with ENOENT.
The CLI agent does not need the web dashboard to run, and the web
dashboard is already exercised by the linux/macos/windows CI jobs.
Skip the UI install + build when TERMUX_VERSION is set, and skip the
ui/dist copy when the dir doesn't exist.
Clean up the debug noise from the Termux step now that the root
cause is known. Keep the file-based su invocation (works), the
python + build-essential install (needed for better-sqlite3 gyp),
and the @rollup/wasm-node symlink swap (works).
npm run build's 'tsup' command isn't finding node_modules/.bin/tsup in PATH inside the Termux container (ash). The diagnostic run proved the symlink exists and node node_modules/tsup/dist/cli-default.js works. Replace npm run build with the two explicit node invocations.
Some Termux mirrors are unreliable. If pkg update hits a bad mirror, fall back and continue — the install step will still work with cached indices.
The service worker file writes to dist/web/ui/sw.js, but on Termux the entire ui/ dir is skipped (no esbuild prebuilt). Guard the writeFileSync behind an existsSync check.
npm run script invocations and vitest rely on node_modules/.bin being in PATH. In Termux's ash shell this isn't set up automatically. Add it explicitly after cd into the project dir so all bin scripts (tsup, vitest, etc.) are found.
…itest npm test spawns 'sh -c vitest run' but the child shell doesn't inherit the PATH we set. npx resolves the binary directly.
npx spawns 'sh -c vitest' which still can't find the binary. Invoke vitest.mjs directly with node, same as tsup.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.