Keep tools current and serve Xcode MCP over HTTP - #34
Open
MrSaneApps wants to merge 40 commits into
Open
MrSaneApps wants to merge 40 commits into
MrSaneApps wants to merge 40 commits into
Conversation
Weekly keep-current applies npm pins and in-major Firecrawl bumps. Mini recurring jobs run on LaunchAgent plus Grok, with Codex heartbeats paused. Xcode MCP uses the Mini HTTP singleton through the existing tunnel instead of a fresh SSH stdio spawn. Machine cleanup plans generated junk by kind, not free space.
Bring in locked Mini screenshot evidence and the CWS watcher recovery from main. Keep the 2026-08-21 Grok/Cursor keep-current work.
Grok was importing Claude hooks that matched tool_name=Bash and no-opped on GROK_HOOK_EVENT, so shell guards never saw run_terminal_command. Shared guards now parse Claude, Grok, and Cursor payloads. Grok loads native ~/.grok/hooks and turns off Claude/Cursor hook import. Cursor keeps its adapters. Claude and Codex registrations stay in place.
Owner-approved SANEBAR_RELEASE_POLICY_ONLY ships can reuse a Mini release_preflight receipt without customer-UI, verify, upgrade-path, or fingerprint binds. Forward the env to Mini so candidate preflight stays in the same mode.
Stop a T&Z Mini pending marker from injecting stop follow-ups into unrelated chats by storing state under conversation_id, ignoring the legacy global file, and treating detector self-tests plus AX/simctl reads as non-mutations. Co-authored-by: Cursor <cursoragent@cursor.com>
Require livez plus orphan :3111 reclaim in the Mini supervisor, schedule an Air memory-only watch with notify-on-red, treat AgentMemory API polls as GUI feedback, and surface dirty/unpushed finish-line apps in operator_brief. Co-authored-by: Cursor <cursoragent@cursor.com>
Compare 5-minute load to core count on Air and Mini, ignore expected build and app work, and notify only on Air after two consecutive hits.
Unblock hosted-file post-release checks for Clip (product 779223) and allow an explicit lemon_variant_id pin in products.yml. Co-authored-by: Cursor <cursoragent@cursor.com>
Block ad-hoc Cloudflare and NVIDIA chat calls unless llm_bakeoff, a fresh research receipt, or an explicit owner override is present—stops winging vendor APIs that kept failing on wrong call shape. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Document CF+NVIDIA burn on Mini LaunchAgent and allowlist overnight_quota/draft_claim in the LLM API guard. Co-authored-by: Cursor <cursoragent@cursor.com>
Drop KeepAlive wording so recurring-jobs matches translations AI_CROSSCHECK and the Mini LaunchAgent install script. Co-authored-by: Cursor <cursoragent@cursor.com>
Peekaboo v4 removed the list/image subcommands, which broke the visual workspace guard probe and the smoke capture commands at runtime even though the old tests still passed. Move all call sites to the 4.3.3 syntax (window list, app list, menubar list, see --no-elements) and update the tests and docs to match. Also included in this unit: mini-local routing plus reduced-motion and canonical lefthook pinning for web captures, deterministic session identity for locked evidence captures, and the 5s Tailscale ping timeout the proxy already uses.
Three guards required scripts/hooks/core/hook_payload, which was never added, so every guard in the chain crashed with LoadError and the dispatcher failed closed-open (raw captures and layout violations passed). Add the shared payload parser with cross-client shell/edit classification plus its unit tests. Also in this unit: shared local-UI guard (Air app-edit, pasteboard, and local-GUI blocks with Mini-remote exemptions), no-prompt Keychain policy, persistence-debt handoff tracking, subagent brief-completeness gate, TCC language corrections, and UTF-8 stdin handling.
Session start/end, cleanup routes, and MCP checks point at Grok and SaneMaster instead of retired Codex paths. Standing maintenance authorization recorded. Development and architecture notes cover mini-local capture, reduced-motion receipts, memory-guard skip rules, keep-current pins, and retired duplicate jobs. Handoff ledger current with no secret material.
Deploy retires duplicate disk-clean jobs and agents. Memory guard skips runs during active builds or open coding clients. Nightly adds a bounded keep-current section. AgentMemory supervision requires livez health and reclaims orphan port listeners by exact PID. Fixtures cover the livez requirement without network. Test registry covers the new suites.
Validation reads live projects from meta/PROJECT_MAP instead of a hardcoded list, covers websites, and stops hardcoding /Users/sj. Release SOP gains the Lemon Squeezy hosted-file handoff with official source citations, matching the staged-upload lane.
portal_targets silently resolved app id 1847 to the release-peer checkout instead of canonical SaneClip (companion one-liner in the peer repo disables setapp there). Abort loudly on duplicate app ids, pin Xcode SDKROOT for the clang residue fixture, and add setapp_config_test.rb (registered in test_registry.json). Suites: upload 39/39, config 2/2, verify_guard 59/59.
Air-verified 2026-09-17: com.saneapps.mini-screenshot agent queue request/receipt recipe (exit 0, 1920x1080 PNG); wrapper Screen Recording fallback guidance.
Secrets model corrected (Keychain-first), ARCHITECTURE gate/count, RELEASE_SOP fallback+numbering, BOOTSTRAP ZIP revision, registry precedence, hook README behavior rows, handoff trimmed to active cards.
Wires sane_push_guard.rb into the bash guard chain and registers its suite. Fails closed on non-fast-forward plain pushes; fails open for force-family, detached HEAD, and unreachable remotes so git still reports the real error. Prevents the push-reject-fetch-rebase loop.
…ging Drops automatic-release-after-approval enforcement so versions stay manual until the owner approves. Adds appstore.iap.price_usd config lookup with 14.99 default and UTF-8-safe env parsing. Lemon Squeezy staging now verifies SHA-256 and retains earlier archives pending remote proof instead of deleting them. Release script requires an affirmative published-file snapshot.
…ardening Adds SaneRuntimeLog supervisor (owns only its log child, bounded capture with receipt) and routes test_mode/sane_test through it with --quiet-logs/--log-seconds. Removes TCC database manipulation and screensaver-default mutation from the work-session guard; bounded caffeinate with PID ownership checks instead. Verify checks the OS-reported executable for test processes and retains per-phase bundles. Machine cleanup plans by artifact kind with a server-busy refuse path. Customer UI contract revokes legacy executor receipts and rejects declaration-only artifacts.
…curring jobs Dependency baseline becomes keep-current: version pin bumps, safe same-major Firecrawl auto-bump with receipts, no-prompt keychain policy, and an Air-only weekly LaunchAgent. Session guardian gains a sustained-CPU watch with expected-work allowlist plus --install. Control-plane sync (Codex/Grok/Cursor) never overwrites peer-owned files and fails closed on conflicts or transfer errors. Adds LaunchAgent recurring-job runners (review watch, X scout, email campaigns, agent heartbeats) with single-instance locks, human- readable review-watch emails, dimension-aware funnel reporting, and hardened SEO/social/X-scout audits.
- Forward 37917 (serena) alongside existing MCP forwards - ssh -n -S none so tunnel restarts cannot silently die as mux slaves - ConnectTimeout 3->15 for slow Mini wakeups
Firecrawl CLI belongs on Mini as well as Air now that Grok TUI runs here. AgentMemory MCP talks to the Mini worker on loopback instead of hanging on Cloudflare Access at session start. Cloud login stays --login only. Peekaboo 4.4.0 is a Mini binary install (GitHub tarball); brew still cannot keg-upgrade without Xcode 27.
Declaration-only artifacts cannot prove runtime execution; revokes legacy executor receipts.
Pairs with 5d67fd0 (tunnel ConnectTimeout 15); rest of this file merges separately.
…de-mcp # Conflicts: # AGENTS.md # SESSION_HANDOFF.md # scripts/automation/agent_heartbeat_test.rb # scripts/automation/dependency_baseline.rb # scripts/automation/dependency_baseline_test.rb # scripts/automation/heartbeats/saneapps-launch-ops.md # scripts/automation/install-recurring-agents.sh # scripts/automation/recurring-jobs.md # scripts/automation/sync-control-plane.sh # scripts/automation/sync-cursor-mini.sh # scripts/automation/sync-grok-mini.sh # scripts/grok-bin/README.md # scripts/grok-bin/xcode-mcp-frame.py # scripts/hooks/core/hook_payload.rb # scripts/hooks/sane_bash_guards.rb # scripts/hooks/sane_layout_guard.rb # scripts/mcp_singleton_bridge.cjs # scripts/mini/README.md # scripts/mini/SCREENSHOT_TOOLS.md # scripts/sanemaster/command_registry.rb # templates/RELEASE_SOP.md
Keep Air behavior (approval-prefix guards, until-direct assert, annotate fixture, registry rewrite) and Mini additions (payload/brief-linter registry entries).
Grok TUI runs on Mini as well as Air; shared pin stays 1.23.3.
Orwell compression, no policy deleted. Frees delegation-startup budget (home+codex+project must fit Muse 64KiB cap).
Q10 check fails when ~/AGENTS.md + ~/.codex/AGENTS.md exceed 62KiB (warn 60KiB); cap math: 65536 incl ~2KiB framing.
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
1 participant
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.
Summary
keep_currentapplies npm pins and in-major Firecrawl bumps; Mini nightly does the Mini role.http://127.0.0.1:37915/mcp, forwarded through the existing AgentMemory tunnel, instead of a fresh SSH stdio spawn that timed out at Grok session start.Test plan
ruby scripts/mcp_contract_test.rb11/11ruby scripts/automation/memory_sync_test.rb10/10ruby scripts/automation/air_mini_acceptance_test.rb10/10ruby scripts/automation/dependency_baseline_test.rb33/33ruby scripts/automation/agent_heartbeat_test.rb13/13ruby scripts/sanemaster/machine_cleanup_test.rb23/23ruby scripts/sanemaster/machine_cleanup_retention_test.rb7/7GET /healthzon 37915 ok; Air initialize through the tunnel HTTP 200