2f25443a9f1d7d6e0a79119cc034d6a3b8860f9a
5
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
9742884308 |
feat(orchestrator): firewall view with signed settings changes (001 S6)
Requirement 001 S6: the hub gets a Firewall view of the Claude Code permission and sandbox settings. The operator signs each change with the passkey; a host settings writer (systemd path unit) applies only exact, signed changes to allowlisted settings files through O_NOFOLLOW directory and file descriptors, and a timer refreshes the view. Every signed text and every page sink refuses or marks hidden characters: Cc, Cf, Zl, Zp, every space but U+0020 and the Default_Ignorable set, with U+FE0E/U+FE0F allowed only after one of the 359 non-ASCII emoji bases of Unicode 16 (shared table, validated at import, the same pair rule on the page). Removal of a deny, ask or denyRead entry is exempt only as an exact single remove. The permission hook prints the hub's refusal reason, cleaned of hidden characters. Accepted MEDUSA false positives (independent review, operator pick): 38 this round, mainly the Unicode defence tables read as attack patterns, generic MM-SCAN-058 heuristics, WEB-FLASK-004 at view.py:213 (ints from the validated table via tojson), EXF26-TOOL-004 on the non-secret WebAuthn challenge, PRIV-SCAN-060 on the office-action shape check and PQC007 on a picker key. 🎯 Fix: 95 % 💥 Break: 3 % 💸 Cost: – ⏱️ Time: – 🛡️ Security: 100 % 🏆 Score: 92 🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 % Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
d9eb121d12 |
feat(orchestrator): S5 Tasks with choices, Gantt, estimates and burn charts
Requirement 001 phase 6, S5: - Tasks tab: every pending ask, rule request, operator row and choice card shows under "waits on Operator"; a choice request goes out on a fixed channel button, post and call with an agent multi-select, and a click on an option sends a passkey-signed answer the agent checks with operator_message. The unsigned /api/decisions/<id>/choose route is gone, and a decision, choice answer or choice recipient whose name holds "*" is refused, so a signed answer never reaches every agent. - Gantt view: one bar per task from start to end (running to now), one arrow per explicit, branch-base and queue-order edge, hatched forecast bars (always, points only, real only) from the median duration or points times median hours per point. - rasci.py set --after/--drop-after stores task dependencies and refuses unknown tasks and cycles. - Estimation toggle (No estimate default, Story points) is stored in the hub settings; burndown and burnup per whole register, list filter or one agent or line, counting tasks or summing complexity with the unestimated tasks as their own line. - New skill kvb-org-queue-estimate: "Request estimates" (story points mode, passkey-signed board action) or the skill sends exactly one newsletter listing every unestimated task with its RASCI parties; a party name that is no plain name shows as "unnamed party". - Business value per task or in bulk is stored only with a verified passkey assertion over the exact ids and values; WSJF uses it instead of the median. The task list sorts by every decision-50 mode both ways. - e2e: the walk's pending ask lives an hour instead of 60 s, and the CDP and BiDi drivers retry removing the browser profile while the browser exits (ENOTEMPTY). Evidence: make test 709 OK, ruff and markdownlint clean (/tmp/s5-make-test-fix-2.log); e2e 26 OK in Chromium and Firefox (/tmp/s5-e2e-fix-2.log); MEDUSA 50 files, an independent skeptic rejected four holes (unsigned choose route, unsigned estimates request, free-text party names, "*" recipients incl. normalizer bypasses), all fixed and re-verified, the 8 remaining pattern hits accepted, rescan 0 open. 🎯 Fix: 92 % 💥 Break: 4 % 💸 Cost: 100 % ⏱️ Time: – 🛡️ Security: 100 % 🏆 Score: 88 🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 % Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
6f940b4415 |
feat(orchestrator): rename kvb-hub to Agent Orchestrator with the phase 2 surface
Surface (requirement 001 phase 2): navigation Stream (Inbox, Regeln) / Agents / Artifacts with keys 1-3 and hash routes (#graph and #tree redirect), header with title, command palette and settings only, server-side badges (unsnoozed inbox, beats within 900 s, open jobs), a Mayday banner with "aufheben", four stream chips with agent:/art:/seit: search, Filter and time menus, the inbox in main, an agents toolbar (Funk an alle with Newsletter, Broadcast, Pan-Pan, Security, Mayday; Priorität; Tag/Nacht), Post or Call in the agent panel, Ambiente in the settings and read-aloud over the seven classes with Mayday locked. Rename (decision 9): hub/ is orchestrator/, the package kvb_orchestrator, the unit kvb-orchestrator-runner.service, scripts/orchestrator.sh, environment KVB_ORCHESTRATOR_*, make targets orchestrator-*, the MCP server, image and compose project kvb-orchestrator, the guard header X-Kvb-Orchestrator and the message marker [kvb-orchestrator msg <id>]. install runs migrate first: it stops the old container, removes its image and code, moves state, approvals and pinned keys, swaps the MCP registration, the PermissionRequest hook and the runner unit, and changes nothing on a second run (scratch-HOME test with stub docker, claude and systemctl). Saved views fall back to the old localStorage key once. MEDUSA acceptances carried to the renamed paths: 82 rows whose lines are unchanged (reason prefixed "renamed from <old path>, line unchanged"), and 7 rows whose text changed only in the product name, the package name, a temp prefix or a new class argument, each upheld by one kvb-think-criticize skeptic (approval.py:60, messaging.py:53, server.py:56, tests/hub_support.py:64, passkey.py:205, page-agent.js:75, language.py:46). Cost 101 %: the longer name adds 15 tokens of skill text; the operator approved it with the commit. 🎯 Fix: 92 % 💥 Break: 4 % 💸 Cost: 101 % ⏱️ Time: – 🛡️ Security: 100 % 💰 Value: ? 🏆 Score: 87 🚦 Gate: ⚠️ cost 101 % > 100: ask the operator before the commit Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
afd96fcd72 |
fix(hub): close seven bugs found in the refactor review
The PermissionRequest hook posted asks cut to the hub's field limits, so a signature covered less than Claude Code would run; it now posts nothing it cannot sign in full and leaves the decision to Claude Code with a reason on stderr. fields.seconds names the field the caller used in its error. Serving a stored page maps an OSError to 500 instead of a dropped connection. The natural-language line marks the register as seen before it builds the inbox, like the page does. do_POST runs every header guard (host, origin, client, content type, MCP) before it reads the body, so a refused request never costs its upload. The runner's reader closes stdout at EOF, because a close from the executing thread blocked on the buffer lock and could SIGPIPE a command still running. The page title counts only inbox items that are not snoozed. Committed on the operator's word despite the gate: cost and time are estimates under 98 % other load and include eight new test cases. 🎯 Fix: 88 % 💥 Break: 3 % 💸 Cost: 106 % ⏱️ Time: 113 % 🛡️ Security: 100 % 🏆 Score: 71 🚦 Gate: ⛔ fix 88 % < 90; cost 106 % > 100, time 113 % > 100: ask the operator before the commit ``` /home/kevinveenbirkenbach/Repositories/github.com/kevinveenbirkenbach/skills/.worktrees/feature/kvb-hub ``` Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
8ba45efd04 |
feat(hub): kvb-hub review hub with passkey-signed approvals
The kvb-hub is a local review hub in a container bound to 127.0.0.1: one page, JSON APIs and an MCP endpoint for the operator's inbox, asks, decisions, staged-commit reviews, the radio stream, the agent graph, the worktree tree, push-to-talk and read-aloud radio (Piper, whisper.cpp) and an optional LLM command line. The hub is only a relay. The host executors in hub/client (the runner user service, the PermissionRequest hook and the approval CLI the commit skills call) verify a WebAuthn passkey signature against keys pinned outside every sandbox, refuse replays and fail closed. scripts/hub.sh with make hub-install, hub-check, hub-pin, hub-runner and hub-hook installs and checks the hub; hub-hook writes ~/.claude/settings.json after a backup. make watch-check shows a stuck watch hook. The new kvb-git-diff skill and the commit skills route reviews through the hub; the test_hub_* suites replace test_hub.py. Security review (kvb-think-criticize, three skeptics) fixed four holes: signed rules auto-approved chained shell commands; the client and LLM openers followed redirects, the LLM one with its bearer key; the worktree scan ran git fsmonitor and filter commands from agent-writable repos; compose set no memory, pids or health limits. Documented until core requirement 093, on the operator's word: the hub page is trusted while signing and pinning, local processes can forge the unsigned operator actions, and the container reads the host /tmp. Committed on the operator's word despite the gate: the passkey path is not yet verified live. Accepted MEDUSA false positives (137 findings on 115 lines, rows in security-accepted.jsonl): EXA002 x17, MEDUSA-MM-SCAN-058 x12, B603 x8, MEDUSA-FTSC-SCAN-089 x7, MEDUSA-PRIV-SCAN-762 x6, AIC-EXFIL-005 x5, EXA006 x5, MEDUSA-FTSCAN-231 x5, WEB-SSRF x5, MEDUSA-INF-SCAN-2057 x3, RH005 x3, TC001 x3, LO002 x2, MEDUSA-AEXP-SCAN-096 x2, MEDUSA-CGS-SCAN-028 x2, MEDUSA-INF-SCAN-1350 x2, MEDUSA-INF-SCAN-2369 x2, MEDUSA-INF-SCAN-2440 x2, MEDUSA-INF-SCAN-2638 x2, MLSC-SERIAL-002 x2, PL002 x2, PQC005 x2, SAST-PATH-001 x2, TC004 x2, TC005 x2, TC006 x2, TC007 x2, AA-146, EXA007, FINE-050, GB-ENCODE-001, GB-ENCODE-003, MEDUSA-CGS-SCAN-029, MEDUSA-CGS-SCAN-103, MEDUSA-DP-SCAN-1034, MEDUSA-DP-SCAN-1916, MEDUSA-DP-SCAN-954, MEDUSA-EXF26-TOOL-004, MEDUSA-FT-TRIGGER-005, MEDUSA-INF-SCAN-2206, MEDUSA-INF-SCAN-2457, MEDUSA-INF-WEB-003, MEDUSA-INFRA-SCAN-030, MEDUSA-MP-SCAN-1135, MEDUSA-RLHF-POISON-002, MEDUSA-RT-SCAN-006, MM-ENCODE-001, MP-124, PQC001, PQC002, PQC006, RP001, TC009, TC010, WEB-FLASK-004. 🎯 Fix: 80 % 💥 Break: 4 % 💸 Cost: 129 % ⏱️ Time: – 🛡️ Security: 100 % 💰 Value: ? 🏆 Score: 60 🚦 Gate: ⛔ fix 80 % < 90; cost 129 % > 100: ask the operator before the commit ``` /home/kevinveenbirkenbach/Repositories/github.com/kevinveenbirkenbach/skills/.worktrees/feature/kvb-hub ``` Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |