The grant order ignored live load: a prio-1 hotfix that fit the budget but waited on RAM or CPU counted as granted for the entries after it, so CPU-light jobs behind it could start, keep the load up and starve it. Such a hotfix now holds back every entry after it until it starts (operator pick 2026-11-11 on Main's report). Each holder stores the reason of its last poll, and list shows it, so a job the load holds back no longer reads as granted, without list sampling the CPU.
🎯 Fix: 92 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 88
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
operator.py add accepted prose, slash commands, aliases and <slot> commands as runnable rows, so the operator got bash blocks that do nothing or fail. A command must now start, after VAR=value words, with a program on PATH, a path to one or a shell builtin, and hold no <slot>; anything else is refused unless the row is a decision. Older rows of that kind stay in the table but get no bash block and are skipped by run. Ordered by the operator on 2026-11-11.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A MEDUSA finding without a line number (a Dockerfile-level rule, for example) had no fingerprint, so security.py accept could never waive it and the score stayed below 100 after review. Such a finding is now keyed by the sha1 of the whole file; any edit of the file brings it back.
🎯 Fix: 95 %
💥 Break: 2 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 93
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A command with 5+ runs is booked at its own p90 RAM and CPU share, an unlearned one at its kind's p90. The old rule took the larger of kind and command, and for an unlearned command the largest command of its kind: test-nostack-788 (docker build) sat 90 min at position 1, booked at 42.6 % RAM from a legacy gate.py hold row while test-nostack measures 5.6 %; skills make test (3.3 %) was booked at the test kind's 35 %.
🎯 Fix: 93 %
💥 Break: 5 %
💸 Cost: 99 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
evaluate counted every unblocked waiting entry as granted for the entries behind it, but only that entry's own wait() poll turns it active; a living holder that stopped polling kept its slot and share forever and wedged the queue. Every poll now stamps polled; a waiting entry silent for CLAIM_SECONDS (3 polls) reserves nothing and is listed last, and once it polls again it takes a fresh seq and enqueue time, so it rejoins at the back of its rank.
🎯 Fix: 92 %
💥 Break: 3 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cache_files walked every cache_dirs entry with os.fwalk, which raises FileNotFoundError for a folder that does not exist (~/.cache/ms-playwright on this host), so cleanup --caches crashed and trim_caches never ran at disk stage warn. A root that cannot be walked is now skipped and the other roots are still trimmed.
🎯 Fix: 95 %
💥 Break: 2 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 93
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The inbox selects several items by checkbox and denies them together through one passkey proof; the cursor stays on the clicked row after the redraw. Keyboard d denies the selection when one exists.
🎯 Fix: 90 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 86
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With two passkey requests open at once, the hub received one request's payload with the other request's assertion and refused it with the challenge is not the one of the payload. signedProof now refuses a second signing while one runs and tells the operator to finish the first.
🎯 Fix: 89 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 85
🚦 Gate: ⛔ fix 89 % < 90: operator approved the commit
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
install.sh removed ~/.claude/skills and ~/.agents/skills and created them anew. The Agent Orchestrator bind-mounts ~/.claude/skills, and a new inode left that mount showing the deleted directory, so the hub's radio send failed with radio.py not found after every make install. The installer now keeps each directory and deletes its contents before copying.
🎯 Fix: 93 %
💥 Break: 3 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 90
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Day/Night menu lists every claimed agent with a checkbox and an All box; Drain, Suspend and Resume act on the ticked agents only, are disabled when none is ticked, and the agent popup carries the same three buttons for that one agent. Each action is passkey-signed with its agent list, night.py takes --target per session and keeps a per-agent part, and each chosen agent gets a call naming the action and its skill. The night skills say how to read a per-agent night.
🎯 Fix: 90 %
💥 Break: 5 %
💸 Cost: 106 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 81
🚦 Gate: ⚠️ cost 106 % > 100: operator approved the commit
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every Agents view marks each agent as working, alive, sleeping, suspending, unreachable or dead with an animated symbol; a Status menu hides states and switches the 3D faces, kept through a reload. The hub's container sees neither ~/.claude/sessions nor host processes, so the host timer kvb-orchestrator-sessions.timer runs client/sessions.py every 15 seconds while the page polls: it reads the session records without following symlinks, only regular files up to 64 KiB, keeps name, status and statusUpdatedAt within type and length bounds, checks the pid's process start, and writes agent-sessions.json atomically. The hub ignores a snapshot that is missing, malformed, older than five minutes or dated in the future. make orchestrator-install enables the timer and orchestrator-check reports it.
🎯 Fix: 90 %
💥 Break: 5 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 86
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pronounced() rewrites a radio text for speech in four passes: paths become file or folder with name and parent (/tmp as temp, ~/ as home), dates with or without a clock become a date-time group converted to Zulu, times become four ICAO digits with Zulu for UTC, and hex runs, version numbers and the remaining digits are spelled in NATO and ICAO words with Decimal between digits. Invalid dates and times stay as written; the displayed text is unchanged.
🎯 Fix: 90 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 86
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A command that exits non-zero or cannot start sets its row stopped with the exit under failed and posts it to every agent holding A, asking for the corrected command; its last stderr lines go along only when the operator answers y, since stderr can hold a credential. Exit 130 or a signal keeps the row open. A cancel needs a reason and posts it the same way. add refuses commands with control characters and run never offers such a row, so a newline can no longer run a hidden second command. A leftover background process holding stderr no longer blocks run. run.md describes what run shows and asks.
🎯 Fix: 90 %
💥 Break: 4 %
💸 Cost: 140 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 62
🚦 Gate: ⚠️ cost 140 % > 100: operator approved the commit
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The operator found the command hard to spot in the run output. run now prints each step's command bold cyan when stdout is a terminal and NO_COLOR is unset or empty (no-color.org); piped or captured output stays byte-identical. Only the command line is colored. The new test runs run on a pseudo-terminal with and without NO_COLOR and piped. Operator order relayed by the Maintainer session on 2026-11-06; the change was written by a subagent and reviewed here.
🎯 Fix: 95 %
💥 Break: 1 %
💸 Cost: –
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 94
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
```
/home/kevinveenbirkenbach/Repositories/github.com/kevinveenbirkenbach/skills/.worktrees/feature/kvb-org-operator-run-color
```
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Agents name the directories of operator rows and can write the repos behind them. Before the operator answered, list and run started git there unsandboxed, and git ran commands the repo config names: core.fsmonitor, the post-index-change hook on an index refresh, filter clean drivers, the pager, external diff and textconv drivers, gpg through log.showSignature, and a lazy fetch of a missing object that runs remote uploadpack through sh. Preview output and task names also reached the terminal raw, so escape sequences could overwrite the displayed command. Every git call before the yes now runs as git -c core.fsmonitor=false -c core.hooksPath=/dev/null -c log.showSignature=false --no-pager --no-optional-locks with GIT_NO_LAZY_FETCH=1, diffs add --no-ext-diff --no-textconv --ignore-submodules=all, a repo whose config names filter commands gets no working-tree preview, and preview output, task, check, table cells and listed commands go through visible(). The approved command itself still runs through $SHELL -ic unchanged. Found by a skeptic while scanning the color change; each hole was reproduced red before the fix.
Accepted MEDUSA false positives after kvb-think-criticize (rows in security-accepted.jsonl): MEDUSA-MM-SCAN-058 x3, MEDUSA-GENAI-SCAN-1270 x4, MEDUSA-WM-SCAN-354, MEDUSA-CGS-SCAN-040.
🎯 Fix: 90 %
💥 Break: 3 %
💸 Cost: –
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 87
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
```
/home/kevinveenbirkenbach/Repositories/github.com/kevinveenbirkenbach/skills/.worktrees/feature/kvb-org-operator-run-color
```
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-release listens once, writes one brief since the previous tag and hands each channel its slice; kvb-release-article writes the long form and kvb-release-post the short posts, each listening before it brainstorms. A project's own release skill sets file names, title and post structure, and an empty channel list means every channel the project covers. kvb-release-changelog moves its lint steps into lint.md.
🎯 Fix: 92 %
💥 Break: 3 %
💸 Cost: 279 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 32
🚦 Gate: ⚠️ cost 279 % > 100: operator approved the commit
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The header radio button opens the radio spot above the main area of the current tab. The Send button, the toolbar radio button and the Radio tab are gone, the mayday clears inside the spot without a Cancel mayday label, and the radio station probe waits for a visible row.
🎯 Fix: 92 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 88
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Requirement 001 decisions 52 to 58 (operator 2026-11-09):
- Radio spot: the Agents toolbar radio button expands inline with the one
composer (every channel, post and call recipients, Record, Send), a gear
with the radio settings (tones, squelch, read aloud, moved out of the
general settings) and Cancel mayday, which leaves the toolbar; the Radio
tab keeps speakers, stations and the log plus a jump to the spot.
- Day/Night menu: the badge and Drain, Suspend and Resume only; each
explanation is the button's tooltip.
- Tasks toolbar: views, filter, estimation, sort and business value in one
wrapping row with a Request choices dropdown; the status chips stay the
second row.
- Burndown and burnup: a trend selector (off, linear regression, velocity
over N days) stored in the hub settings, the trend lines and the forecast
date on the chart, and a tooltip per point with the work packages that
started, ended or joined there.
- ETA: every list row shows its ETA (end when done, else start or now plus
the forecast length behind its predecessors), sharing one predecessor
walk with the Gantt.
- Tasks 3D: a sixth view on the 3D engine with the dependency edges in the
Gantt colours, layouts Force and Time.
- Steady 3D camera: the view fits once per graph build, refreshes keep the
operator's camera.
- e2e: Firefox runs with its profile as cwd, so its crashreporter no longer
drops core dumps into the repository.
- A forecast beyond the last second a Date can hold is no forecast, so a
nearly flat regression no longer aborts the Tasks render.
Evidence: make test Ran 717 OK, ruff clean, markdownlint 0 errors; e2e 30 of
30 OK in Chromium and Firefox at 1920, 1366 and 390 px; MEDUSA 0 findings
after one skeptic-accepted lexical EXA006.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A new release category groups what announces a release (changelog,
article, posts); the changelog skill is its first member. The naming
test lists the category and the README organigram places the skill
under it.
BREAKING CHANGE: /kvb-git-changelog is now /kvb-release-changelog;
make install replaces the installed copy, kvb-skill-cleanup removes the
legacy directory.
Evidence: make test green, ruff and markdownlint clean
(/tmp/rename-changelog-make-test-2.log); no reference to the old name
left in the tree; MEDUSA 5 files, 0 findings.
🎯 Fix: 97 %
💥 Break: 3 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 94
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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>
Requirement 001 phase 6, S4 (decision 29):
- Skills renamed: kvb-org-evening -> kvb-org-drain, kvb-org-sunset ->
kvb-org-suspend, kvb-org-sunrise -> kvb-org-resume, kvb-agent-relax
-> kvb-agent-drain, kvb-agent-sleep -> kvb-agent-suspend,
kvb-agent-wakeup -> kvb-agent-resume; their text uses the new verbs and
every reference in skills, README, orchestrator and tests follows.
- night.py takes drain, suspend, resume, resume --alone and status. A
state file written before the rename keeps working: mode evening reads
as drain, sunset as suspend, the awake list as resumed and is rewritten
on the next save; the hub's council.night_mode maps the same values.
mode: day and handover_night_<session>.md keep their names.
- Orchestrator: the Day/Night menu shows Drain, Suspend and Resume, each
confirmed, signed through /api/board/sign with the passkey and run as
the existing argv job; the old action names are refused.
- kvb-skill-cleanup: git rates two renames at 44 and 50 % similarity,
under its default threshold, so legacy.py detects renames from 40 %.
- tests/test_night_names.py fails on any old name in the tree outside a
rename record "old ->".
BREAKING CHANGE: the six old skill names and night.py's clear and wake
are gone. Run make install (it reinstalls the orchestrator too), then
kvb-skill-cleanup to remove the old installed copies; until the install,
the running hub calls the old night.py path.
Evidence: make test 679 OK, ruff and markdownlint clean
(/tmp/s4-make-test-1.log); e2e 22 OK in Chromium and Firefox
(/tmp/s4-e2e-3.log); MEDUSA 23 files, the 2 findings are the
byte-identical atomic state write that moved with night.py, already
accepted after a skeptic at its old path, rescan 0 open.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Requirement 001 phase 6, S3:
- faces.js derives a face from the agent's stable key (its session id,
or name:<name> without a session), so a rename keeps the face: shape
4, colour 7, eyes 5, brows 4, mouth 5, hair or headwear 6, accessory 5,
84,000 combinations. It emits constant SVG tags and attributes only.
- The emoji avatar is gone: popup, Orbit and Chord nodes, the radio
view, stream and palette draw the SVG face (the 3D views keep their
marker).
- While the radio reads an agent's message, its faces carry the class
speaking and the mouth runs the talk keyframes; a silent face does
not move, and prefers-reduced-motion stops the mouth.
Evidence: make test 673 OK, ruff and markdownlint clean
(/tmp/s3-make-test-1.log); e2e 20 OK in Chromium and Firefox, none
skipped (/tmp/s3-e2e-3.log); screenshots
/tmp/orchestrator-phase4-shots/*-speaking.png; MEDUSA 13 files, 2
findings ("PLA" in "templates", the UI call openAgent) accepted after
an independent skeptic that found no markup sink or selector injection
in the face code.
🎯 Fix: 94 %
💥 Break: 3 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 91
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Requirement 001 phase 6, S2:
- Agent popup: a click on an agent opens one fading <dialog> with face,
name, earlier names, offices, tasks, claims, subagents, history and
the composer; Escape or a click outside closes it. The agent drawer
and the Details side panel are gone.
- Node popup: a click on a node in a graph view enlarges it and marks it
red, and an SVG leader line ties it to the translucent popup; a click
elsewhere restores the node. popup.js places the popup right, left,
below or above the node, caps its height to the room there and
returns the line.
- Search: in the 3D views a search field with suggestions replaces the
list of every node, one chip per distinct label.
- Fit: body.fit now applies at every width; without the side panel the
orbit view otherwise grew a page scrollbar at 390 px.
- Two page modules had silently shadowed each other's functions of the
same name while this was built; test_no_two_page_modules_declare_the_same_function
now guards that. Artifact ids carry NUL, which CSS.escape turns into
U+FFFD, so the node lookup compares dataset values.
- Composer: a message goes to the agent whose popup it was written in.
deliver() read the global ui.agent after awaiting the upload, so
closing the popup or opening another agent during an upload sent the
signed message to null or to the other agent (found by the security
skeptic, present before, easier now that a backdrop click closes).
Limits: the 3D leader line is placed when the popup opens and does not
follow the camera while the layout settles; the search keeps the first
node per label, so a second repository's main branch is found in the
graph, not through the search.
Evidence: make test 667 OK, ruff and markdownlint clean
(/tmp/s2-make-test-1.log); e2e 19 OK in Chromium and Firefox at 1920,
1366 and 390 px, none skipped, rerun after the composer fix
(/tmp/s2-e2e-4.log); screenshots in /tmp/orchestrator-phase4-shots/;
MEDUSA 22 files, 6 findings (a "PLA" pattern in "templates", the UI
names openAgent and drawAgent) accepted after an independent skeptic
that also found no XSS or selector injection in the popup code.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Requirement 001 phase 6, S0:
- Speech: the security class shows as "Sécurité" and is spoken as in
maritime radio, "say-cure-ee-tay" three times. A hex run of 7 or more
characters with a letter and a digit (commit, session, message ids) is
spelled in the NATO alphabet; every other digit is spoken on its own
in ICAO pronunciation (42305: four two tree zero fife). The spelling
runs inside clip() before the length cut, so a phrase still fits
SPEECH_LIMIT; the displayed text keeps its digits and hashes.
- English: every orchestrator label, tooltip, message, error, script
output, document and test string is English (Regeln -> Firewall,
Hierarchie -> Hierarchy, Steuerung -> Controls, Ausführen -> Run, ...);
the page declares lang="en" and formats dates as en-GB. Task filter
ids are English. German stays only in the skill trigger phrases that
quote the operator, and as identifiers: the view keys (view=) and the
leitstand id.
- tests/test_english.py scans every tracked text file for umlauts and
German words, with an allowlist of the quoted trigger phrases.
BREAKING CHANGE: the Stream search tokens art: and seit: are now kind:
and since:; an old art: or seit: token counts as a plain word.
Evidence: make test 657 OK, ruff and markdownlint clean
(/tmp/s0-make-test-1.log); e2e 17 OK, none skipped (/tmp/s0-e2e-1.log);
MEDUSA 86 files, 24 findings on reworded strings accepted after an
independent skeptic (21 path:line, 0 rejected, no behaviour change in
signing, approving, rules, denying, offices, events), rescan 0 open.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Requirement 001 phase 6, S1:
- Fit: Agents and Artifacts in a graph view fill the window from 861 px
up (body.fit flex column); the e2e walk probes #agents/orbit at every
width for a page no taller than the window.
- Rename once: a session seen under two names by one source (two claims)
flipped its name on every state build, so the Stream logged a rename
each poll. Names.observe keeps only the newest sighting per session
and source, and the timeline emits each (session, before, after) pair
once.
- Deutschlandfunk: root cause was the signed media-src, which never
listed the redirect target f111.rndfnk.com that st01.sslstream.dlf.de
answers with (fixed in ba97f7c). The e2e now replays that chain, a
302 from one https origin to a second, in Chromium and Firefox, with
self-signed throwaway certificates the drivers accept.
- Squelch: a settings slider 0..1 scales every tone; at 0 no beep plays,
music on or off (Voice.beep skips a silent tone). Stored settings
without the key load with 1.0.
- Tree: the missing tree file error names `pkgmgr branch tree` and the
release and install step that makes the timer write it.
Evidence: make test 655 OK, ruff and markdownlint clean
(/tmp/s1-make-test-1.log); e2e 17 OK, none skipped (/tmp/s1-e2e-3.log);
MEDUSA 14 files, 0 findings.
🎯 Fix: 94 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 90
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With depth = time, #agents/3d drew subagents behind the session that
spawned them. A subagent took the mtime of its meta.json, which is
rewritten when it finishes, and a session took the moment the hub first
saw its name, or the window end when it had none.
Sessions now take the first line of their transcript and subagents the
first line of theirs, both from /api/subagents, with the old values as
fallback. SpaceView.notBeforeParent moves a spawned node without a time,
or with one before its parent's, to its parent's time, nested subagents
included, so no subagent lies before the agent that spawned it. The 3D
signature carries the time, so the nodes move once the lifetimes load.
🎯 Fix: 93 %
💥 Break: 2 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 91
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Deutschlandfunk stayed silent: its signed URL answers 302 to
f111.rndfnk.com, an origin the signed media-src never listed, so the
browser blocked the redirected load and reported only the station's own
URL in securitypolicyviolation.
The hub now resolves each https stream's redirect chain server side when
a settings save adds or changes it and stores the extra origins as the
stream's via list. media-src lists them next to the signed origins. A CDN
host of three or more labels widens to its siblings (*.rndfnk.com); an
IP, a two-label host and a name under a country code's generic second
level stay exact.
Hardening:
- Only the signed URL itself may lie at a non-global address; every
redirect hop dials a global address only, checked on the address
actually dialed (NAT64 and IPv4-compatible IPv6 count by the embedded
IPv4 address).
- A resolve answers within three socket timeouts even for a server that
drips its headers; a late probe goes on in a daemon thread and a url
still in flight is not probed twice.
- A failed resolution stores via null and retries on a save ten minutes
later; the radio page says so instead of blaming the CSP.
- A save whose settings changed while it was resolved fails instead of
overwriting them (compare and set under the settings lock).
- A new stream URL at an already signed origin takes the passkey too.
- Settings written before via existed load as via [] and never dial.
Residuals:
- The passkey proof is spent before the resolve, so a save racing
another one fails and the operator signs again.
- A save that hits a due retry blocks up to twelve seconds and may
reload the page.
- A page loaded before the upgrade throws on the missing via until it
reloads.
- A private multi-tenant suffix such as github.io still widens.
- getaddrinfo has no timeout; it runs inside the daemon thread.
Operator step after install: remove Deutschlandfunk in Radio, Sender
bearbeiten, save, add it again and save with the passkey.
🎯 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>
Root cause of the 2026-11-07 20:14 OOM: the queue granted jobs per kind
only, so several heavy kinds and direct heavy commands outside the queue
ran side by side until the kernel killed the editor.
- queue.py: every grant checks the sum of the learned RAM and CPU
footprints of active and granted jobs against budget_percent; an
unlearned kind takes the whole budget and runs alone. RAM used plus
what young jobs (RAMP_SECONDS) may still take must stay below
load_limits. Memory stage stop blocks every grant.
- watch.py: memory stage from MemAvailable, swap, PSI full and the swap
of the container holding most of it; the block hook classifies
builds, make install/verify-*, and pytest or unittest beyond one file
as heavy through wrappers, sh -c, command substitutions and heredocs
a shell reads.
- history.py: footprints per kind and signature, seed RAM for pending
shares.
- operator.py: [h]old queues a heavy row through queue.py hold and asks
again on its grant, with a flushed tty and a timeout; sudo rows, which
need the terminal, stay direct.
Residuals: a shell reading code from a pipe or here-string, its own
script, a program named by a substitution, quotes inside an expanding
heredoc, an escaped backslash before a substitution; a lone job waits
while the host is over load_limits.
🎯 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>
Add a Tasks tab that lists the kvb-org-rasci register next to the inbox
and the heavy-job queue, in list, board and RASCI matrix views, each
task with its roles, WSJF, open work and the merge flow of its worktree.
- Task priority pins and WSJF estimates are signed with the operator's
passkey and applied by the runner through rasci.py; the hub verifies
the signature itself before it queues a job or logs the change, the
runner verifies it again and runs only the checked argv, from /, with
a pinned PATH and a 30 s process-group limit.
- The runner refuses task changes into a council that is a symlink, not
owned by its uid or given by a relative path, and while the installed
council.py, nofollow.py or rasci.py differ from the tree.
- Only tasks the hub can sign whole are editable (task slug and length).
- Page lookups keyed by register text use Object.hasOwn, so a status or
party named like an Object.prototype member cannot break the views.
- The inbox screen renders through the tasks screen, so its task filter
shows with "Wartet auf Operator" pressed.
Tests: 604 unit tests OK, e2e 15/15 OK (Chromium and Firefox).
🎯 Fix: 93 %
💥 Break: 5 %
💸 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>
Stream -> Radio gets one line per station with state, play and stop,
autoplay and CSP refusals as row state and toast, and the speaking agent
pulsing there and in the Orbit and Chord views. The composer sends on all
seven channels; push-to-talk records with MediaRecorder, whisper.cpp
transcribes off the request thread under per-minute and per-hour caps,
and the recording travels only when "Aufnahme anhängen" is ticked.
Every operator message, the mayday clear, the day and night board
actions and a settings save that adds a stream origin take the passkey;
the hub verifies each signature once against the pinned keys before it
relays or acts. The container gets TMPDIR=/scratch so that verification
no longer fails on the read-only root.
A denial now signs its recipient: the inbox and the signing step resolve
it through one function, a claim set naming several agents leaves a
direct denial without recipient, and a recipient that changed after the
inbox showed it is refused with 409. operator_message returns only the
signed reason for a denial, never the agent's own words the radio row
quotes, and the permission hook hashes the signed recipient too.
🎯 Fix: 95 %
💥 Break: 3 %
💸 Cost: 99 %
⏱️ Time: 68 %
🛡️ Security: 100 %
🏆 Score: 100
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The inbox shows the console output of a command under its "Ausführen"
row, kept to the same tail the job list shows. A deny picks its radio
channel (post by default, call, newsletter, broadcast, pan-pan,
security) and an optional reason; the passkey signs denial, channel and
reason, the hub verifies the proof before it stores or sends anything,
and the PermissionRequest hook quotes a reason only when it verifies.
Agents ask for an auto-approval rule through the MCP tool rule_request;
the request shows in the inbox with a dry run over the last seven days
and is accepted only with a passkey assertion the hub verifies against
the pinned keys and the hub origin, for every rule kind. Requests lapse
with the ask lifetime and are capped per agent and in total.
The image runs as nobody by default and carries the health check
command; compose keeps the timing.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 98 %
⏱️ Time: 75 %
🛡️ Security: 100 %
🏆 Score: 100
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Agents views gain a Hierarchie view: an org chart of the chair and
the line offices read from the council claims. The operator appoints,
hands over and recalls an office there; the page signs each change with
the passkey and a host unit (client/offices.py, offices path unit)
verifies the signature, removes a live holder itself and runs a plain
council.py claim for the successor. council.py loses the --operator
flag, so no agent can override a live holder.
The agent order is now separate from task priority: rasci.py agents
order stores it in agent-order.json beside the register through the
nofollow helpers and never writes a --prio pin. It only breaks ties of
equal task rank, also among the queue grants. The Hierarchie view edits
it as a drag list.
Host units run the installed skill scripts, so orchestrator.sh keeps the
offices unit off while the installed council.py, nofollow.py or rasci.py
differ from the tree. Job files, council files and hub records are read
with O_NOFOLLOW and O_NONBLOCK and size caps; nofollow.rewrite replaces
a claim without creating directories, so a beat never revives a released
office.
🎯 Fix: 92 %
💥 Break: 4 %
💸 Cost: 100 %
⏱️ Time: 79 %
🛡️ Security: 100 %
🏆 Score: 100
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The council root is writable by every agent, and operator.py runs
rasci.database() outside the sandbox. A symlinked rasci.json made it
read any JSON file the operator can read and write it back into the
council as a regular file; a symlinked rasci.lock made it create files
at the link target; a FIFO hung every register call.
load() now reads through nofollow.read_text and database() locks and
writes through nofollow.locked and nofollow.replace. A register that is
no object of task objects, or nested past the recursion limit, exits 2
instead of a traceback. nofollow.open_below also refuses a file with
more than one link, so a hard link to a file outside the council is
refused like a symlink.
🎯 Fix: 96 %
💥 Break: 3 %
💸 Cost: 99 %
⏱️ Time: 95 %
🛡️ Security: 100 %
🏆 Score: 99
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 4 of requirement 001. Agents and Artifacts each offer Orbit, Chord, Swimlane, Tree and 3D: the spawn tree (page-spawn.js) and the artifacts orbit (page-artifacts.js, kvb_orchestrator/artifacts.py) replace the "folgt" placeholders, 3D runs on the vendored 3d-force-graph and three.js (MIT, LICENSE files kept, served from the image under the unchanged CSP), and a Bootstrap Icons sprite (icons.py, MIT) replaces the emoji glyphs, also in the meditation page.
Signed branch deletes: the Tree view offers a delete only for merged-clean branches without a gc skip reason; the operator signs the exact item with the passkey, the hub writes a job without any command (kvb_orchestrator/cleanup.py), and the host path unit runs client/cleanup.py, which verifies the signature, dry-runs pkgmgr branch cleanup for that exact branch, refuses patterns, option names, changed tips, other worktrees and plans naming other branches, deletes, and rewrites the branch tree. pkgmgr runs with core.hooksPath=/dev/null, core.fsmonitor=false and the config hook events disabled through GIT_CONFIG_COUNT, because the cwd is a repository agents can write and pkgmgr releases before 4c8b137 delete without scan mode (test with a positive control that both hook kinds fire without it).
3D labels go into the tooltip as a text node: the vendored tooltip renders a string label as HTML.
Tests: 503 OK, Chromium (CDP) and Firefox (BiDi) walks over every screen at three widths.
MEDUSA acceptances (kvb-think-criticize skeptic 2026-11-07): WEB-FLASK-004 view.py:135 (file template, autoescape on, state via tojson), MEDUSA-MM-SCAN-058 server.py:69/:85 (substring pla in --templates and Places), RH005 client/cleanup.py:199 (raise SystemExit(main())), B704 icons.py:45 (sprite baked into the image, names checked against it), EXA002 page-artifacts.js:221 (openAgent only opens the drawer), STG002 page-constants.js:64 (U+2212 in a static label), PL002 page-graph3d.js:44 (CSS class space-tools; the real tooltip sink fixed), B603 client/cleanup.py:75 (argv without shell; hook hole closed).
🎯 Fix: 92 %
💥 Break: 5 %
💸 Cost: 100 %
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 87
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New skill with ansible_tasks.py: gaps, durations, kinds, kind and task rank a run's log by wall-clock gaps or profile_tasks lines; profile-report merges cProfile dumps of the controller and its forked workers (sitecustomize hook); bench-carrier and the bench playbook measure per-task overhead against variable size. kvb-code-tune hands its Ansible timing to it.
🎯 Fix: 92 %
💥 Break: 2 %
💸 Cost: new
⏱️ Time: new
🛡️ Security: 100 %
🏆 Score: 90
🚦 Gate: cost new, time new: cleared by the operator's commit word for kvb-check-ansible (2026-11-07)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 3 of requirement 001. The orchestrator no longer starts git: orchestrator/kvb_orchestrator/tree.py maps the document that pkgmgr branch tree writes (repos, branches, worktrees, detached heads, delete_ready) onto the Tree view, re-reads it when mtime or size change, and reports a missing or broken file as an empty tree with the error. compose.yml mounts the tree directory read-only (a file mount would keep the replaced inode). /api/state touches state/tree-heartbeat; kvb-orchestrator-tree.timer runs every minute and its oneshot service runs only while that heartbeat is younger than 5 minutes, with PATH=/usr/bin:/bin and the pkgmgr venv that the install builds from the newest v* tag a fully trusted GPG key signed, cloned fresh from PKGMGR_REPO because the local checkout's config is sandbox-writable.
Dialectic verdict 2026-11-06 (2 rounds, 5 skeptics, ~85 %): hardened git runner in pkgmgr, gc classification shared with the tree scan, forge facts only from an unlocked Secret Service token with a 600 s cache, atomic tree output refreshed by open/close/drop/gc, host units outside sandbox-writable paths, orchestrator reads the file only. Losing objections: extending the host runner (it would run bash -c on signed text), a new payload kind, copying the pkgmgr checkout, a tree limited to .worktrees. Operator answers: signed-tag venv, UI delete keeps volumes and is local only, ignored files beyond known caches block a delete, scan only while the page polls. Delete jobs, gc_item signing and the UI delete path follow in phase 4.
MEDUSA acceptances (kvb-think-criticize 2026-11-07, 1 skeptic): MEDUSA-MM-SCAN-058 orchestrator/server.py:64 and :80 false positives (case-insensitive substring pla in --templates and Places; --tree-file is a read-only JSON file, no model loading).
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: –
⏱️ Time: –
🛡️ Security: 100 %
🏆 Score: 89
🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 %
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
