watch.py admits heavy work only while PSI, swap-out rate and RAM forecast stay under the configured limits; config.default.json is the single default the install writes once (mode 600, no clobber) and every hook, admit and queue command falls back to. Browser runners and package-manager script forms count as heavy, preflight derives its probes from the light and heavy tool lists, and the shipped default carries no prune command, buildx path or host-specific key. Tests run once on unify/skills-20261114 after the merge (operator order 2026-11-14).
🎯 Fix: 85 %
💥 Break: 8 %
💸 Tokens: 99 %
⏱️ Velocity: –
🛡️ Security: 100 %
🏆 Score: 79
🚦 Gate: ⛔ fix 85 % < 90, break 8 % > 5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Operator order 2026-11-13: once the operator ran a command, the agent that queued it must hear it directly, without the operator relaying it. Every row close now sends one radio call to the stored initiator (party name plus the session id add pinned; legacy rows by the A name): run and hub ran (done, exit 0), check still open, the check-pass sweep of list and run, done, cancel, failure and hub failed. The call names row, outcome and exit code and never carries command output; shared stderr stays in failed.output of the register. radio.py send takes --sid for a call to exactly one address. Accepted MEDUSA false positives: CGS-SCAN-029 and CGS-SCAN-030 at operator.py:930, which runs the row command only after the operator's per-row answer.
🎯 Fix: 93 %
💥 Break: 4 %
💸 Cost: 105 %
⏱️ Time: 100 %
🛡️ Security: 100 %
🏆 Score: 85
🚦 Gate: ⚠️ cost 105 % > 100: ask the operator before the commit
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
operator.py stores the adding session's name and id with the row. Done,
cancel and failure go to that stored name and pinned session id, never
re-resolved: a radio post by default, a call when that session picked it
with radio.py feedback set call. The caller's own parts are dropped. The
hub counts unanswered calls per session id.
MEDUSA: 34 findings accepted after an independent review, 0 true
positives: 3x CGS-SCAN-029 by design (operator-confirmed runner in
operator.py execute and its call), 2x CGS-SCAN-030 (exec inside the name
execute), CODEGEN-SCAN-003 and CGS-SCAN-028 (open inside Popen),
3x CGS-SCAN-040 (prompt loops end), 4x GENAI-SCAN-1270 (rasci.database()
JSON register), 20x FLA-SCAN-055 (dba inside the word feedback).
🎯 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>
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 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>
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>
operator.py run walks the open rows in list order and shows each step's directory, command (control characters made visible), reason, check and a preview of the change (commits and diffstat to push, files an install changes, what a removal deletes, the repository state) before asking yes, skip, diff or quit; an approved command runs through $SHELL -ic so aliases work, and the row closes once its check passes. list prints one bash block per step plus the run block. The new check kind decision keeps decisions and browser steps out of blocks and run. The skill now requires one row per step and moves the check kinds and the peer sweep into side files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Legacy register rows carry a prose title and no directory. They stay in the table but no longer land in the copy-paste bash block as if they were commands; with no runnable row the block is omitted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Collecting the operator's pending commands meant messaging every agent, waking each for a full turn. operator.py add records a step as register task operator/<slug> (command, absolute directory, reason, operator as R, the adding session as A) with a check from a fixed list: pushed:<repo> (nothing ahead of the upstream), installed:<source>=<installed> (same files, same bytes), binary:<name>, absent:<path> or manual. A stored shell command is refused on purpose, together with relative paths and shell metacharacters, so no agent can run code in another's session through the shared register. list runs the checks, closes what passed, and prints the open steps in dependency order as a table and one bash block; done closes a manual step.
The skill now has every agent add a step the moment it arises and prints list on invocation; a peer sweep by message only on the operator's explicit request. test_operator covers every check kind, the open states and the refusals.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>