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>
The banned constructs per language were a table the agent read into context and applied by eye. defaults.py scans the added lines of git diff HEAD (or the index with --staged, or whole files with --file): Python through ast (parameter defaults, .get(k, v), getattr(o, n, v), os.environ.get / os.getenv with a value, add_argument(default=); None, False, empty strings and empty containers stay permitted, a mutable container as a parameter default is still reported), the other languages through masked regular expressions that skip strings and comments (shell, .env, Make, Dockerfile, Compose, YAML, Jinja, JS/TS, C#, Java, Rust, SQL, Terraform). Each hit prints file:line, language, construct and the replacement, marked candidate where it is often logic (??, || or or with a literal) or justified? next to a comment; exit 1 on hits.
The skill's detection step calls the script and the agent judges every hit; Go zero values, Helm values, Docker ENV placeholders and the judgement stay prose. constructs.md is gone. test_defaults covers every language, the permitted values, --staged, --file and the exit codes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The operator's rule (2026-11-03): nothing is declared 90/5 before every make test* target of the project ran green on the exact head. kvb-check-gate holds it as the single source and gate.py backs it: list parses the Makefile make would read and its includes statically (never running make, whose $(shell ...) would execute), status reports green, red or missing per target for HEAD and says gate: green only when all are green on a clean tracked tree, and run executes known targets one by one as make -C with output written straight to a log (no pipe hiding the exit code), marking a target red on a non-zero exit, a make: *** or make[N]: *** line, or a project --fail-regex such as infinito-nexus' FAILED TARGETS. Results live per sha in the git common dir under a lock, shared by linked worktrees; a new commit or a target that edits tracked files invalidates them.
kvb-check-confidence and kvb-git-pr-label point to gate.py status instead of restating the rule; the README places the skill. test_gate covers parsing with includes, green, red and fail-pattern runs, stale results after a commit, dirty trees, worktrees and usage errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every loaded SKILL.md and every listed description is paid on every turn of every session, and every delivered message costs its recipient a full turn. scripts/budget.py gains SKILL_WORDS = 600, enforced by test_lock, and test_naming forbids a skill naming its own slash command; 53 descriptions and 10 body lines drop it. The 18 longest skills move details into 31 side files read only in their situation (organize's capacity, succession, peers, offices, relay and lineage; cleanup's docker, worktrees, execute and ask; code-tune, changelog, triage, dream, split and others), so all SKILL.md files shrink from 33866 to 26291 words without dropping a rule; an independent review of HEAD against the new text found no loss in 60 of 64 skills and three weakenings, now restored (WSJF re-estimates after an objective change, cleanup's volume checklist, essence's absolute dates).
The kvb-org-organize Peer protocol sends status that asks nothing to the news board, answers nothing that asks nothing, keeps messages terse, and lets light runs admit themselves in the operator's wording; the skills follow it (operator reads the register, relax and wakeup reply through the news, evening collects them there, wsjf asks only for missing estimates, cleanup and triage batch their questions). Skills that explore code ask the codebase-memory-mcp graph first, subagent briefs scale with the stakes and cap their answers, and the skills that need the full register or council tables call show --wide or list --long.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The new skill measures before it changes anything (listing and SKILL.md sizes, hook output, messages and news, subagent runs, context re-read per turn) and works seven levers in order: listing, length, messages by the Peer protocol, subagents, graph first, hooks and the context window; every change waits for the operator's pick, and another repository's skills get a register task instead of a write. kvb-optimize runs it right after kvb-skill-essence.
context.py reports per UTC day, from the transcripts Claude Code keeps, the turns (one per message id, although the transcript repeats usage per content block), context tokens, median, p90, compactions and the shares of turns and tokens above each threshold, so no cron is needed. enable-plugins.js sets CLAUDE_CODE_AUTO_COMPACT_WINDOW to 500000 only where it is unset, as the operator's trial value; silence proposes a new one only with the series as evidence, never on its own. test_context and test_dream cover both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
install.sh copies every skills-lock.json skill into .claude/skills while enable-plugins.js also enables the caveman and ponytail plugins that ship the same skills, so Claude listed 13 long descriptions twice on every turn of every session. enable-plugins.js now removes from the settings' .claude/skills every locked skill whose source is an enabled plugin's repository, taken from MARKETPLACES; .agents/skills keeps the copies for Codex and Gemini, which have no Claude plugin. The install message names it. test_dream checks that plugin copies go and other skills stay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Agents read these tables on every status, organize and matrix step. rasci.py show now prints Prio, WSJF, Task, Status, Worker and A (task slug never cut, title cell 40 characters, worker and A 24), down from 1817 to 1003 words on the live register; show --wide prints the old 14-column table byte for byte. council.py list prints office, holder, state and a 40-character work cell with whitespace collapsed, 1236 to 276 words; list --long prints the old form. The old assertions now run against --wide and --long; new tests pin the compact defaults, cutting, --all, --mine and pipes in titles, and fail against the previous scripts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every delivered SendMessage costs its recipient a full turn of re-reading its context. news.py keeps status that asks nothing in <council root>/news.jsonl: post adds an item for every session or --to named ones (at most 500 kept, under a lock), and its UserPromptSubmit hook prints the unseen items of the last 24 hours (at most 10, own posts left out) with the session's next turn and marks them seen per session; the hook never fails a prompt. show prints without marking. enable-plugins.js registers the hook; test_news covers single delivery, addressing, own posts and broken input.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
find_fix.py decides in_head with one rev-list, runs -S as parallel slices and builds the candidate lines in parallel. Verified before the merge on the precomputed merge tree: test_ci_tools 7/7, byte-identical output on 3 calls in the skills repo and 2 in infinito-nexus-core, -S on core 62.4 s to 9.6 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-org-evening and kvb-org-sunset narrow the organisation to the session the operator types them in; kvb-org-sunrise undoes it. night.py keeps the state in <council root>/night.json: set makes the caller active and lists the other live sessions, status tells each session its part (active, awake, relax, sleep), wake lifts one session alone, clear ends the night; a missing or invalid state means day. Evening messages every other session to run kvb-agent-relax (finish the task, end its loops, start nothing); sunset messages them in waves of three to run kvb-agent-sleep (park at the next safe point, never cut a running job short, end loops, write a ten-line handover into the project's memory directory, reply asleep), and the operator then sets inbound messages to hold. At night the active session admits its own runs through watch.py and leaves the sleepers' offices alone. Sunrise switches messages back to accept, clears the state and sends kvb-agent-wakeup, which restores loops, jobs and roles from the handover; the operator can also wake one session alone by typing it there.
A peer's request to do less now needs no operator approval in the kvb-org-organize Peer protocol, and kvb-org-robot stops a run at its next safe point when the night reaches it. The README gains the agent category's state meaning and the new nodes; test_night covers set, status, wake, clear and a bad state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
git apply --unidiff-zero places an add-only hunk, which has no context to match, at its new-side start line. That line counts every earlier hunk of the diff, so each hunk select() dropped shifted the later add-only hunks down by its size: on the real README a staged node landed in the wrong Mermaid subgraph. renumber() rewrites each kept header from the old start plus the net lines of the hunks kept before it (+1 for a pure insertion, -1 for a pure deletion), with counts that match the kept lines. A new test drops an early hunk and a line of a mixed one and compares the whole staged file; against the previous script it fails with the line one place too low.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The new category agent holds skills that act on an agent's own setup. kvb-agent-optimize sets up the general code index codebase-memory-mcp (MIT, fully local) for every repository in scope and tightens its agent instructions: agents.py reports whether AGENTS.md exists, whether CLAUDE.md and GEMINI.md point to it, dead relative links, whether it names the index, whether the binary is installed and the git state of the shared snapshot; the operator installs the index through kvb-org-operator; the agent indexes with index_repository and check_index_coverage, asks whether the snapshot is shared, sends code questions to the graph first, audits the instructions with prompt-audit and confirms every change. A domain graph the generic index cannot see goes to the line that owns the repository. kvb-optimize runs it as step 9, so kvb-dream does too. test_naming lists the category, test_dream the optimizer, and test_agent_optimize covers the checks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rasci.py gains agents, which groups the open tasks by worker in rank order (operator pins first, then WSJF) and prints rank, agent, count and tasks, and order "A,B,...", which checks the names against that list and, under the register lock, pins the listed agents' tasks 1..n in the chosen order, keeping each agent's own task order, and drops the pins of every agent left out so they rank by WSJF after them. The new kvb-org-priority skill shows the current order, offers keep, up to two reasoned suggestions or the operator's own order through Other, applies it and tells the chair and every reordered agent. kvb-org-rasci documents both commands and names order as the one exception to writing only your own rows. test_rasci covers the order, the pins in show, the dropped pins and the usage errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-git-pr-create opens a pull request for the current branch: it assesses the branch, hands the push to the operator (gipsn) and waits until git ls-remote shows the head, writes a body with one Closes line per verified issue, confirms once, and opens it ready at the kvb-org-split bar or as a draft below it, then runs kvb-git-pr-label. kvb-git-pr-label sets hotfix on a pull request that targets the default branch and fixes a failure there, and 90/5 once it meets the split bar, each with its evidence, creating the labels once and marking a draft ready when it gains 90/5. kvb-git-issue-solve solves open issues one sidetree each through kvb-org-split, skipping fixed ones and those a peer already works on, and opens one pull request per issue.
kvb-git-close-issues becomes kvb-git-issue-close, following the kvb-git-<object>-<verb> scheme: with a pull request it works as before; without one it covers every open issue, and the new done.py lists the default-branch commits that reference each (one git log, foreign repositories excluded). tests/test_close_issues.py moves to tests/test_issue_close.py and gains TestDone on a temporary repository. Six skill descriptions are shortened, keeping their triggers, so the listing stays within its 1500-word budget (1517 to 1493).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Almost all of find_fix's time went to git child processes: one merge-base per hit (2105 x 132 ms for a broad --grep), a single-threaded git log --all -S, and one branch --contains per printed candidate, run one after another. One rev-list HEAD set now decides in_head, -S runs as strided slices of rev-list --all through a thread pool and is sorted back into git log --all order, and the candidate lines are built in parallel. Output stays byte-identical on four real calls (3x to 142x faster). Tests pin the -S order and that the filters are alternatives.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A read-only run of issues.py on infinito-nexus/core#1326 listed #0 and #1 from CI row names like WordPress#0 and docs#1, reported #1324 twice (plain and as infinito-nexus/core#1324 from the linked issue), and offered the pull request's own number. A bare # now needs no word character or slash before it and a number from 1, the required --repo maps the pull request's own repository to the plain form and merges the sources, and the pull request's number is skipped; the skill passes --repo and reads number. On #1326 only #1324 closing remains.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stage_hunks.py takes git diff -U0 of one file, keeps the added lines of an add-only hunk that match a regex and every other hunk whole when one of its changed lines matches, and applies the result to the index with git apply --cached --unidiff-zero --recount, or prints it with --dry-run; exit 1 when nothing matches. Step 4 of kvb-git-commit-next names it for a file that carries two bundles, which replaces the hand-written partial patches needed five times on 2026-11-03. tests/test_stage_hunks.py covers matching, the unstaged rest and the exit codes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A scan on the real host offered 4276 candidates, 4253 of them files of two repository copies in one peer's scratchpad. scratch_candidates now walks with os.walk, skips any directory holding .git, roles, plugins, pyproject.toml, package.json or a Makefile, and descends at most one level below a scratchpad; the real scan drops to 59 candidates. test_transcend adds a nested helper that stays, a deeper one and a repository copy that drop out (negative control against the unfiltered scan fails).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/budget.py holds the listing limits (1500 words in total, 40 per skill) and the one way to count a description, prints the total and the longest descriptions, reads the staged versions with --index, and exits 1 over budget. make budget [INDEX=1] runs it; test_lock.py takes its limits and counting from the script instead of its own copy. It replaces the one-liner that was run by hand a dozen times on 2026-11-03.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A claim made with --session <name> stored the name, but the PreToolUse gate compares the hook payload's session_id, so the gate refused every edit of the claiming session and a later release without the flag exited 3. caller() now takes CLAUDE_CODE_SESSION_ID whenever it is set and falls back to --session only for other runtimes. test_council covers a claim and release with --session under a Claude Code session (negative control against the old caller fails). Claims made with --session before this fix go stale after an hour without a heartbeat and can then be taken over.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every scheduled Update run since 2026-10-31 failed with 'Missing required argument: source': --agent takes several values and swallowed the source that followed it. update.sh now passes the source first. test_update.py had pinned the source as the last argument and so encoded the bug; it now requires the source right after add and claude-code as the only agent (negative control against the old order fails).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every new skill had to trim the trigger phrases of others to stay under 1300 words; the operator raised the total to 1500. The per-skill limit of 40 words stays.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
branches.py lists every open branch of the remote with its lag behind the base and whether it belongs to the operator: a pull request by the operator's login, or a branch whose own commits all carry the operator's email; merged branches drop out and fork pull requests are marked. The skill merges the base into the operator's own branches directly and into others only after consent through kvb-think-listen, never rebases, aborts on a conflict for kvb-git-merge-fix, and hands every push and gh pr update-branch to kvb-org-operator. tests/test_update_branches.py builds a bare origin with own, foreign, merged, PR and fork cases. Operator order 2026-11-03; three descriptions trimmed for the listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
issues.py reads gh pr view --json from stdin and lists every issue the pull request references: closing keywords and mentions in the body and the commits (own repo, owner/repo#n and issue URLs), a trailing number in the branch name and the issues GitHub already links. The skill verifies each one against the diff and its test or CI evidence, asks the operator, then closes a fix that reached the default branch with a comment naming the pull request and commit, or adds Closes lines to the body of a pull request that is still open or merged elsewhere. tests/test_close_issues.py covers every reference kind. Operator order 2026-11-03; two descriptions trimmed for the listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hooks/reminder.sh claims the current clock hour with an atomic mkdir under ~/.claude/.coach-hours/, so only the first prompt of the hour across all agents gets the COACH HOUR hint (at most three lines, most urgent area first, then the request); earlier hours are removed. enable-plugins.js registers it next to the kvb-dream reminder. test_coach covers once per hour, silence within the hour and the cleanup; test_dream checks the single registration. Operator order 2026-11-03.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-coach-meditation writes a local session page with its settings embedded (xdg-open drops a file:// query): generative pentatonic ambience with reverb over an isochronic or binaural alpha layer (8-13 Hz), a WebGL Julia fractal that zooms with the breathing cue (box, 4-7-8 or none), breath-synced glowing type in URW Gothic, +/- keys for the picture's brightness kept in localStorage, and a progress line along the bottom edge. meditation.py validates every setting and refuses a session while one is missing; after every session kvb-think-listen collects feedback that becomes one set per answer, logged in the kvb-coach profile. Adds the category coach: every coaching skill beyond kvb-coach is named kvb-coach-<name>. tests/test_meditation.py keeps script and page in step. Operator order 2026-11-03; two descriptions trimmed for the listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-coach reads the situation (local time, hours at work today, the profile, open operator commands from kvb-org-operator, delegable kvb-org-rasci tasks) and suggests at most one next step per area: psyche, body, sleep, food, the operator's own tasks, and tasks to delegate to agents; short, friendly, with emojis, ordered by what the situation needs and biased to what the operator likes. It runs kvb-think-listen often to learn preferences. coach.py keeps the host-local profile in the gitignored .state/ of the skills repository: likes and dislikes per area (one replaces the other), a capped done-log, and the hours since the last activity per area; tests/test_coach.py covers it. General wellbeing only, no medical advice. Operator order 2026-11-03; two descriptions trimmed for the listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-skill-dream becomes the uncategorized collector kvb-dream. It reads the last 24 hours of every repository and session the agent touched, writes each finding into the memory of its own project, hands its findings to the new kvb-optimize (skill-cleanup, transcend, skill-auto, skill-tune, skill-compile, skill-essence, code-refactor, code-tune, think-gaps, one round of questions) and then runs kvb-cleanup and kvb-git-branch-gc. The separate kvb-skill-tune reminder is gone: one reminder reads the session id from the hook payload and hints each agent at most every twelve hours, and the global install enables kvb-dream-notify.timer, a systemd user timer that recommends /kvb-dream in a GNOME notification daily at 17:00 local time and catches up once after a missed run. enable-plugins.js retires the tune reminder hooks and repoints every earlier dream hook path. Operator order 2026-11-03; three descriptions trimmed for the listing budget.
BREAKING CHANGE: /kvb-skill-dream is now /kvb-dream; kvb-skill-tune no longer has its own reminder hook.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
legacy.py reads the skills repository's git history (renames and deletions of skills/*/SKILL.md), follows rename chains to the current skill, and classifies every installed skill directory as current (shipped or locked), project (tracked in the repository that holds the root), legacy (with its successor), removed, or unknown. kvb-skill-cleanup scans ~/.claude/skills, ~/.agents/skills, ~/.codex/skills and the projects' .claude and .agents skills, asks the operator, and deletes by re-running the installer, by rm -r where the sandbox allows, or through kvb-org-operator; a rescan verifies. tests/test_skill_cleanup.py covers a two-step rename chain, a deletion, a locked, a project and an unknown skill, and invalid input; a broken one-step resolver fails it. Operator order 2026-11-03; four descriptions trimmed for the listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-skills-dream, -tune, -auto, -essence and -transcend become kvb-skill-*, and tests/test_naming.py now requires the category skill. enable-plugins.js registers the reminder hooks under the new names and repoints hook commands of the previous kvb-skills-* paths as well as the older autotune/dream paths; test_dream covers both generations. References in kvb-check-confidence, kvb-org-perform, kvb-org-rasci, kvb-transcend, the tests and the README organigram and skill map follow. The header comments of both reminder hooks and enable-plugins.js are dropped.
BREAKING CHANGE: /kvb-skills-dream, /kvb-skills-tune, /kvb-skills-auto, /kvb-skills-essence and /kvb-skills-transcend are now /kvb-skill-*.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds kvb-transcend (collector), kvb-skills-transcend (into the skills repo) and kvb-code-transcend (into the current repo). transcend.py scans the own scratchpad, peer scratchpads and the agent homes for scripts, tests and skills the skills repo does not ship, skips secret-named and oversized files, and remembers carried/dropped decisions by content digest in the gitignored .state/transcend.json, so a rescan offers only new or changed candidates. Wired into kvb-skills-tune (own scripts), kvb-code-tune step 6 (measurements) and kvb-skills-dream step 7. Seven descriptions are tightened to stay inside the 1300-word listing budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The operator dropped the conversation shortcuts (sobu, soco, socobu, socon, sona, sote) as unnecessary. Deletes the skill and tests/test_shortcuts.py, drops the tokens from the kvb-code-no-comments triggers, the shortcut branch from kvb-skills-auto step 3, the shortcuts row option from kvb-skills-tune, and the nodes and edge from the README organigram and skill map.
BREAKING CHANGE: the shortcut tokens no longer expand; use /kvb-git-commit-next and /kvb-git-commit-no-verify.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kvb-org-wsjf: the chair asks its subordinates for Fibonacci estimates of business value and complexity, rasci.py estimate records them per party, and WSJF = median value / median complexity ranks the tasks. rasci.py wsjf prints the WSJF table with OKR and the task id; the register's show takes its Prio from that rank (an operator --prio pin first) and adds a WSJF column. Four descriptions were tightened to keep the listing budget.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
rasci.py keeps the organisation's tasks in <council root>/rasci.json: RASCI roles, objective with key results, priority, worker, start, end, workdir, branch, tools and artifacts; done is refused while a key result is open. council.py writes every claim, takeover and release through, and show lists the live claims as one row per task with R, A, S, C and I columns. kvb-org-artefacts manages the artifacts in the same table for kvb-cleanup. The org, plan, triage, tune, cleanup, branch-gc and dream skills now record into and read from the register; five descriptions were tightened to keep the listing budget.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
All 38 first-party skills move to kvb-<category>-<name> with the categories think, create, code, ci, git, check, org and skills; kvb-cleanup and kvb-help are the only uncategorized names. tests/test_naming.py pins the schema, the Agent Skills name rules and the README organigram. enable-plugins.js repoints hook commands of renamed skills in settings.local.json so a stale script path cannot block tools.
BREAKING CHANGE: the old skill names and slash commands (/commit-next, /dream, /autotune, ...) are gone; use their kvb- successors listed in the README organigram.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
autotune, autoskill and confidence each carried their own idea of what an
improvement is, and autotune's two efficiency rules were the same sentence
twice. They now route to one skill that names three axes to move at once —
fewer tokens, less time, higher quality — and the rules that decide when a
proposal counts: at least one axis better, none damaged, quality never the
currency, observed friction instead of a guessed percentage.
Applying its own rule, autotune gets shorter rather than longer. The test
holds the coupling, so a later rewrite cannot silently drop a route.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
