[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6BSQ2D38FA6AJM1YAFXMFQH":3,"skill-same:01M6BSQ2D38FA6AJM1YAFXMFQH":95},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":38,"linkOnly":39,"markdownContent":40,"markdownBody":41,"frontMatter":42,"refName":45,"commitSha":46,"skillDirectory":47,"forksCount":48,"repositoryHtmlUrl":49,"repositoryHomepageUrl":50,"ownerHtmlUrl":51,"ownerBlogUrl":52,"ownerDescription":6,"updatedAt":53,"install":54,"resources":66},"01M6BSQ2D38FA6AJM1YAFXMFQH","bilan",null,"Measure whether this session actually finished — the 0-10 completion number, computed from git, the carnet ledger, LIMITATION markers and CI instead of narrated. Use before reporting a completion number, when asked \"where are we\", at the end of a piece of work, and to find what a session that died left behind.","Atmosphere\u002Fatmosphere","Atmosphere","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F318873?v=4","01M6BSPYD9QQA0JP3Z913W9PRV","Atmosphere Framework","01M6BSPYHYTTF1B7JPD17B8N5S",3834,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002FSKILL.md",[],[5,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37],"use","when","where are we","at","true","number","ledger","markers","scope","end","git","session","find","work","sweep","json","carnet","reporting","ci","2026-11-10 10:12:30",false,"---\nname: bilan\ndescription: Measure whether this session actually finished — the 0-10 completion number, computed from git, the carnet ledger, LIMITATION markers and CI instead of narrated. Use before reporting a completion number, when asked \"where are we\", at the end of a piece of work, and to find what a session that died left behind.\nargument-hint: \"[sweep] [scope] [--cheap] [--json]\"\nuser-invocable: true\n---\n\n# Bilan\n\n**The completion number is this script's number, never yours.**\n\nIt used to be narrated: it came from the session's account of itself, so a session that\n*believed* it was done reported 8 or 10 while it still held open carnet issues, had commits\nsitting unpushed, or had background work still running. Every one of those facts is\nmachine-checkable, and `bilan.sh` checks them.\n\n```bash\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh            # full: local facts, one fetch, one CI query\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh --cheap    # local only, no network\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh sweep      # what a session that DIED left behind\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh scope      # the files the limitation register scans, as bilan resolves them\n```\n\nExit code: `0` = 10\u002F10 · `1` = incomplete · `2` = the script itself failed.\n\n**`--cheap` is the status-line run**, and it gives the full run's number wherever local evidence\ncan settle the fact. Where only the network can, the cheap run states what it cannot know instead\nof guessing: commits that look unpushed against a fetch more than five minutes old (it caps at 9,\n\"may already be on the remote\", where the full run fetches first), a filed issue someone else may\nhave closed, or a `limitation` label applied outside `carnet.sh` (cheap keeps the cap on both), and\nwhether the issue a new marker names is live (cheap does not judge it). CI is printed, never\nscored, so skipping it changes what is shown, not the score.\n\nEvery scoring run also writes `score\u003CTAB>epoch\u003CTAB>top cap` to\n`${CLAUDE_CONFIG_DIR:-~\u002F.claude}\u002Fbilan\u002F\u003Csession-id>.score`, so a status-line command can show the\nnumber without running anything: it only has to refresh the file with `bilan.sh --cheap --json`\nwhen it ages.\n\n## The rules\n\n1. **Run it before you give a number.** Report the score it prints. If you believe a cap is\n   wrong, say so in words *and still report the script's number* — arguing with the measurement\n   is a conversation, overriding it silently is the failure this exists to stop.\n2. **A cap is a fact, not an opinion.** Each one prints its own evidence and its own remedy.\n   The score is `min()` over the caps, so one open carnet issue holds the whole session at 6 no\n   matter how much else landed.\n3. **Failures never deduct.** A red that is now green, a mistake found and fixed, a rough path —\n   none of it lowers the number. The score measures *completion*, and only completion.\n4. **Friction is printed, never scored.** Tool errors, interrupts and denials are counted from\n   the transcript and shown for context. They do not cap anything.\n\n## What caps the score\n\n| Evidence | Caps at |\n|---|---|\n| carnet issue claimed by this session, neither closed nor released | **6** |\n| carnet issue **filed** by this session and still open (unless a registered limitation) — or, in the full run, whose state the tracker would not give | **6** |\n| `LIMITATION(registre#…)` marker added in a file the register scans, naming no live issue — missing, closed, a pull request, or not labelled `limitation` (full run only), or no issue number at all | **6** |\n| `LIMITATION(` marker added in a file with an extension the register scans, while the register cannot say what it scans (`.registre` not checked out) | **6** |\n| background task still running | **7** |\n| tracked files modified and uncommitted | **7** |\n| todo still `pending` or `in_progress` | **7** |\n| commits not pushed — to the upstream, or, on a branch without one (or whose upstream was deleted), not on `origin\u002Fmain` | **8** |\n| nothing measurable — no commit, no todo | **9** |\n| untracked files, a stash made on this branch during this session, a local branch whose upstream is gone — this checkout's own, or one no worktree has checked out | **9** |\n| commits to push but the validation marker is missing, older than the pre-push hook's TTL, or for another sha | **9** |\n| commits that look unpushed against a stale fetch (`--cheap` only) | **9** |\n| dev stack this session started still up — only in a checkout that ships `bin\u002Fdev-processes.sh`; this repo has none, so here it never fires | **9** |\n\n## This repo's shape\n\n**Worktrees and branches without an upstream.** Work here commonly happens in\n`.claude\u002Fworktrees\u002F*`, on branches that have no upstream at all: a branch is typically landed on\nmain with `git push origin \u003Cbranch>:main` rather than published. So a branch without an upstream\nis measured against `origin\u002Fmain` — its commits that are not there are unpushed work, and the\nremedy says how this repo lands it. Measured against an upstream it does not have, a worktree's\nwork would never be reported at all.\n\n**The validation marker** is `scripts\u002Fpre-push-validate.sh`'s: one line, `\u003Cepoch> \u003Csha>`, in the\ncheckout's own git dir. In a worktree that is `.git\u002Fworktrees\u002F\u003Cname>\u002Fvalidation-passed`, so a\nmarker stamped from the main checkout never validates a worktree's HEAD — the pre-push hook would\nrefuse that push, and bilan says so first. The TTL is read from `.githooks\u002Fpre-push` rather than\nrestated, so the two cannot disagree about when a marker has expired.\n\n**The stash is one stack.** Every worktree pushes onto the same `refs\u002Fstash`, so a stash made on\nanother branch was made in another checkout: it is stated and never counted. For this session's\nown stash the remedy is to apply or drop it by its exact ref — a bare `git stash pop` in a shared\nstack can take a peer's work.\n\n**The ledger directory is machine-wide.** `${CLAUDE_CONFIG_DIR:-~\u002F.claude}\u002Fcarnet-claims\u002F` holds\nthe ledgers of every repo whose sessions use carnet, and each line names the register its issue\nlives in. An issue is a (number, tracker) pair: a question about it goes to its own tracker, and\nan issue on another repo's register is named as `owner\u002Frepo#n`, never as this repo's `carnet#n`.\nEvery question goes over REST (`gh api`), never `gh issue`: that is GraphQL, which a cloud\nsession's proxy refuses, and a refusal read as an answer kept a closed issue's cap. A tracker that\ndoes not answer at all keeps the cap and says so, rather than reporting the issue as open.\n\n## Uncommitted files that are not this session's\n\nSeveral sessions can share one checkout. A peer's mid-edit file used to hold every other session\nat 7 — and those sessions were right to refuse to touch it, so they could not reach 10 no matter\nwhat they did. That was a category error: the score measures *this session's* completion, and a\npeer's in-flight file is not this session's incompleteness.\n\nTwo mechanisms fix it, and the first needs nothing from you.\n\n**The baseline.** The SessionStart hook records which files were already dirty or untracked — every\nuntracked file by name — and which commits were already unpushed, when the session opened. A path\ndirty before the session existed is definitionally not its work — that much *is* machine-decidable.\nThose are stated as a note and never scored. The hook records it once per session: SessionStart\nfires again on resume and after a compaction, and re-recording there would declare the session's\nown uncommitted work inherited. Its paths and its HEAD describe the checkout the session opened in\nand no other. In a worktree the session reaches afterwards no path is inherited — a path dirty in\nthe main checkout says nothing about the same-named file there, and a peer's pre-existing worktree\nis what `ack` is for — and whether the session committed anything there is read from that\ncheckout's own HEAD reflog: where its HEAD stood when the session opened, or the commit it was made\nat. The unpushed commits it recorded are inherited everywhere: a sha names the same commit in every\ncheckout, so a peer's commit already unpushed when the session opened is not the session's to land\nfrom a worktree made on top of it either.\n\n**`ack`, for what goes dirty afterwards.** A peer editing during your session is not covered by\nthe baseline, so you say so once:\n\n```bash\nbilan.sh ack --why \"a peer's version bump, written into this shared checkout at 10:16\"\n```\n\nThat **clears** the cap rather than softening it — the tracked-files cap and the untracked-files\ncap alike — and carries the reason into every later report. It is keyed to the exact set of paths\neach cap names — what the baseline already calls inherited is not in it — so dirtying one more\nfile brings that cap straight back, and ownership is a property of the files rather than their\ncontents, so a peer changing those same files again stays covered. Neither cap's remedy ever says\nto delete a file: the ones it names might be a peer's.\n\n**Why not attribute automatically?** It was tried and it does not work. Claude Code records the\npaths a session touched under `file-history-snapshot.trackedFileBackups`, but only for the\nEdit\u002FWrite tools, so a session that writes its files through Bash leaves that map empty.\nAttributing on it would call a session's own work a peer's and stop blocking — the worst\ndirection to be wrong in.\n\nThe dev-stack cap follows the same rule where a checkout has one: a stack already running under\nthe same pid when the session opened is a peer's, stated and never scored, because the only way\nto clear such a cap would be to take the peer's servers down.\n\n## The three call sites\n\n**`\u002Fbilan`** — on demand, the full measurement including CI.\n\n**The startup sweep** (`hooks\u002Fsession-start-sweep.sh`) — records the baseline, then looks for\nwhat a dead session left behind. A `kill -9`, a closed terminal or an exhausted context fires no\nexit hook, so the dead session can never report on itself; the session that opens after it looks\ninstead, across every worktree of this repo and every ledger on the machine. It reports only this\nrepo's register, and drops a dead session's claim once the tracker says the issue is closed, so a\nlong-resolved issue does not keep reappearing as abandoned.\n\nA ledger belongs to its session, so the sweep rewrites or deletes one only when that session has\ncertainly ended. A session is judged by its id, not by the pid its ledger recorded — carnet writes\nthat pid once, and `claude --resume` keeps the id under a new one — from the session files Claude\nCode keeps under its config dir (`$CLAUDE_CONFIG_DIR` and every `~\u002F.claude*`): a file naming the id\nwith a live pid is a running session, and the sweep leaves it alone. When the sweep cannot tell —\na session file still names the session under a pid that does not run here (a leftover of a crash,\nor a session in another pid namespace sharing the config dir, such as a container), the ledger's\nrecorded pid still runs with no session file to say whose it is, or the ledger names no pid — it\nreports the claims as possibly still held, and clears nothing. The host name carnet stamps into the\nledger decides nothing: the ledger sits in this machine's config dir, and a machine's name changes\nunder it — a Mac renames itself when its name is taken on the network — so a session this machine\nran under an earlier name is still judged, and cleared, as this machine's.\n\nThe ledger directory is shared, and not every sweep that reads it keeps to its own register. This\none asks each claim's own tracker and leaves another register's lines alone. The sweeps of\ndravr-platform and mirroir-mcp, as they stand, do not: each asks its own tracker about every claim\nnumber in a ledger whose session it takes for ended, whatever register the line names. A claim held\nhere can then be reported there as that repo's own `carnet#n`, or be deleted — and the ledger with\nit, once nothing is left — because that repo's issue of the same number is closed; if that sweep\nruns first, this one then finds nothing to report. The claim marker on the tracker is untouched:\n`carnet.sh status \u003Cn>` still shows who holds the issue and whether that session has ended.\n\n**The Stop gate** (`hooks\u002Fstop-gate.sh`) — **disarmed, and deliberately not wired.** It would\nrefuse a stop while the score was 8 or below. That is only as good as the number: a gate grading\nsomething the session did not do — a shared checkout's HEAD, a peer's red CI — blocks every\nsession in that checkout over a commit none of them made, and each block spends a session's last\nturn arguing with a number about somebody else's work. A wrong number costs a paragraph; a wrong\nnumber that can *stop the session* costs every turn it had left, and fixing one misattribution\ndoes not remove the class. The script is kept, and its tests run, so re-arming it would be a\ndecision about a script that works.\n\n## Work that leaves no trace\n\nTwo kinds of incompleteness are invisible to git, the ledger and CI, and both have produced a\nfalse 10:\n\n**Background tasks.** Claude Code writes each one's stream to the session's `tasks\u002F` directory,\nbeside its scratchpad (`…\u002F\u003Csession-id>\u002Ftasks\u002F\u003Cid>.output`), and closes it with\n`[exited with code N]` or `[killed]`; no marker and a live holder means it never ended. The\ndirectory is per session, so each terminal answers only for its own work. A session working in a\nworktree has it keyed by the directory the session was launched from — usually the main\ncheckout — and bilan looks there too. Closing a session with a task running throws that work\naway, so it caps at **7**. bilan runs *inside* one of those streams itself, so a file held by\nanything in its own process ancestry is this invocation, not a task.\n\n**Issues the session filed.** `carnet.sh create` writes a `filed` line to the ledger and `close`\nremoves it, so what remains is what this session opened and did not fix. That caps at **6** —\nlevel with an issue still held, because filing instead of fixing is the same unfinished work\nwearing a label. The standing rule is *fix first, file only the residue*; if something genuinely\ncannot be fixed here, that is a decision to put in front of the project maintainer, not a cap to\nslip past.\n\n**Except a registered limitation, which is the one filed issue that is not work owed.** The\nLIMITATION procedure *requires* an open issue for as long as a marker names it, so this cap would\npunish a session for obeying it — and there would be nothing it could do, because the fix is to\nclose an issue the rules say must stay open.\n\nSo the exemption needs **both** halves: the issue carries the `limitation` label, *and* a\n`LIMITATION(registre#n)` marker names that issue **in a file the register scans**. A label alone\nstill caps, so a bug cannot be relabelled out of the score; a marker naming a dead issue is caught\nby the marker check from the other side. Both together mean the issue is a register entry rather\nthan deferred work, and it prints as a **note** — the gap stays visible instead of disappearing\ninto a clean pass, which is the entire point of registering it. The exemption only speaks for this\nrepo's register: the marker half is this checkout's scope.\n\n**Both halves are local, so `--cheap` credits it too.** The full run reads the label from the\ntracker, falling back to the ledger only when the tracker cannot be asked at all. `--cheap` reads\nthe line `carnet.sh` writes to the session ledger when it applies the label — `create --label\nlimitation` or `label \u003Cn> +limitation` — and removes on `-limitation`. A label applied any other\nway (the web UI) reaches only the full run.\n\n## Where a marker counts\n\n**It is the register's decision, not bilan's.** The register is llm-registre, a git submodule at\n`.registre`; its gate scans the directories its callers name, for the extensions `registre.toml`\nconfigures, minus test, bench, example and generated trees. bilan asks the gate for that set\ninstead of keeping its own copy of the exclusions: a copy drifts from the gate both ways, and a\nmarker one tool honoured would be invisible to the other.\n\n- **The files.** A gate new enough to have `--list-files` is asked for them. An older gate is\n  probed: it runs once with a recording ripgrep first on `PATH`, and bilan lists files with exactly\n  the directories and globs the gate composed — which is what `--list-files` itself runs. Either\n  way nothing is copied. `bilan.sh scope` prints the answer and how it was reached.\n- **The directories.** The ones the CI lane passes the gate\n  (`.github\u002Fworkflows\u002Flimitation-register.yml`, the only workflow read), read out of the workflow\n  rather than restated. `scan_dirs` in `registre.toml`, and `REGISTRE_SCAN_DIRS` over it, replace\n  them for bilan alone: the gate this repo pins reads neither, and both of its callers name the\n  directories on the command line, so either setting narrows what bilan verifies and never what the\n  gate scans. The suite checks that bilan resolves the same directories\n  `scripts\u002Fpre-push-validate.sh` runs the gate on, so a committed `scan_dirs` that disagrees fails a\n  test; the environment variable is a local override no test sees.\n- **Test trees are outside it.** A gap in test *coverage* is therefore marked on the production\n  item the tests leave uncovered, not on the test that fails to cover it.\n- **Only what the session added.** Markers are read from the session's diff against the merge\n  base, committed and not, so a checkout that is merely behind is neither credited with nor blamed\n  for a line it has not merged.\n- **When the register cannot say** — `.registre` is not checked out, which is the usual state of a\n  fresh worktree (`git submodule update --init .registre`), or the gate listed nothing — a marker\n  this session wrote cannot be verified, and it caps at 6. Only a marker in a file whose extension\n  the register scans: prose, scripts and this skill's own tree are outside every register's scope,\n  so a session that merely documents the convention is not held to it. Once the submodule is\n  populated in a worktree, a plain `git worktree remove` refuses it (git will not remove a\n  worktree containing submodules): check it holds nothing uncommitted, then `--force`.\n- **The tracker is private.** A token that cannot read it leaves a new marker unverified; the cap\n  then says the tracker was unreadable (`gh auth status`) rather than blaming the marker.\n\n## When there is nothing to measure\n\nA score of 10 means *everything I checked is done*. When nothing was checkable, 10 means nothing\nat all — and that is how a session reported 10\u002F10 with its artifact unwritten. bilan reads the\nrepo, the register and CI; a session whose work is research, a design, a document or a published\nartifact touches none of the three, so every check came back clean because every check came back\nempty.\n\nSo a session that made no commit and declared no todo caps at **9**, labelled *nothing\nmeasurable*. It caps rather than blocks, because answering a question really is a complete\nsession; what it must not do is issue a verdict it never earned. Declaring the work as a todo,\nor committing something, makes it measurable again.\n\nThe verdict is measured against an ask, so it waits for one. A status line renders before the\nfirst prompt arrives, and without that condition every session would open at \"9\u002F10 nothing\nmeasurable\" for having done nothing in its first second. A session nobody has asked anything of\nscores clean.\n\nEvery report also carries the **ask**, verbatim: `asked:` is the latest prompt the project\nmaintainer typed, and `opened:` the first, when they differ. bilan cannot judge whether the work\nsatisfies it — that would be narration again, the thing it exists to treat — but it can refuse to\nlet a session claim completion without the request in view. The latest counts because a long\nsession is not what it opened with. A scheduled wakeup, a task notification or a peer message is\ntext the session or the harness wrote, not an ask, and the transcript marks it so.\n\n## CI, and one thing it cannot see\n\n**CI is printed, never scored.** A shared checkout has one HEAD and many sessions, all committing\nas the same author, so whose commit the tip is cannot be recovered from git. Grading the tip caps\nevery session in the checkout over one peer's red; grading it only when HEAD moved since the\nsession opened is no better, because HEAD moves when peers push, and a session is then graded on\na tip that landed after its own push. The verdict is shown; the number stays about work the\nsession can act on. For CI on your own commit, ask for that sha by name.\n\nIt is read from the check-runs API by full sha, every page: `gh run list --commit` with a short sha\nanswers with an empty list, and a failure on the second page of a commit's check runs is still a\nfailure. A sha with no check run prints nothing — absent, not green. It is asked whether or not the\nbranch has an upstream: a worktree branch landed with `git push origin \u003Cbranch>:main` has main's\ntip as its HEAD, and that commit's CI is shown like any other.\n\n**Background work is scanned in the session's `tasks\u002F` directory only.** Whether every kind of\nin-flight work can be detected there — a Monitor, a subagent, whose entry is a link to its\ntranscript and never carries an exit marker — has not been confirmed. Treat the running-work cap\nas covering background Bash tasks; a session that knows it has other work in flight should say\nso rather than trust the absence of a cap.\n\n## What it does not do\n\nIt cannot tell you whether the work is *good*, only whether it is *finished*. A green bilan on\na wrong implementation is still a wrong implementation — that is what `\u002Fcode-review` is for.\n\nIt cannot see a deliverable the session never declared. A todo, a commit or an issue makes work\nvisible to it; an artifact written and published with none of those does not. The *nothing\nmeasurable* cap is how it says so out loud instead of scoring an empty check clean.\n","# Bilan\n\n**The completion number is this script's number, never yours.**\n\nIt used to be narrated: it came from the session's account of itself, so a session that\n*believed* it was done reported 8 or 10 while it still held open carnet issues, had commits\nsitting unpushed, or had background work still running. Every one of those facts is\nmachine-checkable, and `bilan.sh` checks them.\n\n```bash\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh            # full: local facts, one fetch, one CI query\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh --cheap    # local only, no network\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh sweep      # what a session that DIED left behind\n.agents\u002Fskills\u002Fbilan\u002Fbilan.sh scope      # the files the limitation register scans, as bilan resolves them\n```\n\nExit code: `0` = 10\u002F10 · `1` = incomplete · `2` = the script itself failed.\n\n**`--cheap` is the status-line run**, and it gives the full run's number wherever local evidence\ncan settle the fact. Where only the network can, the cheap run states what it cannot know instead\nof guessing: commits that look unpushed against a fetch more than five minutes old (it caps at 9,\n\"may already be on the remote\", where the full run fetches first), a filed issue someone else may\nhave closed, or a `limitation` label applied outside `carnet.sh` (cheap keeps the cap on both), and\nwhether the issue a new marker names is live (cheap does not judge it). CI is printed, never\nscored, so skipping it changes what is shown, not the score.\n\nEvery scoring run also writes `score\u003CTAB>epoch\u003CTAB>top cap` to\n`${CLAUDE_CONFIG_DIR:-~\u002F.claude}\u002Fbilan\u002F\u003Csession-id>.score`, so a status-line command can show the\nnumber without running anything: it only has to refresh the file with `bilan.sh --cheap --json`\nwhen it ages.\n\n## The rules\n\n1. **Run it before you give a number.** Report the score it prints. If you believe a cap is\n   wrong, say so in words *and still report the script's number* — arguing with the measurement\n   is a conversation, overriding it silently is the failure this exists to stop.\n2. **A cap is a fact, not an opinion.** Each one prints its own evidence and its own remedy.\n   The score is `min()` over the caps, so one open carnet issue holds the whole session at 6 no\n   matter how much else landed.\n3. **Failures never deduct.** A red that is now green, a mistake found and fixed, a rough path —\n   none of it lowers the number. The score measures *completion*, and only completion.\n4. **Friction is printed, never scored.** Tool errors, interrupts and denials are counted from\n   the transcript and shown for context. They do not cap anything.\n\n## What caps the score\n\n| Evidence | Caps at |\n|---|---|\n| carnet issue claimed by this session, neither closed nor released | **6** |\n| carnet issue **filed** by this session and still open (unless a registered limitation) — or, in the full run, whose state the tracker would not give | **6** |\n| `LIMITATION(registre#…)` marker added in a file the register scans, naming no live issue — missing, closed, a pull request, or not labelled `limitation` (full run only), or no issue number at all | **6** |\n| `LIMITATION(` marker added in a file with an extension the register scans, while the register cannot say what it scans (`.registre` not checked out) | **6** |\n| background task still running | **7** |\n| tracked files modified and uncommitted | **7** |\n| todo still `pending` or `in_progress` | **7** |\n| commits not pushed — to the upstream, or, on a branch without one (or whose upstream was deleted), not on `origin\u002Fmain` | **8** |\n| nothing measurable — no commit, no todo | **9** |\n| untracked files, a stash made on this branch during this session, a local branch whose upstream is gone — this checkout's own, or one no worktree has checked out | **9** |\n| commits to push but the validation marker is missing, older than the pre-push hook's TTL, or for another sha | **9** |\n| commits that look unpushed against a stale fetch (`--cheap` only) | **9** |\n| dev stack this session started still up — only in a checkout that ships `bin\u002Fdev-processes.sh`; this repo has none, so here it never fires | **9** |\n\n## This repo's shape\n\n**Worktrees and branches without an upstream.** Work here commonly happens in\n`.claude\u002Fworktrees\u002F*`, on branches that have no upstream at all: a branch is typically landed on\nmain with `git push origin \u003Cbranch>:main` rather than published. So a branch without an upstream\nis measured against `origin\u002Fmain` — its commits that are not there are unpushed work, and the\nremedy says how this repo lands it. Measured against an upstream it does not have, a worktree's\nwork would never be reported at all.\n\n**The validation marker** is `scripts\u002Fpre-push-validate.sh`'s: one line, `\u003Cepoch> \u003Csha>`, in the\ncheckout's own git dir. In a worktree that is `.git\u002Fworktrees\u002F\u003Cname>\u002Fvalidation-passed`, so a\nmarker stamped from the main checkout never validates a worktree's HEAD — the pre-push hook would\nrefuse that push, and bilan says so first. The TTL is read from `.githooks\u002Fpre-push` rather than\nrestated, so the two cannot disagree about when a marker has expired.\n\n**The stash is one stack.** Every worktree pushes onto the same `refs\u002Fstash`, so a stash made on\nanother branch was made in another checkout: it is stated and never counted. For this session's\nown stash the remedy is to apply or drop it by its exact ref — a bare `git stash pop` in a shared\nstack can take a peer's work.\n\n**The ledger directory is machine-wide.** `${CLAUDE_CONFIG_DIR:-~\u002F.claude}\u002Fcarnet-claims\u002F` holds\nthe ledgers of every repo whose sessions use carnet, and each line names the register its issue\nlives in. An issue is a (number, tracker) pair: a question about it goes to its own tracker, and\nan issue on another repo's register is named as `owner\u002Frepo#n`, never as this repo's `carnet#n`.\nEvery question goes over REST (`gh api`), never `gh issue`: that is GraphQL, which a cloud\nsession's proxy refuses, and a refusal read as an answer kept a closed issue's cap. A tracker that\ndoes not answer at all keeps the cap and says so, rather than reporting the issue as open.\n\n## Uncommitted files that are not this session's\n\nSeveral sessions can share one checkout. A peer's mid-edit file used to hold every other session\nat 7 — and those sessions were right to refuse to touch it, so they could not reach 10 no matter\nwhat they did. That was a category error: the score measures *this session's* completion, and a\npeer's in-flight file is not this session's incompleteness.\n\nTwo mechanisms fix it, and the first needs nothing from you.\n\n**The baseline.** The SessionStart hook records which files were already dirty or untracked — every\nuntracked file by name — and which commits were already unpushed, when the session opened. A path\ndirty before the session existed is definitionally not its work — that much *is* machine-decidable.\nThose are stated as a note and never scored. The hook records it once per session: SessionStart\nfires again on resume and after a compaction, and re-recording there would declare the session's\nown uncommitted work inherited. Its paths and its HEAD describe the checkout the session opened in\nand no other. In a worktree the session reaches afterwards no path is inherited — a path dirty in\nthe main checkout says nothing about the same-named file there, and a peer's pre-existing worktree\nis what `ack` is for — and whether the session committed anything there is read from that\ncheckout's own HEAD reflog: where its HEAD stood when the session opened, or the commit it was made\nat. The unpushed commits it recorded are inherited everywhere: a sha names the same commit in every\ncheckout, so a peer's commit already unpushed when the session opened is not the session's to land\nfrom a worktree made on top of it either.\n\n**`ack`, for what goes dirty afterwards.** A peer editing during your session is not covered by\nthe baseline, so you say so once:\n\n```bash\nbilan.sh ack --why \"a peer's version bump, written into this shared checkout at 10:16\"\n```\n\nThat **clears** the cap rather than softening it — the tracked-files cap and the untracked-files\ncap alike — and carries the reason into every later report. It is keyed to the exact set of paths\neach cap names — what the baseline already calls inherited is not in it — so dirtying one more\nfile brings that cap straight back, and ownership is a property of the files rather than their\ncontents, so a peer changing those same files again stays covered. Neither cap's remedy ever says\nto delete a file: the ones it names might be a peer's.\n\n**Why not attribute automatically?** It was tried and it does not work. Claude Code records the\npaths a session touched under `file-history-snapshot.trackedFileBackups`, but only for the\nEdit\u002FWrite tools, so a session that writes its files through Bash leaves that map empty.\nAttributing on it would call a session's own work a peer's and stop blocking — the worst\ndirection to be wrong in.\n\nThe dev-stack cap follows the same rule where a checkout has one: a stack already running under\nthe same pid when the session opened is a peer's, stated and never scored, because the only way\nto clear such a cap would be to take the peer's servers down.\n\n## The three call sites\n\n**`\u002Fbilan`** — on demand, the full measurement including CI.\n\n**The startup sweep** (`hooks\u002Fsession-start-sweep.sh`) — records the baseline, then looks for\nwhat a dead session left behind. A `kill -9`, a closed terminal or an exhausted context fires no\nexit hook, so the dead session can never report on itself; the session that opens after it looks\ninstead, across every worktree of this repo and every ledger on the machine. It reports only this\nrepo's register, and drops a dead session's claim once the tracker says the issue is closed, so a\nlong-resolved issue does not keep reappearing as abandoned.\n\nA ledger belongs to its session, so the sweep rewrites or deletes one only when that session has\ncertainly ended. A session is judged by its id, not by the pid its ledger recorded — carnet writes\nthat pid once, and `claude --resume` keeps the id under a new one — from the session files Claude\nCode keeps under its config dir (`$CLAUDE_CONFIG_DIR` and every `~\u002F.claude*`): a file naming the id\nwith a live pid is a running session, and the sweep leaves it alone. When the sweep cannot tell —\na session file still names the session under a pid that does not run here (a leftover of a crash,\nor a session in another pid namespace sharing the config dir, such as a container), the ledger's\nrecorded pid still runs with no session file to say whose it is, or the ledger names no pid — it\nreports the claims as possibly still held, and clears nothing. The host name carnet stamps into the\nledger decides nothing: the ledger sits in this machine's config dir, and a machine's name changes\nunder it — a Mac renames itself when its name is taken on the network — so a session this machine\nran under an earlier name is still judged, and cleared, as this machine's.\n\nThe ledger directory is shared, and not every sweep that reads it keeps to its own register. This\none asks each claim's own tracker and leaves another register's lines alone. The sweeps of\ndravr-platform and mirroir-mcp, as they stand, do not: each asks its own tracker about every claim\nnumber in a ledger whose session it takes for ended, whatever register the line names. A claim held\nhere can then be reported there as that repo's own `carnet#n`, or be deleted — and the ledger with\nit, once nothing is left — because that repo's issue of the same number is closed; if that sweep\nruns first, this one then finds nothing to report. The claim marker on the tracker is untouched:\n`carnet.sh status \u003Cn>` still shows who holds the issue and whether that session has ended.\n\n**The Stop gate** (`hooks\u002Fstop-gate.sh`) — **disarmed, and deliberately not wired.** It would\nrefuse a stop while the score was 8 or below. That is only as good as the number: a gate grading\nsomething the session did not do — a shared checkout's HEAD, a peer's red CI — blocks every\nsession in that checkout over a commit none of them made, and each block spends a session's last\nturn arguing with a number about somebody else's work. A wrong number costs a paragraph; a wrong\nnumber that can *stop the session* costs every turn it had left, and fixing one misattribution\ndoes not remove the class. The script is kept, and its tests run, so re-arming it would be a\ndecision about a script that works.\n\n## Work that leaves no trace\n\nTwo kinds of incompleteness are invisible to git, the ledger and CI, and both have produced a\nfalse 10:\n\n**Background tasks.** Claude Code writes each one's stream to the session's `tasks\u002F` directory,\nbeside its scratchpad (`…\u002F\u003Csession-id>\u002Ftasks\u002F\u003Cid>.output`), and closes it with\n`[exited with code N]` or `[killed]`; no marker and a live holder means it never ended. The\ndirectory is per session, so each terminal answers only for its own work. A session working in a\nworktree has it keyed by the directory the session was launched from — usually the main\ncheckout — and bilan looks there too. Closing a session with a task running throws that work\naway, so it caps at **7**. bilan runs *inside* one of those streams itself, so a file held by\nanything in its own process ancestry is this invocation, not a task.\n\n**Issues the session filed.** `carnet.sh create` writes a `filed` line to the ledger and `close`\nremoves it, so what remains is what this session opened and did not fix. That caps at **6** —\nlevel with an issue still held, because filing instead of fixing is the same unfinished work\nwearing a label. The standing rule is *fix first, file only the residue*; if something genuinely\ncannot be fixed here, that is a decision to put in front of the project maintainer, not a cap to\nslip past.\n\n**Except a registered limitation, which is the one filed issue that is not work owed.** The\nLIMITATION procedure *requires* an open issue for as long as a marker names it, so this cap would\npunish a session for obeying it — and there would be nothing it could do, because the fix is to\nclose an issue the rules say must stay open.\n\nSo the exemption needs **both** halves: the issue carries the `limitation` label, *and* a\n`LIMITATION(registre#n)` marker names that issue **in a file the register scans**. A label alone\nstill caps, so a bug cannot be relabelled out of the score; a marker naming a dead issue is caught\nby the marker check from the other side. Both together mean the issue is a register entry rather\nthan deferred work, and it prints as a **note** — the gap stays visible instead of disappearing\ninto a clean pass, which is the entire point of registering it. The exemption only speaks for this\nrepo's register: the marker half is this checkout's scope.\n\n**Both halves are local, so `--cheap` credits it too.** The full run reads the label from the\ntracker, falling back to the ledger only when the tracker cannot be asked at all. `--cheap` reads\nthe line `carnet.sh` writes to the session ledger when it applies the label — `create --label\nlimitation` or `label \u003Cn> +limitation` — and removes on `-limitation`. A label applied any other\nway (the web UI) reaches only the full run.\n\n## Where a marker counts\n\n**It is the register's decision, not bilan's.** The register is llm-registre, a git submodule at\n`.registre`; its gate scans the directories its callers name, for the extensions `registre.toml`\nconfigures, minus test, bench, example and generated trees. bilan asks the gate for that set\ninstead of keeping its own copy of the exclusions: a copy drifts from the gate both ways, and a\nmarker one tool honoured would be invisible to the other.\n\n- **The files.** A gate new enough to have `--list-files` is asked for them. An older gate is\n  probed: it runs once with a recording ripgrep first on `PATH`, and bilan lists files with exactly\n  the directories and globs the gate composed — which is what `--list-files` itself runs. Either\n  way nothing is copied. `bilan.sh scope` prints the answer and how it was reached.\n- **The directories.** The ones the CI lane passes the gate\n  (`.github\u002Fworkflows\u002Flimitation-register.yml`, the only workflow read), read out of the workflow\n  rather than restated. `scan_dirs` in `registre.toml`, and `REGISTRE_SCAN_DIRS` over it, replace\n  them for bilan alone: the gate this repo pins reads neither, and both of its callers name the\n  directories on the command line, so either setting narrows what bilan verifies and never what the\n  gate scans. The suite checks that bilan resolves the same directories\n  `scripts\u002Fpre-push-validate.sh` runs the gate on, so a committed `scan_dirs` that disagrees fails a\n  test; the environment variable is a local override no test sees.\n- **Test trees are outside it.** A gap in test *coverage* is therefore marked on the production\n  item the tests leave uncovered, not on the test that fails to cover it.\n- **Only what the session added.** Markers are read from the session's diff against the merge\n  base, committed and not, so a checkout that is merely behind is neither credited with nor blamed\n  for a line it has not merged.\n- **When the register cannot say** — `.registre` is not checked out, which is the usual state of a\n  fresh worktree (`git submodule update --init .registre`), or the gate listed nothing — a marker\n  this session wrote cannot be verified, and it caps at 6. Only a marker in a file whose extension\n  the register scans: prose, scripts and this skill's own tree are outside every register's scope,\n  so a session that merely documents the convention is not held to it. Once the submodule is\n  populated in a worktree, a plain `git worktree remove` refuses it (git will not remove a\n  worktree containing submodules): check it holds nothing uncommitted, then `--force`.\n- **The tracker is private.** A token that cannot read it leaves a new marker unverified; the cap\n  then says the tracker was unreadable (`gh auth status`) rather than blaming the marker.\n\n## When there is nothing to measure\n\nA score of 10 means *everything I checked is done*. When nothing was checkable, 10 means nothing\nat all — and that is how a session reported 10\u002F10 with its artifact unwritten. bilan reads the\nrepo, the register and CI; a session whose work is research, a design, a document or a published\nartifact touches none of the three, so every check came back clean because every check came back\nempty.\n\nSo a session that made no commit and declared no todo caps at **9**, labelled *nothing\nmeasurable*. It caps rather than blocks, because answering a question really is a complete\nsession; what it must not do is issue a verdict it never earned. Declaring the work as a todo,\nor committing something, makes it measurable again.\n\nThe verdict is measured against an ask, so it waits for one. A status line renders before the\nfirst prompt arrives, and without that condition every session would open at \"9\u002F10 nothing\nmeasurable\" for having done nothing in its first second. A session nobody has asked anything of\nscores clean.\n\nEvery report also carries the **ask**, verbatim: `asked:` is the latest prompt the project\nmaintainer typed, and `opened:` the first, when they differ. bilan cannot judge whether the work\nsatisfies it — that would be narration again, the thing it exists to treat — but it can refuse to\nlet a session claim completion without the request in view. The latest counts because a long\nsession is not what it opened with. A scheduled wakeup, a task notification or a peer message is\ntext the session or the harness wrote, not an ask, and the transcript marks it so.\n\n## CI, and one thing it cannot see\n\n**CI is printed, never scored.** A shared checkout has one HEAD and many sessions, all committing\nas the same author, so whose commit the tip is cannot be recovered from git. Grading the tip caps\nevery session in the checkout over one peer's red; grading it only when HEAD moved since the\nsession opened is no better, because HEAD moves when peers push, and a session is then graded on\na tip that landed after its own push. The verdict is shown; the number stays about work the\nsession can act on. For CI on your own commit, ask for that sha by name.\n\nIt is read from the check-runs API by full sha, every page: `gh run list --commit` with a short sha\nanswers with an empty list, and a failure on the second page of a commit's check runs is still a\nfailure. A sha with no check run prints nothing — absent, not green. It is asked whether or not the\nbranch has an upstream: a worktree branch landed with `git push origin \u003Cbranch>:main` has main's\ntip as its HEAD, and that commit's CI is shown like any other.\n\n**Background work is scanned in the session's `tasks\u002F` directory only.** Whether every kind of\nin-flight work can be detected there — a Monitor, a subagent, whose entry is a link to its\ntranscript and never carries an exit marker — has not been confirmed. Treat the running-work cap\nas covering background Bash tasks; a session that knows it has other work in flight should say\nso rather than trust the absence of a cap.\n\n## What it does not do\n\nIt cannot tell you whether the work is *good*, only whether it is *finished*. A green bilan on\na wrong implementation is still a wrong implementation — that is what `\u002Fcode-review` is for.\n\nIt cannot see a deliverable the session never declared. A todo, a commit or an issue makes work\nvisible to it; an artifact written and published with none of those does not. The *nothing\nmeasurable* cap is how it says so out loud instead of scoring an empty check clean.\n",{"name":5,"description":7,"argument-hint":43,"user-invocable":44},"[sweep] [scope] [--cheap] [--json]",true,"main","4f29b171768db38405b89eaee6b6f7c70c6822e4",".agents\u002Fskills\u002Fbilan",765,"https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere","https:\u002F\u002Fasync-io.live","https:\u002F\u002Fgithub.com\u002FAtmosphere","http:\u002F\u002Fasync-io.org","2026-11-15 01:07:57",{"source":8,"skillName":5,"commands":55,"prompt":65,"zipAvailable":44},[56,59,62],{"tool":57,"command":58},"npx","npx skills add Atmosphere\u002Fatmosphere --skill bilan",{"tool":60,"command":61},"bunx","bunx skills add Atmosphere\u002Fatmosphere --skill bilan",{"tool":63,"command":64},"pnpm","pnpm dlx skills add Atmosphere\u002Fatmosphere --skill bilan","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002FSKILL.md",[67,73,77,81,87,91],{"path":68,"resourceType":69,"resourceTypeDesc":70,"sizeBytes":71,"sourceHtmlUrl":72},"bilan.sh","SCRIPT","脚本",85685,"https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002Fbilan.sh",{"path":74,"resourceType":69,"resourceTypeDesc":70,"sizeBytes":75,"sourceHtmlUrl":76},"hooks\u002Fsession-start-sweep.sh",2127,"https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002Fhooks\u002Fsession-start-sweep.sh",{"path":78,"resourceType":69,"resourceTypeDesc":70,"sizeBytes":79,"sourceHtmlUrl":80},"hooks\u002Fstop-gate.sh",5556,"https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002Fhooks\u002Fstop-gate.sh",{"path":82,"resourceType":83,"resourceTypeDesc":84,"sizeBytes":85,"sourceHtmlUrl":86},"references\u002Fai-snapshot-workflow.md","REFERENCE","参考文档",6732,"https:\u002F\u002Fgithub.com\u002FChachamaru127\u002Fclaude-code-harness\u002Fblob\u002Fmain\u002Fopencode\u002Fskills\u002Fagent-browser\u002Freferences\u002Fai-snapshot-workflow.md",{"path":88,"resourceType":83,"resourceTypeDesc":84,"sizeBytes":89,"sourceHtmlUrl":90},"references\u002Fbrowser-automation.md",6648,"https:\u002F\u002Fgithub.com\u002FChachamaru127\u002Fclaude-code-harness\u002Fblob\u002Fmain\u002Fopencode\u002Fskills\u002Fagent-browser\u002Freferences\u002Fbrowser-automation.md",{"path":92,"resourceType":69,"resourceTypeDesc":70,"sizeBytes":93,"sourceHtmlUrl":94},"test.sh",82561,"https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fbilan\u002Ftest.sh",[96,109,118,141,164,187],{"skillNo":97,"name":98,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":106,"publishedAt":38},"01M6BSSQ9ZE9KG4GEBK6CXCB37","support-agent","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fsamples\u002Fspring-boot-orchestration-demo\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Fsupport-agent\u002FSKILL.md",[101],{"categoryNo":102,"categoryName":103,"taxonomyType":104,"taxonomyTypeDesc":105},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[107,108],"support agent","support",{"skillNo":110,"name":111,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":115,"publishedAt":38},"01M6BSSMTRCM898EM0E3X68ZDJ","billing-agent","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fsamples\u002Fspring-boot-orchestration-demo\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Fbilling-agent\u002FSKILL.md",[114],{"categoryNo":102,"categoryName":103,"taxonomyType":104,"taxonomyTypeDesc":105},[116,117],"billing agent","billing",{"skillNo":119,"name":120,"nameZh":6,"description":121,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":122,"categories":123,"keywords":125,"publishedAt":38},"01M6BSSJ5B0S8BAQG6MKJAA7YN","omnichannel-chat","General-purpose AI assistant that answers naturally over web chat and external messaging platforms (Telegram, Slack, Discord, WhatsApp, Messenger). One agent, every channel.","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fsamples\u002Fspring-boot-channels-chat\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Fomnichannel-chat\u002FSKILL.md",[124],{"categoryNo":102,"categoryName":103,"taxonomyType":104,"taxonomyTypeDesc":105},[126,127,128,129,130,131,132,133,134,135,136,137,138,139,140],"omnichannel chat","chat","assistant","discord","slack","whatsapp","web","messaging","ai assistant","channel","telegram","general","multi","messenger","multi-channel",{"skillNo":142,"name":143,"nameZh":6,"description":144,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":145,"categories":146,"keywords":148,"publishedAt":38},"01M6BSSFCE0HX9FVQ2PVAQP63T","writer-agent","Business writer that synthesizes research, strategy, and financial findings into executive briefings and reports. Use after upstream specialist agents have produced their data.","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fmodules\u002Fskills\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Fwriter-agent\u002FSKILL.md",[147],{"categoryNo":102,"categoryName":103,"taxonomyType":104,"taxonomyTypeDesc":105},[149,150,151,152,153,154,155,156,157,19,158,159,160,161,162,163],"writer","writer agent","reports","research","business writer","briefing","upstream","Synthesis","data","business","strategy","executive-briefing","briefings","specialist","writing",{"skillNo":165,"name":166,"nameZh":6,"description":167,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":38},"01M6BSSCT499BPVKA5XM9R7NGV","tool-assistant","Tool-calling assistant with backend functions for time, city time, weather, and temperature conversion, plus cost metering and approval workflows. Use when a question needs real-time data via tool calls.","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fmodules\u002Fskills\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Ftool-assistant\u002FSKILL.md",[],[171,172,128,173,174,175,20,176,177,178,179,180,181,182,19,183,184,185,157,186],"tool assistant","tool","metering","weather","functions","for","calls","function-calling","with","time","calling","conversion","question","real-time","Tools","workflows",{"skillNo":188,"name":98,"nameZh":6,"description":189,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":190,"categories":191,"keywords":193,"publishedAt":38},"01M6BSSA9MPT2656NHQG3VJAAJ","Front-line support desk for account questions, technical troubleshooting, and general inquiries. Use as the default customer entry point; routes billing questions to the billing-agent via handoff.","https:\u002F\u002Fgithub.com\u002FAtmosphere\u002Fatmosphere\u002Fblob\u002Fmain\u002Fmodules\u002Fskills\u002Fsrc\u002Fmain\u002Fresources\u002FMETA-INF\u002Fskills\u002Fsupport-agent\u002FSKILL.md",[192],{"categoryNo":102,"categoryName":103,"taxonomyType":104,"taxonomyTypeDesc":105},[108,107,194,195,196,176,197,198,117,199,200,201,137,19,202,203,204,205,206],"as","point","entry","front","questions","via","line","default","technical","troubleshooting","routing","customer","handoff"]