[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M72ADQDB89DNXR7QB6YR5YTC":3,"skill-same:01M72ADQDB89DNXR7QB6YR5YTC":55},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M72ADQDB89DNXR7QB6YR5YTC","xskill-helper",null,"Use this skill for xskill. Trigger when the user wants to init, connect or join a team server, generate or rewrite a team Skill from trajectories, search download upload or import team Skills, look up this user or a teammate's past chats across Claude Code Codex Cursor OpenCode Trae DeepSeek Harness and ngagent, upgrade the client, or debug a stuck daemon. Team chats: xskill traj search then traj read. This machine only: add --local to read ~\u002F.xskill\u002F*_sessions. Invoke as \u002Fxskill-helper.","SkillNerds\u002Fxskill","SkillNerds","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F283534061?v=4","01M72ADC90GC2G52CXN8DKGP5Q","01M72ADC9KGMTQHYDEGS3FG98H",129,"MIT","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fsrc\u002Fxskill\u002Fdata\u002Fskill\u002Fxskill-helper\u002FSKILL.md",[],[],"2026-11-10 18:39:20",false,"---\nname: xskill-helper\ndescription: >-\n  Use this skill for xskill. Trigger when the user wants to init, connect or\n  join a team server, generate or rewrite a team Skill from trajectories,\n  search download upload or import team Skills, look up this user or a\n  teammate's past chats across Claude Code Codex Cursor OpenCode Trae\n  DeepSeek Harness and ngagent, upgrade the client, or debug a stuck daemon.\n  Team chats: xskill traj search then traj read. This machine only: add\n  --local to read ~\u002F.xskill\u002F*_sessions. Invoke as \u002Fxskill-helper.\n---\n\n# xskill-helper\n\nxskill is a thin client + background daemon that (1) mounts your team's shared\nSkills into every AI-agent tool you use (Claude Code, Codex, OpenCode, Cursor,\nTrae, DeepSeek Harness) and (2) collects your agent trajectories and syncs\nthem to a team server; you choose which projects are uploaded (see\n\"Controlling which trajectories are uploaded\"). You join a server once, then\nit keeps skills in sync and auto-updates itself.\n\n`xskill init` can install this guide into chosen harnesses on this machine.\n`xskill connect` also installs it after a team join. Invoke it as\n`\u002Fxskill-helper` from Claude Code, Codex, Cursor, and the other supported\ntools.\n\nAll state lives under `~\u002F.xskill\u002F`:\n\n| File | What |\n|---|---|\n| `~\u002F.xskill\u002Flocal_init.json` | local harness scan marker (no team server required) |\n| `~\u002F.xskill\u002Fteam_client.json` | connection identity (server_url, client_id, join_token, server privacy mode) — survives restarts |\n| `~\u002F.xskill\u002Fprivacy.json` | local upload rules: mode + per-project allow\u002Fdeny — never sent to the server |\n| `~\u002F.xskill\u002Fconnect_daemon.json` | current background daemon (pid \u002F host task) — used by `status`\u002F`stop` |\n| `~\u002F.xskill\u002Flogs\u002Fxskill.*.log` | split logs (one file per component) |\n| `~\u002F.xskill\u002Fskill\u002F` | the local skill repo everything is mounted from |\n\n## Getting started\n\nAfter `pip install xskill`, local trajectory search does not need a team\nserver. The first `xskill traj search` checks for a connection. If there is\nno `xskill connect` and this machine has not been initialized, it scans\ndetected harnesses, converts their sessions into `~\u002F.xskill\u002F*_sessions`,\nand builds the local index. After that, `xskill traj search` and\n`xskill traj read` work on this computer.\n\nFor a guided setup that lists harnesses, turns on trajectory processing, and\nasks which agents should get this guide:\n\n```bash\nxskill init\n```\n\n`xskill init -y` is the non-interactive form: scan every detected harness\nand install `\u002Fxskill-helper` into all of them. `--no-skill` skips the\nguide. `--skills-only` only installs the guide. `--force` rescans even if\nthe local index already exists.\n\nInteractive `xskill init` then asks whether to join a team server. Enter,\nskip, or an empty address\u002Ftoken all leave the machine usable locally. Do\nnot treat a skipped connect as failure. If they want to connect later, or\nrun a standalone team server:\n\n```bash\nxskill connect \u003Chost:port> --token \u003Ctoken> --name \u003Cuser-id>\nxskill serve --server\n```\n\nNever send the user to a public hub or any address you were not given. If\nthey do not have host, token, or name, show the connect command as the\nexample and ask their server operator. The operator prints the token with\n`xskill serve --server`. Do not invent an address or token, and do not\npaste one from the internet.\n\nAfter a successful handshake, connect installs this guide and starts the\nbackground daemon. Later reconnects can omit the address and token; they\nreuse `~\u002F.xskill\u002Fteam_client.json`.\n\nIn the agent, invoke this guide as `\u002Fxskill-helper`.\n\n## Controlling which trajectories are uploaded\n\nRules live only on this machine. A *mode* decides what happens to projects\nwithout a rule; the server and this machine each have one, and the stricter\nwins (`allowlist` is stricter than `denylist`):\n\n- `allowlist` — nothing is uploaded except projects you `allow`.\n- `denylist` — everything is uploaded except projects you `deny` (server default).\n\nA server set to `allowlist` requires it from every client on this version or\nnewer; older clients do not know the field.\n\n```bash\nxskill privacy status                 # effective mode + every discovered project and its decision\nxskill privacy mode allowlist         # this machine: only upload allowed projects (server cannot loosen it)\ncd ~\u002Fcode\u002Fmy-service && xskill privacy allow   # allow this project (and subdirectories)\nxskill privacy deny ~\u002Fcode\u002Fsecret     # never read or upload this project\nxskill privacy clear ~\u002Fcode\u002Fsecret    # drop the rule, back to the mode default\nxskill privacy review                 # interactive walk through discovered projects\nxskill connect \u003Chost:port> --token \u003Ct> --privacy allowlist   # set the mode before the first upload\n```\n\nA skipped trajectory is never read, never uploaded and never marked as\nuploaded, so allowing it later uploads it on the next scan. A trajectory whose\nsidecar records no working directory (Cursor and Trae never do) cannot be\nmatched to a project: it follows the mode default and `status` lists it\nseparately.\nRules never delete what was already uploaded.\n\n## Generate or rewrite a Skill\n\n`generate` does not invent a Skill from scratch. It reads trajectories the team\nserver already allows this client to see, and the instruction only describes\nwhat to create or change. The job may wait for a free SkillEdit seat; the CLI\nstreams queue and run logs. When it finishes, the Skill is committed to main\nand pinned on the initiator's recommendation list.\n\n```bash\nxskill generate \"创建一个排查 Python 内存泄漏的 Skill，包含常用诊断命令\"\nxskill generate \"改写现有的 python-memory-debug Skill，补充 Windows 排查步骤\"\nxskill generate --name alice,bob \"根据这些用户的成功案例生成数据库迁移 Skill\"\n```\n\n`--name` is a comma-separated list of employee ids the agent should read first.\nOmit it and the agent may search every trajectory the server authorizes. If the\nCLI says the server is too old, ask the operator to upgrade the team server.\n\n## Import an existing Skill\n\n`import` takes a local Skill folder (or a parent that contains several) into\nthe team's own repo. This is not `upload`: import becomes a first-party Skill\non main, while upload only lands in the user's SkillHub share.\n\n```bash\nxskill import .\u002Fmy-skill\nxskill import .\u002Fskills-parent --json\n```\n\n## Searching and reading trajectories\n\nThis is how the agent looks up past chats. xskill collects sessions from\nevery harness it detected on this machine and converts them to\n`traj_*.md`. After `xskill connect` it also uploads them to the team\nserver. One search covers Claude Code, Codex, Cursor, OpenCode, Trae,\nDeepSeek Harness, and ngagent. Do not tell the user to open each tool's\nown history UI.\n\nThe default entry is the trajectory, not Atom.\n\n`xskill traj search` is full-text over `traj_*.md`, not the first-user-query\nindex. Default listing is one block per hit: `traj_id`, then the first\nmatch with three lines before and after (the hit line is marked `*`).\nSame-name prefixes are spread. A page shows at most 30 hits. `--page N`\nturns the page. `--cards` returns index cards (source, line count, user\nturns, tools, then `L 问` \u002F `L 答`), at most 8 per page; cards are not\nclose reading. Close-read with\n`xskill traj read \u003Ctraj_id> --offset-start \u003CL from the card>`.\n\nNo team server (after pip, or after `xskill init`): search reads\n`~\u002F.xskill\u002F*_sessions`. The first search on this machine runs the same\nharness scan as `xskill init` if that has not happened yet.\n\nTeam (online, after `xskill connect`): search reads uploaded\n`traj_*.md` this user and teammates can see. `--name alice,bob`\nnarrows to those employee ids. Reading someone else's file needs the\nserver switch `team.server.allow_read_others`; otherwise the CLI\nprints that the server has not opened others' trajectories.\n\nThis machine only (offline, or skip the server): add `--local`.\n`xskill traj search --local` and `xskill traj read --local \u003Ctraj_id>`\nuse `~\u002F.xskill\u002F*_sessions` and do not call the team server. If this\nmachine has not been initialized, the first `--local` search or read\nscans harnesses and converts sessions. `--local` also prints the\nconcrete `*_sessions` directories. After that, grep those folders\nwith this harness's own search tools (`traj_*.md`); do not invent a\nserver path.\n\nPutting the word traj after `xskill search` searches skills, not\ntrajectories. Use `xskill traj search`.\n\n`xskill traj read` takes `--offset-start` and `--offset-end`\n(1-based, half-open). Each reply prints the current window and the\ntotal window. One call returns at most 200 lines.\n\nLocal harness directories (this machine):\n\n| Harness | Bridged markdown |\n|---|---|\n| Claude Code | `~\u002F.xskill\u002Fcc_sessions\u002Ftraj_*.md` |\n| Codex | `~\u002F.xskill\u002Fcodex_sessions\u002Ftraj_*.md` |\n| Cursor | `~\u002F.xskill\u002Fcursor_sessions\u002Ftraj_*.md` |\n| OpenCode | `~\u002F.xskill\u002Fopencode_sessions\u002Ftraj_*.md` |\n| ngagent | `~\u002F.xskill\u002Fngagent_sessions\u002Ftraj_*.md` |\n| nga3 | `~\u002F.xskill\u002Fnga3_sessions\u002Ftraj_*.md` |\n| Trae | `~\u002F.xskill\u002Ftrae_sessions\u002Ftraj_*.md` |\n| DeepSeek Harness | `~\u002F.xskill\u002Fdsh_sessions\u002Ftraj_*.md` |\n\n```bash\nxskill traj search 内存泄漏\nxskill traj search patentdagger --page 2\nxskill traj search patentdagger --cards\nxskill traj search --cards traj_cc_patentdagger_43773fc8\nxskill traj search \"alembic 半迁移\" --cards --page 2\nxskill traj search --name alice,bob 发票核对 --json\nxskill traj read traj_cc_alice_memleak\nxskill traj read traj_cc_alice_memleak --offset-start 12 --offset-end 88\nxskill traj read --local traj_cc_alice_memleak\nxskill traj read --local traj_cc_alice_memleak --offset-start 12 --offset-end 88\n```\n\n`--name` only applies in team mode. Search does not download files.\nPaste a `traj_id` into `xskill generate` when the instruction should\nname the evidence. Do not print server paths.\n\n## Advanced: atoms\n\nAtom is a split-agent product. Granularity may change across versions.\nPrefer `xskill traj search` unless a script already has an `atom_id`.\n\n```bash\nxskill atom search 内存泄漏\nxskill atom search --name alice,bob 发票核对\nxskill atom read atom_t_0001\nxskill atom read atom_t_0001 --offset-start 40 --json\n```\n\n## Searching & sharing team skills\n\n```bash\nxskill init                    # this machine: scan harnesses, convert traj, optional helper\nxskill search \u003Cquery...>       # search team skills; returns metadata only\nxskill traj search \u003Cquery...>  # no team: local index; after connect: this user and teammates\nxskill traj read \u003Ctraj_id>     # read trajectory lines; prints current and total range\nxskill traj read --local \u003Cid>  # this machine ~\u002F.xskill\u002F*_sessions, no team server\nxskill search \u003Cquery...> --download  # legacy 10-slot LRU download + auto-install\nxskill download \u003Cskill-id>     # persist one result; interactively select harnesses\nxskill download \u003Cskill-id> --agent claude-code --agent codex -y  # for agents\u002Fscripts\nxskill search auth retry -k 3  # top-3 results (max 10)\nxskill upload .\u002Fmy-skill       # package a SKILL.md folder and share to the team\nxskill dashboard               # print a passwordless link into the server dashboard\nxskill stats                   # token usage & estimated cost\n```\n\nAlready connected clients search SkillHub. Standalone machines search the local\nlibrary (semantic index, falling back to BM25). `search --download` is the old\nrolling-slot path; prefer `download \u003Cskill-id>` to keep a Skill permanently.\n\n## Upgrading\n\n```bash\npip install -U xskill        # manual upgrade to the latest PyPI release\nxskill update                # check PyPI now, upgrade + restart if newer exists\n```\n\nThe daemon auto-checks PyPI every hour and upgrades itself when a newer\nversion is out; disable with `xskill connect --no-auto-update`. Behind a\ncorporate proxy, add `--use-proxy` (default is direct connection, bypassing the\nSWG proxy). Background hosting differs per platform — Windows uses a Scheduled\nTask, Linux\u002FWSL uses a systemd user service. See\n[references\u002Fplatforms.md](references\u002Fplatforms.md).\n\n## Debugging\n\n```bash\nxskill connect --foreground          # run the daemon loop in the foreground, live logs\nxskill connect --foreground --debug  # + verbose logging\n```\n\n| Symptom | Command | Expected |\n|---|---|---|\n| Is it running \u002F connected? | `xskill status` | prints background task + pid, or \"not running\" |\n| My trajectories never reach the server | `xskill privacy status` | shows the effective mode and the decision per project; `allow` the project or switch mode |\n| Skills not updating | `xskill connect --foreground` | watch the reconcile loop; look for `copy-mode` \u002F `fell back to copy` warnings |\n| Stop it | `xskill stop` | tears down the Scheduled Task \u002F systemd unit |\n| Restart clean | `xskill stop` then `xskill connect` | re-handshakes and re-daemonizes |\n\nLogs are at `~\u002F.xskill\u002Flogs\u002Fxskill.*.log`. See\n[references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) for the\nblack-window, copy-mode, and can't-connect issues in detail.\n\n## Where skills land per tool\n\n| Tool | Install dir | Mount |\n|---|---|---|\n| Claude Code | `~\u002F.claude\u002Fskills\u002F\u003Cname>\u002F` | symlink → junction (Win) → copy |\n| Codex | `~\u002F.agents\u002Fskills\u002F\u003Cname>\u002F` (shared) | same |\n| OpenCode | `~\u002F.agents\u002Fskills\u002F\u003Cname>\u002F` (shared) | same |\n| Cursor | `~\u002F.cursor\u002Fskills\u002F\u003Cname>\u002F` | same |\n| Trae | IDE workspace \u002F `~\u002F.trae-cn` (auto-detected) | same |\n| DeepSeek Harness | `~\u002F.dsh\u002Fskills\u002F\u003Cname>\u002F` | same |\n\nsymlink\u002Fjunction installs are live — server updates show up instantly and your\nedits round-trip back. A copy-mode install is a snapshot: it logs a warning,\nupdates do not propagate live, and local edits do not round-trip. Copy mode\nonly happens when neither symlink nor junction can be made (cross-drive, or\nnon-NTFS). On Windows, enable Developer Mode so symlinks work — see\n[references\u002Fplatforms.md](references\u002Fplatforms.md).\n","# xskill-helper\n\nxskill is a thin client + background daemon that (1) mounts your team's shared\nSkills into every AI-agent tool you use (Claude Code, Codex, OpenCode, Cursor,\nTrae, DeepSeek Harness) and (2) collects your agent trajectories and syncs\nthem to a team server; you choose which projects are uploaded (see\n\"Controlling which trajectories are uploaded\"). You join a server once, then\nit keeps skills in sync and auto-updates itself.\n\n`xskill init` can install this guide into chosen harnesses on this machine.\n`xskill connect` also installs it after a team join. Invoke it as\n`\u002Fxskill-helper` from Claude Code, Codex, Cursor, and the other supported\ntools.\n\nAll state lives under `~\u002F.xskill\u002F`:\n\n| File | What |\n|---|---|\n| `~\u002F.xskill\u002Flocal_init.json` | local harness scan marker (no team server required) |\n| `~\u002F.xskill\u002Fteam_client.json` | connection identity (server_url, client_id, join_token, server privacy mode) — survives restarts |\n| `~\u002F.xskill\u002Fprivacy.json` | local upload rules: mode + per-project allow\u002Fdeny — never sent to the server |\n| `~\u002F.xskill\u002Fconnect_daemon.json` | current background daemon (pid \u002F host task) — used by `status`\u002F`stop` |\n| `~\u002F.xskill\u002Flogs\u002Fxskill.*.log` | split logs (one file per component) |\n| `~\u002F.xskill\u002Fskill\u002F` | the local skill repo everything is mounted from |\n\n## Getting started\n\nAfter `pip install xskill`, local trajectory search does not need a team\nserver. The first `xskill traj search` checks for a connection. If there is\nno `xskill connect` and this machine has not been initialized, it scans\ndetected harnesses, converts their sessions into `~\u002F.xskill\u002F*_sessions`,\nand builds the local index. After that, `xskill traj search` and\n`xskill traj read` work on this computer.\n\nFor a guided setup that lists harnesses, turns on trajectory processing, and\nasks which agents should get this guide:\n\n```bash\nxskill init\n```\n\n`xskill init -y` is the non-interactive form: scan every detected harness\nand install `\u002Fxskill-helper` into all of them. `--no-skill` skips the\nguide. `--skills-only` only installs the guide. `--force` rescans even if\nthe local index already exists.\n\nInteractive `xskill init` then asks whether to join a team server. Enter,\nskip, or an empty address\u002Ftoken all leave the machine usable locally. Do\nnot treat a skipped connect as failure. If they want to connect later, or\nrun a standalone team server:\n\n```bash\nxskill connect \u003Chost:port> --token \u003Ctoken> --name \u003Cuser-id>\nxskill serve --server\n```\n\nNever send the user to a public hub or any address you were not given. If\nthey do not have host, token, or name, show the connect command as the\nexample and ask their server operator. The operator prints the token with\n`xskill serve --server`. Do not invent an address or token, and do not\npaste one from the internet.\n\nAfter a successful handshake, connect installs this guide and starts the\nbackground daemon. Later reconnects can omit the address and token; they\nreuse `~\u002F.xskill\u002Fteam_client.json`.\n\nIn the agent, invoke this guide as `\u002Fxskill-helper`.\n\n## Controlling which trajectories are uploaded\n\nRules live only on this machine. A *mode* decides what happens to projects\nwithout a rule; the server and this machine each have one, and the stricter\nwins (`allowlist` is stricter than `denylist`):\n\n- `allowlist` — nothing is uploaded except projects you `allow`.\n- `denylist` — everything is uploaded except projects you `deny` (server default).\n\nA server set to `allowlist` requires it from every client on this version or\nnewer; older clients do not know the field.\n\n```bash\nxskill privacy status                 # effective mode + every discovered project and its decision\nxskill privacy mode allowlist         # this machine: only upload allowed projects (server cannot loosen it)\ncd ~\u002Fcode\u002Fmy-service && xskill privacy allow   # allow this project (and subdirectories)\nxskill privacy deny ~\u002Fcode\u002Fsecret     # never read or upload this project\nxskill privacy clear ~\u002Fcode\u002Fsecret    # drop the rule, back to the mode default\nxskill privacy review                 # interactive walk through discovered projects\nxskill connect \u003Chost:port> --token \u003Ct> --privacy allowlist   # set the mode before the first upload\n```\n\nA skipped trajectory is never read, never uploaded and never marked as\nuploaded, so allowing it later uploads it on the next scan. A trajectory whose\nsidecar records no working directory (Cursor and Trae never do) cannot be\nmatched to a project: it follows the mode default and `status` lists it\nseparately.\nRules never delete what was already uploaded.\n\n## Generate or rewrite a Skill\n\n`generate` does not invent a Skill from scratch. It reads trajectories the team\nserver already allows this client to see, and the instruction only describes\nwhat to create or change. The job may wait for a free SkillEdit seat; the CLI\nstreams queue and run logs. When it finishes, the Skill is committed to main\nand pinned on the initiator's recommendation list.\n\n```bash\nxskill generate \"创建一个排查 Python 内存泄漏的 Skill，包含常用诊断命令\"\nxskill generate \"改写现有的 python-memory-debug Skill，补充 Windows 排查步骤\"\nxskill generate --name alice,bob \"根据这些用户的成功案例生成数据库迁移 Skill\"\n```\n\n`--name` is a comma-separated list of employee ids the agent should read first.\nOmit it and the agent may search every trajectory the server authorizes. If the\nCLI says the server is too old, ask the operator to upgrade the team server.\n\n## Import an existing Skill\n\n`import` takes a local Skill folder (or a parent that contains several) into\nthe team's own repo. This is not `upload`: import becomes a first-party Skill\non main, while upload only lands in the user's SkillHub share.\n\n```bash\nxskill import .\u002Fmy-skill\nxskill import .\u002Fskills-parent --json\n```\n\n## Searching and reading trajectories\n\nThis is how the agent looks up past chats. xskill collects sessions from\nevery harness it detected on this machine and converts them to\n`traj_*.md`. After `xskill connect` it also uploads them to the team\nserver. One search covers Claude Code, Codex, Cursor, OpenCode, Trae,\nDeepSeek Harness, and ngagent. Do not tell the user to open each tool's\nown history UI.\n\nThe default entry is the trajectory, not Atom.\n\n`xskill traj search` is full-text over `traj_*.md`, not the first-user-query\nindex. Default listing is one block per hit: `traj_id`, then the first\nmatch with three lines before and after (the hit line is marked `*`).\nSame-name prefixes are spread. A page shows at most 30 hits. `--page N`\nturns the page. `--cards` returns index cards (source, line count, user\nturns, tools, then `L 问` \u002F `L 答`), at most 8 per page; cards are not\nclose reading. Close-read with\n`xskill traj read \u003Ctraj_id> --offset-start \u003CL from the card>`.\n\nNo team server (after pip, or after `xskill init`): search reads\n`~\u002F.xskill\u002F*_sessions`. The first search on this machine runs the same\nharness scan as `xskill init` if that has not happened yet.\n\nTeam (online, after `xskill connect`): search reads uploaded\n`traj_*.md` this user and teammates can see. `--name alice,bob`\nnarrows to those employee ids. Reading someone else's file needs the\nserver switch `team.server.allow_read_others`; otherwise the CLI\nprints that the server has not opened others' trajectories.\n\nThis machine only (offline, or skip the server): add `--local`.\n`xskill traj search --local` and `xskill traj read --local \u003Ctraj_id>`\nuse `~\u002F.xskill\u002F*_sessions` and do not call the team server. If this\nmachine has not been initialized, the first `--local` search or read\nscans harnesses and converts sessions. `--local` also prints the\nconcrete `*_sessions` directories. After that, grep those folders\nwith this harness's own search tools (`traj_*.md`); do not invent a\nserver path.\n\nPutting the word traj after `xskill search` searches skills, not\ntrajectories. Use `xskill traj search`.\n\n`xskill traj read` takes `--offset-start` and `--offset-end`\n(1-based, half-open). Each reply prints the current window and the\ntotal window. One call returns at most 200 lines.\n\nLocal harness directories (this machine):\n\n| Harness | Bridged markdown |\n|---|---|\n| Claude Code | `~\u002F.xskill\u002Fcc_sessions\u002Ftraj_*.md` |\n| Codex | `~\u002F.xskill\u002Fcodex_sessions\u002Ftraj_*.md` |\n| Cursor | `~\u002F.xskill\u002Fcursor_sessions\u002Ftraj_*.md` |\n| OpenCode | `~\u002F.xskill\u002Fopencode_sessions\u002Ftraj_*.md` |\n| ngagent | `~\u002F.xskill\u002Fngagent_sessions\u002Ftraj_*.md` |\n| nga3 | `~\u002F.xskill\u002Fnga3_sessions\u002Ftraj_*.md` |\n| Trae | `~\u002F.xskill\u002Ftrae_sessions\u002Ftraj_*.md` |\n| DeepSeek Harness | `~\u002F.xskill\u002Fdsh_sessions\u002Ftraj_*.md` |\n\n```bash\nxskill traj search 内存泄漏\nxskill traj search patentdagger --page 2\nxskill traj search patentdagger --cards\nxskill traj search --cards traj_cc_patentdagger_43773fc8\nxskill traj search \"alembic 半迁移\" --cards --page 2\nxskill traj search --name alice,bob 发票核对 --json\nxskill traj read traj_cc_alice_memleak\nxskill traj read traj_cc_alice_memleak --offset-start 12 --offset-end 88\nxskill traj read --local traj_cc_alice_memleak\nxskill traj read --local traj_cc_alice_memleak --offset-start 12 --offset-end 88\n```\n\n`--name` only applies in team mode. Search does not download files.\nPaste a `traj_id` into `xskill generate` when the instruction should\nname the evidence. Do not print server paths.\n\n## Advanced: atoms\n\nAtom is a split-agent product. Granularity may change across versions.\nPrefer `xskill traj search` unless a script already has an `atom_id`.\n\n```bash\nxskill atom search 内存泄漏\nxskill atom search --name alice,bob 发票核对\nxskill atom read atom_t_0001\nxskill atom read atom_t_0001 --offset-start 40 --json\n```\n\n## Searching & sharing team skills\n\n```bash\nxskill init                    # this machine: scan harnesses, convert traj, optional helper\nxskill search \u003Cquery...>       # search team skills; returns metadata only\nxskill traj search \u003Cquery...>  # no team: local index; after connect: this user and teammates\nxskill traj read \u003Ctraj_id>     # read trajectory lines; prints current and total range\nxskill traj read --local \u003Cid>  # this machine ~\u002F.xskill\u002F*_sessions, no team server\nxskill search \u003Cquery...> --download  # legacy 10-slot LRU download + auto-install\nxskill download \u003Cskill-id>     # persist one result; interactively select harnesses\nxskill download \u003Cskill-id> --agent claude-code --agent codex -y  # for agents\u002Fscripts\nxskill search auth retry -k 3  # top-3 results (max 10)\nxskill upload .\u002Fmy-skill       # package a SKILL.md folder and share to the team\nxskill dashboard               # print a passwordless link into the server dashboard\nxskill stats                   # token usage & estimated cost\n```\n\nAlready connected clients search SkillHub. Standalone machines search the local\nlibrary (semantic index, falling back to BM25). `search --download` is the old\nrolling-slot path; prefer `download \u003Cskill-id>` to keep a Skill permanently.\n\n## Upgrading\n\n```bash\npip install -U xskill        # manual upgrade to the latest PyPI release\nxskill update                # check PyPI now, upgrade + restart if newer exists\n```\n\nThe daemon auto-checks PyPI every hour and upgrades itself when a newer\nversion is out; disable with `xskill connect --no-auto-update`. Behind a\ncorporate proxy, add `--use-proxy` (default is direct connection, bypassing the\nSWG proxy). Background hosting differs per platform — Windows uses a Scheduled\nTask, Linux\u002FWSL uses a systemd user service. See\n[references\u002Fplatforms.md](references\u002Fplatforms.md).\n\n## Debugging\n\n```bash\nxskill connect --foreground          # run the daemon loop in the foreground, live logs\nxskill connect --foreground --debug  # + verbose logging\n```\n\n| Symptom | Command | Expected |\n|---|---|---|\n| Is it running \u002F connected? | `xskill status` | prints background task + pid, or \"not running\" |\n| My trajectories never reach the server | `xskill privacy status` | shows the effective mode and the decision per project; `allow` the project or switch mode |\n| Skills not updating | `xskill connect --foreground` | watch the reconcile loop; look for `copy-mode` \u002F `fell back to copy` warnings |\n| Stop it | `xskill stop` | tears down the Scheduled Task \u002F systemd unit |\n| Restart clean | `xskill stop` then `xskill connect` | re-handshakes and re-daemonizes |\n\nLogs are at `~\u002F.xskill\u002Flogs\u002Fxskill.*.log`. See\n[references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) for the\nblack-window, copy-mode, and can't-connect issues in detail.\n\n## Where skills land per tool\n\n| Tool | Install dir | Mount |\n|---|---|---|\n| Claude Code | `~\u002F.claude\u002Fskills\u002F\u003Cname>\u002F` | symlink → junction (Win) → copy |\n| Codex | `~\u002F.agents\u002Fskills\u002F\u003Cname>\u002F` (shared) | same |\n| OpenCode | `~\u002F.agents\u002Fskills\u002F\u003Cname>\u002F` (shared) | same |\n| Cursor | `~\u002F.cursor\u002Fskills\u002F\u003Cname>\u002F` | same |\n| Trae | IDE workspace \u002F `~\u002F.trae-cn` (auto-detected) | same |\n| DeepSeek Harness | `~\u002F.dsh\u002Fskills\u002F\u003Cname>\u002F` | same |\n\nsymlink\u002Fjunction installs are live — server updates show up instantly and your\nedits round-trip back. A copy-mode install is a snapshot: it logs a warning,\nupdates do not propagate live, and local edits do not round-trip. Copy mode\nonly happens when neither symlink nor junction can be made (cross-drive, or\nnon-NTFS). On Windows, enable Developer Mode so symlinks work — see\n[references\u002Fplatforms.md](references\u002Fplatforms.md).\n",{"name":5,"description":7},"main","d438329390736e6ea696c4fe25c53a72a06eda37","src\u002Fxskill\u002Fdata\u002Fskill\u002Fxskill-helper",27,"https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill","https:\u002F\u002Fxskill.wiki\u002F","https:\u002F\u002Fgithub.com\u002FSkillNerds","2026-11-11 18:14:29",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add SkillNerds\u002Fxskill --skill xskill-helper",{"tool":37,"command":38},"bunx","bunx skills add SkillNerds\u002Fxskill --skill xskill-helper",{"tool":40,"command":41},"pnpm","pnpm dlx skills add SkillNerds\u002Fxskill --skill xskill-helper","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fsrc\u002Fxskill\u002Fdata\u002Fskill\u002Fxskill-helper\u002FSKILL.md",true,[45,51],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"references\u002Fplatforms.md","REFERENCE","参考文档",2156,"https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fsrc\u002Fxskill\u002Fdata\u002Fskill\u002Fxskill-helper\u002Freferences\u002Fplatforms.md",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Ftroubleshooting.md",3753,"https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fsrc\u002Fxskill\u002Fdata\u002Fskill\u002Fxskill-helper\u002Freferences\u002Ftroubleshooting.md",[56,63,70,77,84],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":18},"01M72ADW9S00NVWZSWDNW5VZ90","demo-skill","A demo skill used to verify openclaw startup install path","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Ftests\u002Fdocker_e2e\u002Fscenarios\u002Fruntime_openclaw_detect\u002Fpre_state\u002F.xskill\u002Fskill\u002Fdemo-skill\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":18},"01M72ADV32DN09D8FFD049799J","xskill-demo-marker","A unique marker skill installed by xskill, used to verify openclaw sees it","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Ftests\u002Fdocker_e2e\u002Fopenclaw_real_llm\u002Fdemo_skill\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":18},"01M72ADMRHYGZ6Y73BPK82C60A","xskill-dev","Use when developing xskill itself — editing agent tool docstrings, changing the generate agent tool surface, or verifying what schema the Agno framework actually sends to the model. Covers the dump_schema workflow and where tool descriptions come from.","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fskills\u002Fxskill-dev\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":18},"01M72ADHX7EXZ7KV9FG1MEEVG8","using-xskill","Use when installing, configuring, or operating xskill (the `xskill` CLI \u002F `pip install xskill`) — starting the daemon, registering trajectory dirs, joining a team server, understanding how trajectories become Skills, or rebuilding\u002Fre-distilling the skill library after a model change.","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002Fskills\u002Fusing-xskill\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":18},"01M72ADFZV2TBW575YCCKWJDMG","xskill-registry-db-first","XSkill 数据访问原则：文件系统是内容真相，由 worker 同步进库；面板与业务读 请求只走数据库；内核 agent loop 可直接写盘。适用于改 dashboard、recommend、 team、pipeline、ecosystems 等读路径，以及审查是否又引入业务侧扫盘。","https:\u002F\u002Fgithub.com\u002FSkillNerds\u002Fxskill\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fxskill-registry-db-first\u002FSKILL.md",[],[]]