[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7189D8D3CX3FPQEWYRH9D95":3,"skill-same:01M7189D8D3CX3FPQEWYRH9D95":123},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":23,"linkOnly":24,"markdownContent":25,"markdownBody":25,"frontMatter":26,"refName":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":33,"ownerDescription":6,"updatedAt":34,"install":35,"resources":48},"01M7189D8D3CX3FPQEWYRH9D95","sentry",null,"anthropics\u002Fclaude-quickstarts","anthropics","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F76263028?v=4","01M697XGDPXJAVQ8391V2BCYTE","Anthropic","01M7188R62QSTJZJJ3DFX7XFT5",20972,"MIT","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fskill.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[5],"2026-11-10 10:12:30",false,"# Setup tips & tricks: scheduled Sentry triage with an MCP OAuth vault credential\n\nThings that aren't obvious from the docs and tend to cost debugging time. Use\nthe checklist at the bottom when walking someone through the quickstart.\n\n---\n\n## Mental model\n\n### Two Sentry authorizations, on purpose\n\n```\n.\u002Fstart.sh\n  ├─ installs\u002Fupdates the Sentry plugin for Claude Code (.claude\u002Fsettings.json)\n  └─ launches this guided setup\n       │\n       ├─ plugin MCP (Claude Code's own OAuth; \u002Fmcp to sign in):\n       │    list orgs and projects, confirm the target with the user\n       │\n       └─ .\u002Fagents\u002Fsetup.sh --org \u003Cslug> --project \u003Cslug>\n            ├─ sentry-config.json: the two non-secret slugs\n            ├─ ant apply: agent + environment + vault → claude-lock.json\n            └─ oauth_setup.py: browser grant → mcp_oauth credential in the vault\n                               → mcp-oauth-validate probe\n\ncron → deployment → session ──(vault injects token)──▶ mcp.sentry.dev\n                                                        issues + optional Seer RCA\n                                                        └── TRIAGE_REPORT.md\n```\n\nClaude Code owns the plugin's credential and does not expose it; the vault owns\nthe scheduled agent's credential and Anthropic refreshes it. Never inspect or\ncopy Claude Code's credential files, and never put a Sentry token in `.env`,\nthe shell environment, or the agent prompt.\n\n### Where the token lives\n\nAn `mcp_oauth` credential is keyed by `mcp_server_url`. When a session's agent\ndeclares an `mcp_servers` entry with exactly that URL, Anthropic attaches the\ntoken to the connection outside the sandbox. The container never holds it, so\nthere is nothing for a prompt injection to read. Two things must match\ncharacter for character: the agent's `mcp_servers[].url` and the credential's\n`auth.mcp_server_url`, both `https:\u002F\u002Fmcp.sentry.dev\u002Fmcp`.\n\nWhat this doesn't give you: the agent can do anything the grant and the\nenabled tools allow. Both are read-only here. The grant is what the user\napproved on Sentry's consent screen (**Inspect Issues & Events** and **Seer**\nonly) and the `mcp_toolset` is fail-closed with four tools enabled, so tools\nSentry adds later do not appear.\n\n### A deployment is the whole host process\n\nA **deployment** bundles the agent, environment, vault, and initial user\nmessage with a cron schedule. Sessions start themselves on Anthropic infra.\nNothing runs on your machine after `deploy.py`.\n\n---\n\n## Gotchas\n\n### The vault always sends `Authorization: Bearer`\n\nBoth MCP credential types (`mcp_oauth`, `static_bearer`) present the token as\n`Authorization: Bearer \u003Ctoken>`; there is no field for another scheme. On\nSentry's hosted MCP server, `Bearer` means a token from its own OAuth flow,\nwhich is bound to the approving user and inherits that user's standing in the\norg. SSO-enforced or otherwise policy-restricted organizations can reject that\ntoken on every call (seen as 403s on Sentry's own org; assume any SSO org may\nbehave this way until a manual run proves otherwise). The fix those orgs\nexpect, an org auth token or internal integration token, reaches the server\nonly as `Authorization: Sentry-Bearer \u003Ctoken>`, a fixed scheme of\n`mcp.sentry.dev` that no org can change and the vault cannot send. The consent\ncompletes and the credential saves either way, so `oauth_setup.py` runs\n`mcp-oauth-validate` right after create to catch it when the probe can. There\nis no in-vault workaround; the README's \"Known limitations\" has the details and\nthe `sentry-cli` alternative.\n\n### `allow_mcp_servers` is separate from `allowed_hosts`\n\nUnder `networking.type: limited`, the sandbox's `allowed_hosts` gates what\nshell commands can reach, and `allow_mcp_servers: true` is the separate opt-in\nfor the platform's own connection to declared MCP servers. This environment\nkeeps `allowed_hosts: []`, so `bash` has no network and Sentry is reachable\nonly through the MCP tools. Drop `allow_mcp_servers` and the session fails with\n`mcp_egress_blocked_error`.\n\n### The credential is created once, and archiving is how you re-authorize\n\n`setup.sh` lists the vault's credentials and skips OAuth if one already\nmatches the MCP URL. Archived credentials are not listed, so the re-auth path\nis: archive, re-run `setup.sh`, approve in the browser. The vault ID is in\n`claude-lock.json` under `.\u002Fvaults\u002Fsentry-triage.yaml`:\n\n```bash\nant beta:vaults:credentials list --vault-id \u003Cvault-id>\nant beta:vaults:credentials archive --vault-id \u003Cvault-id> --credential-id \u003Ccredential-id>\n.\u002Fagents\u002Fsetup.sh\n```\n\n`mcp_server_url`, `token_endpoint`, and `client_id` are immutable on a\ncredential, which is why replacement rather than update is the path.\n\n### The deployment pins an agent version\n\nAn agent update writes a new agent version, and sessions you start by hand use\nthe latest. The deployment doesn't: it keeps the version it pinned at create\ntime, so a prompt edit alone never reaches scheduled runs. Re-running\n`.\u002Fagents\u002Fsetup.sh` does both halves: `ant apply` publishes the change, then\n`deploy.py` (which setup runs when `.env` has a deployment ID) calls\n`deployments.update` to re-pin to the latest. The bare agent ID means \"latest\nversion\". The same call sends the environment ID, `vault_ids`, and the first\nmessage, because the deployment also keeps the ones it was created with. That\nmatters after you recreate a vault or environment, or change the org or project\nin `sentry-config.json`.\n\n### Cron is wall-clock, with DST edges\n\n`0 9 * * 1-5` in `America\u002FNew_York` fires at 9:00 AM Eastern regardless of\nDST. Times that don't exist on spring-forward day are skipped, and times that\noccur twice on fall-back day fire twice. If that matters, schedule outside 1-3\nAM local or use UTC. Runs may start up to 10 seconds late, granularity is\nper-minute, and an org can have up to 1,000 deployments.\n\nAfter `deployments.create`, check `schedule.upcoming_runs_at` in the response\nto confirm the expression fires when you expect (`deploy.py` prints it).\n\n### Permanent failures auto-pause the deployment\n\n`vault_not_found_error`, `agent_archived_error`, and\n`environment_archived_error` pause the deployment and set `paused_reason`, so a\nmisconfigured deployment doesn't keep failing on schedule indefinitely.\nTransient failures (rate limits, backend errors) don't pause. `runs.py` lists\nboth: every trigger writes a deployment run record with `error.type` when no\nsession was produced.\n\n### `pause` is not `archive`\n\n`pause` stops future scheduled triggers. In-flight sessions keep running, and\nmanual runs still work while paused. `unpause` resumes from the next occurrence\n(missed runs are not backfilled). `archive` is terminal.\n\n### Report files lag the session by a few seconds\n\nThe agent writes to `\u002Fmnt\u002Fsession\u002Foutputs\u002F`, which the Files API captures\nautomatically. Indexing can lag 1-3 seconds after the session goes idle, so\n`run_now.py` retries the empty list a few times before giving up.\n\n### Seer unavailable is not an auth failure\n\nSeer may be unavailable because the organization has not enabled it, the plan\ndoes not include it, repository\u002Fcode mappings are absent, or the issue category\nis not supported. The agent tries once per run, then labels the rest **Model\nhypothesis** instead of retrying. A Seer result is evidence-backed advice, not\npermission to run Autofix or open a pull request.\n\n---\n\n## Setup checklist\n\n1. Confirm `uv`, `ant` 2.0+, `jq`, and Claude Code are installed. Sign in\n   once with [`ant auth login`](https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fcli-sdks-libraries\u002Fcli\u002Fquickstart)\n   or put only `ANTHROPIC_API_KEY` in `.env`. The `ant` CLI and the SDK share\n   the login.\n2. Use the Sentry plugin's MCP tools to list the organizations and projects the\n   user can access. If the plugin's `sentry` server is not authenticated, ask\n   the user to run `\u002Fmcp` and sign in. Ask them to choose the target; do not\n   infer a production project.\n3. `uv sync`.\n4. `.\u002Fagents\u002Fsetup.sh --org \u003Corganization-slug> --project \u003Cproject-slug>` →\n   writes `sentry-config.json`, `ant apply` creates the agent, environment, and\n   vault and writes `claude-lock.json`, then `oauth_setup.py` opens the\n   browser. Explain the second consent before it opens and stop while the user\n   completes it. Read back the `credential: validation status` line; `invalid`\n   means the SSO\u002Fcustom-scheme limitation above.\n5. `uv run python deploy.py` → reads the three IDs from `claude-lock.json`,\n   creates the deployment, appends `CLAUDE_DEPLOYMENT_ID` to `.env`. Check the\n   printed upcoming runs.\n6. `uv run python run_now.py` → streams a manual run and downloads\n   `TRIAGE_REPORT.md` to `reports\u002F\u003Csession_id>\u002F`. Watch for `session.error`,\n   especially `mcp_authentication_failed_error`.\n7. Verify the report: searches are constrained to the selected org and project,\n   top issues are ranked by users affected, Seer status is explicit, each top\n   issue labels either **Seer RCA** or **Model hypothesis**, and no issue state,\n   code, or pull request was changed.\n8. Ask before leaving the cron deployment active. `uv run python runs.py` shows\n   history.\n\nTo change the prompt or model later, edit `agents\u002Fsentry-triage.md` and re-run\n`.\u002Fagents\u002Fsetup.sh`. It publishes the change and re-pins the deployment (see\nthe gotcha above).\n\nTo stop: `uv run python teardown.py` archives the deployment and everything in\n`claude-lock.json` (the vault's credential goes with it), then removes the\ndeployment ID from `.env` and deletes the lockfile. Skip it to leave the\nschedule running. Afterwards `.\u002Fagents\u002Fsetup.sh` and `deploy.py` start from\nscratch. If teardown fails partway, fix the cause and run it again.\n\n## Debugging a failed run\n\n| Symptom | Likely cause |\n|---|---|\n| `mcp_authentication_failed_error` on the first run, or `mcp-oauth-validate` says `invalid` right after setup | The organization rejects the user-bound MCP OAuth token (SSO enforcement or a similar policy). The org-issued token it would accept needs `Sentry-Bearer`, which the vault cannot send. See the README's \"Known limitations\" |\n| `mcp_authentication_failed_error` after runs that used to work | Refresh failed: grant revoked or upstream session lapsed. `mcp-oauth-validate` shows the failing step. Archive the credential and re-run `.\u002Fagents\u002Fsetup.sh` |\n| `mcp_egress_blocked_error` | `allow_mcp_servers: true` missing from the environment's `networking`. Restore it and re-run setup |\n| Session runs but never calls a Sentry tool | `mcp_toolset` missing or its `mcp_server_name` does not match the `mcp_servers` entry, or the tools were disabled in the Console after an agent update |\n| Session parks waiting for a confirmation | A tool's `permission_policy` is `always_ask`. Scheduled runs have no one to answer; set it back to `always_allow` and re-pin |\n| 403 from Sentry inside a tool result | The grant lacks the skill (approve **Inspect Issues & Events** and **Seer**), or the organization rejects user-bound OAuth tokens (see the first row) |\n| `runs.py` shows `vault_not_found_error` or `vault_archived_error` | Vault deleted or archived while the deployment still references it. `ant apply` notices too and refuses to guess: run `ant apply --force --yes vaults` to create a replacement, then `.\u002Fagents\u002Fsetup.sh`, which adds the credential to the new vault and points the deployment's `vault_ids` at it. The failure paused the deployment, so finish with `ant beta:deployments unpause --deployment-id $CLAUDE_DEPLOYMENT_ID` |\n| `ant apply` prints `refusing to apply` | A resource in `claude-lock.json` was changed, archived, or deleted in the Console. The plan says which and why. `--force` overwrites the edit or creates a replacement |\n| Scheduled time passed, no run record | Deployment paused (check `paused_reason`), or you're checking before the up-to-10s jitter |\n| `run_now.py` finds no files | Report indexing lag. The script retries, but if it still comes up empty, check the streamed transcript for whether the agent wrote the file |\n\n---\n\n## Production notes\n\n- Keep the grant read-only. A prompt injection in an issue title can at worst\n  read what the on-call engineer could; with **Triage Issues** approved it\n  could resolve or reassign issues.\n- The report lands in the session's files, not your inbox. For delivery,\n  register a `session.status_idled` webhook, download the report, and post it\n  to Slack (the [`..\u002Fslack`](..\u002Fslack) quickstart has the webhook pattern).\n  Subscribe to `vault_credential.refresh_failed` on the same webhook to hear\n  about a lapsed grant before the next morning's run fails.\n",{},"main","4a4224bdf0e9cfb3332167b69c90457403813625","managed-agents\u002Fsentry",3497,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts","https:\u002F\u002Fgithub.com\u002Fanthropics","https:\u002F\u002Fanthropic.com","2026-11-07 03:37:57",{"source":7,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add anthropics\u002Fclaude-quickstarts --skill sentry",{"tool":41,"command":42},"bunx","bunx skills add anthropics\u002Fclaude-quickstarts --skill sentry",{"tool":44,"command":45},"pnpm","pnpm dlx skills add anthropics\u002Fclaude-quickstarts --skill sentry","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fskill.md",true,[49,55,61,65,69,75,79,83,87,91,95,99,103,107,111,115,119],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},".claude\u002Fsettings.json","REFERENCE","参考文档",73,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002F.claude\u002Fsettings.json",{"path":56,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":59,"sourceHtmlUrl":60},".env.example","OTHER","其他",801,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002F.env.example",{"path":62,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":63,"sourceHtmlUrl":64},".gitignore",293,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002F.gitignore",{"path":66,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":67,"sourceHtmlUrl":68},"agents\u002Fsentry-triage.md",4818,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fagents\u002Fsentry-triage.md",{"path":70,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":73,"sourceHtmlUrl":74},"agents\u002Fsetup.sh","SCRIPT","脚本",3187,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fagents\u002Fsetup.sh",{"path":76,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":77,"sourceHtmlUrl":78},"CLAUDE.md",3518,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002FCLAUDE.md",{"path":80,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":81,"sourceHtmlUrl":82},"deploy.py",2720,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fdeploy.py",{"path":84,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":85,"sourceHtmlUrl":86},"environments\u002Fsentry-triage.yaml",455,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fenvironments\u002Fsentry-triage.yaml",{"path":88,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":89,"sourceHtmlUrl":90},"managed_agents.py",4411,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fmanaged_agents.py",{"path":92,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":93,"sourceHtmlUrl":94},"oauth_setup.py",11759,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Foauth_setup.py",{"path":96,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":97,"sourceHtmlUrl":98},"pyproject.toml",265,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fpyproject.toml",{"path":100,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":101,"sourceHtmlUrl":102},"README.md",10781,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002FREADME.md",{"path":104,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":105,"sourceHtmlUrl":106},"run_now.py",2422,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Frun_now.py",{"path":108,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":109,"sourceHtmlUrl":110},"runs.py",971,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fruns.py",{"path":112,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":113,"sourceHtmlUrl":114},"start.sh",1218,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fstart.sh",{"path":116,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":117,"sourceHtmlUrl":118},"teardown.py",2785,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fteardown.py",{"path":120,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":121,"sourceHtmlUrl":122},"vaults\u002Fsentry-triage.yaml",381,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fsentry\u002Fvaults\u002Fsentry-triage.yaml",[124,131,137,146,164,172],{"skillNo":125,"name":126,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":127,"categories":128,"keywords":129,"publishedAt":130},"01M73EH1C0DZGXSXNPQVK9VP4B","mcp-server-typescript","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fmcp-server-typescript\u002Fskill.md",[],[],"2026-11-11 15:11:14",{"skillNo":132,"name":133,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":134,"categories":135,"keywords":136,"publishedAt":130},"01M73EGYDTB5XECYS365QPRG55","linear","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Flinear\u002Fskill.md",[],[],{"skillNo":138,"name":139,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":140,"categories":141,"keywords":145,"publishedAt":23},"01M7189FFG0GVY39QBCTSYS5AZ","slack","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fslack\u002Fskill.md",[142],{"categoryNo":143,"categoryName":144,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000005","邮件与沟通",[139],{"skillNo":147,"name":148,"nameZh":6,"description":149,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":150,"categories":151,"keywords":152,"publishedAt":23},"01M7189B12649K1GD9TGXK7DBY","openshell-policy-advisor","Use when an OpenShell sandbox returns policy_denied, mentions policy.local, or needs a narrow network policy proposal.","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fself-hosted-sandboxes\u002Fopenshell-policy\u002Fsandbox\u002Fopenshell-skills\u002Fpolicy-advisor\u002FSKILL.md",[],[153,154,155,156,157,158,159,160,161,162,163],"policy","advisor","openshell policy advisor","local","network","sandbox","denied","proposal","use","when","returns",{"skillNo":165,"name":166,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":167,"categories":168,"keywords":169,"publishedAt":23},"01M71899E0V3MWC8B65EWJ5P5W","roadtrip-planner","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Froadtrip-planner\u002Fskill.md",[],[170,171],"roadtrip planner","planner",{"skillNo":173,"name":174,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":180,"publishedAt":23},"01M71895KCS8K44E1JFJTAX372","knowledge-wiki","https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-quickstarts\u002Fblob\u002Fmain\u002Fmanaged-agents\u002Fknowledge-wiki\u002Fskill.md",[177],{"categoryNo":178,"categoryName":179,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000021","资料检索与研究",[181,182,183],"knowledge","wiki","knowledge wiki"]