Datasets:
id stringlengths 6 6 | title stringlengths 17 90 | difficulty stringclasses 3
values | category stringclasses 5
values | description stringlengths 114 694 | tools_provided stringlengths 6 30 | initial_state stringlengths 154 1.78k | expected_state stringlengths 136 1k | scoring stringclasses 1
value | max_turns int64 3 20 | optimal_commands int64 1 12 | timeout_seconds int64 30 180 |
|---|---|---|---|---|---|---|---|---|---|---|---|
cb-001 | List open issues in a GitHub repo | easy | project_mgmt | List all currently open issues in the repository "acme-corp/web-platform".
Display the issue number, title, and assignee for each.
| ["gh"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 42, "title": "Fix login redirect loop", "state": "open", "assignee": "alice", "labels": ["bug", "auth"]}, {"number": 43, "title": "Add dark mode support", "state": "open", "assignee": "bob", "labels": ["enhancement", "ui"]}, {"number": 44, "title": "Upd... | {"gh": {"command_history": [{"pattern": "gh issue list.*--repo acme-corp/web-platform.*--state open"}], "output_contains": ["42", "43", "45", "46"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-002 | Create a GitHub issue with title and body | easy | project_mgmt | Create a new issue in the repository "acme-corp/web-platform" with the title
"Add rate limiting to public API endpoints" and a body that describes the need
to prevent abuse on unauthenticated API routes.
| ["gh"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 50, "title": "Existing issue", "state": "open", "assignee": "alice", "labels": []}]}}}} | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"title": "Add rate limiting to public API endpoints", "state": "open", "body_contains": "rate limit"}]}}, "command_history": [{"pattern": "gh issue create.*--repo acme-corp/web-platform.*--title"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-003 | List channels in Slack workspace | easy | communication | List all public channels in the Slack workspace. Show the channel name,
member count, and topic for each channel.
| ["slack"] | {"slack": {"channels": [{"id": "C001", "name": "general", "is_private": false, "num_members": 45, "topic": "Company-wide announcements"}, {"id": "C002", "name": "engineering", "is_private": false, "num_members": 22, "topic": "Engineering discussions"}, {"id": "C003", "name": "design-team", "is_private": true, "num_memb... | {"slack": {"command_history": [{"pattern": "slack channel list"}], "output_contains": ["general", "engineering", "random", "incidents"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-004 | Send a message to a Slack channel | easy | communication | Send the message "Deployment v2.3.1 completed successfully. All health checks passing."
to the #engineering channel in Slack.
| ["slack"] | {"slack": {"channels": [{"id": "C002", "name": "engineering", "is_private": false, "num_members": 22, "topic": "Engineering discussions", "messages": []}, {"id": "C001", "name": "general", "is_private": false, "num_members": 45, "topic": "Company-wide announcements", "messages": []}]}} | {"slack": {"channels": [{"name": "engineering", "messages": [{"text_contains": "Deployment v2.3.1 completed successfully"}]}], "command_history": [{"pattern": "slack message send.*--channel.*engineering"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-005 | List issues assigned to a specific user in Linear | easy | project_mgmt | List all issues in Linear that are currently assigned to the user "sarah@acme.com".
Show the issue identifier, title, status, and priority.
| ["linear"] | {"linear": {"issues": [{"id": "ACM-101", "title": "Implement OAuth2 PKCE flow", "status": "in_progress", "priority": "high", "assignee": "sarah@acme.com", "team": "Platform"}, {"id": "ACM-102", "title": "Database migration script", "status": "todo", "priority": "medium", "assignee": "sarah@acme.com", "team": "Platform"... | {"linear": {"command_history": [{"pattern": "linear issue list.*--assignee.*sarah"}], "output_contains": ["ACM-101", "ACM-102", "ACM-104"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-006 | Create a Linear issue with priority | easy | project_mgmt | Create a new issue in the Linear team "Platform" with the title
"Implement webhook retry logic with exponential backoff", set the priority
to "urgent", and assign it to "sarah@acme.com".
| ["linear"] | {"linear": {"teams": [{"name": "Platform", "key": "ACM", "members": ["sarah@acme.com", "mark@acme.com"]}], "issues": [{"id": "ACM-110", "title": "Existing backlog item", "status": "backlog", "priority": "low", "assignee": "mark@acme.com", "team": "Platform"}]}} | {"linear": {"issues": [{"title": "Implement webhook retry logic with exponential backoff", "priority": "urgent", "assignee": "sarah@acme.com", "team": "Platform"}], "command_history": [{"pattern": "linear issue create.*--team.*Platform.*--priority.*urgent"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-007 | Search for Slack messages containing a keyword | easy | communication | Search the Slack workspace for all messages containing the keyword "outage"
from the last 7 days. Display the channel, sender, timestamp, and message text.
| ["slack"] | {"slack": {"channels": [{"id": "C005", "name": "incidents", "messages": [{"user": "ops-bot", "text": "ALERT: Database outage detected in us-east-1", "ts": "2026-03-10T14:30:00Z"}, {"user": "alice", "text": "Investigating the outage \u2014 looks like connection pool exhaustion", "ts": "2026-03-10T14:35:00Z"}, {"user": "... | {"slack": {"command_history": [{"pattern": "slack message search.*outage"}], "output_contains": ["outage", "incidents", "engineering"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-008 | View a specific GitHub issue by number | easy | project_mgmt | View the full details of issue #42 in the repository "acme-corp/web-platform",
including its title, body, labels, assignee, and comments.
| ["gh"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 42, "title": "Fix login redirect loop", "state": "open", "assignee": "alice", "labels": ["bug", "auth"], "body": "Users are experiencing an infinite redirect loop when logging in\nwith SSO. The issue occurs after the OAuth callback redirects back\nto th... | {"gh": {"command_history": [{"pattern": "gh issue view 42.*--repo acme-corp/web-platform"}], "output_contains": ["Fix login redirect loop", "redirect"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-009 | List projects in Linear | easy | project_mgmt | List all projects in the Linear workspace. Show the project name, status,
lead, and target date for each project.
| ["linear"] | {"linear": {"projects": [{"name": "Q1 Platform Migration", "status": "in_progress", "lead": "sarah@acme.com", "target_date": "2026-03-31", "teams": ["Platform"]}, {"name": "Mobile App v3", "status": "in_progress", "lead": "mark@acme.com", "target_date": "2026-04-15", "teams": ["Mobile"]}, {"name": "Security Audit Remed... | {"linear": {"command_history": [{"pattern": "linear project list"}], "output_contains": ["Q1 Platform Migration", "Mobile App v3", "Security Audit Remediation", "Design System 2.0"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-010 | Send a DM to a user on Slack | easy | communication | Send a direct message to the Slack user "alice" with the text:
"Hey Alice, can you review PR #234 when you get a chance? It's the auth fix we discussed."
| ["slack"] | {"slack": {"users": [{"id": "U001", "name": "alice", "real_name": "Alice Chen", "status": "active"}, {"id": "U002", "name": "bob", "real_name": "Bob Smith", "status": "active"}, {"id": "U003", "name": "carol", "real_name": "Carol Davis", "status": "away"}], "direct_messages": {"alice": [], "bob": []}}} | {"slack": {"direct_messages": {"alice": [{"text_contains": "review PR #234"}]}, "command_history": [{"pattern": "slack message send.*--user.*alice"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-011 | Add a label to a GitHub issue | easy | project_mgmt | Add the label "priority:high" to issue #45 in the repository
"acme-corp/web-platform". The issue is about a memory leak in the
websocket handler.
| ["gh"] | {"gh": {"repos": {"acme-corp/web-platform": {"labels": [{"name": "priority:high", "color": "ff0000"}, {"name": "priority:low", "color": "00ff00"}, {"name": "bug", "color": "d73a4a"}], "issues": [{"number": 45, "title": "Memory leak in websocket handler", "state": "open", "assignee": null, "labels": ["bug"]}, {"number":... | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 45, "labels": ["bug", "priority:high"]}]}}, "command_history": [{"pattern": "gh issue edit 45.*--add-label.*priority:high"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-012 | Update a Linear issue status to done | easy | project_mgmt | Update the Linear issue "ACM-101" (Implement OAuth2 PKCE flow) to mark
its status as "done". The implementation has been completed and merged.
| ["linear"] | {"linear": {"issues": [{"id": "ACM-101", "title": "Implement OAuth2 PKCE flow", "status": "in_progress", "priority": "high", "assignee": "sarah@acme.com", "team": "Platform"}, {"id": "ACM-102", "title": "Database migration script", "status": "todo", "priority": "medium", "assignee": "sarah@acme.com", "team": "Platform"... | {"linear": {"issues": [{"id": "ACM-101", "status": "done"}], "command_history": [{"pattern": "linear issue update ACM-101.*--status.*done"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-013 | List Notion pages | easy | data_ops | List all pages in the Notion workspace. Show the page title, last edited time,
and parent database or page for each entry.
| ["notion"] | {"notion": {"pages": [{"id": "pg-001", "title": "Engineering Runbook", "last_edited": "2026-03-10T16:00:00Z", "parent": "Engineering Wiki", "created_by": "alice"}, {"id": "pg-002", "title": "Q1 OKRs", "last_edited": "2026-03-08T10:00:00Z", "parent": "Company Goals", "created_by": "ceo"}, {"id": "pg-003", "title": "Onbo... | {"notion": {"command_history": [{"pattern": "notion page list"}], "output_contains": ["Engineering Runbook", "Q1 OKRs", "Onboarding Checklist", "Incident Post-Mortem Template", "API Design Guidelines"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-014 | Create a Notion page | easy | data_ops | Create a new Notion page under the "Engineering Wiki" parent with the title
"Redis Caching Strategy". The page content should describe a caching approach
for the user service with TTL policies.
| ["notion"] | {"notion": {"pages": [{"id": "pg-001", "title": "Engineering Runbook", "last_edited": "2026-03-10T16:00:00Z", "parent": "Engineering Wiki", "created_by": "alice"}, {"id": "pg-004", "title": "Incident Post-Mortem Template", "last_edited": "2026-03-11T09:00:00Z", "parent": "Engineering Wiki", "created_by": "bob"}], "data... | {"notion": {"pages": [{"title": "Redis Caching Strategy", "parent": "Engineering Wiki", "content_contains": "caching"}], "command_history": [{"pattern": "notion page create.*--parent.*Engineering Wiki.*--title.*Redis Caching Strategy"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-015 | List Jira issues in a project | easy | project_mgmt | List all issues in the Jira project "INFRA" (Infrastructure). Show the issue
key, summary, status, assignee, and priority for each issue.
| ["jira"] | {"jira": {"projects": [{"key": "INFRA", "name": "Infrastructure", "issues": [{"key": "INFRA-201", "summary": "Upgrade Kubernetes to 1.29", "status": "In Progress", "assignee": "ops-alice", "priority": "High"}, {"key": "INFRA-202", "summary": "Set up Prometheus alerting rules", "status": "To Do", "assignee": "ops-bob", ... | {"jira": {"command_history": [{"pattern": "jira issue list.*--project.*INFRA"}], "output_contains": ["INFRA-201", "INFRA-202", "INFRA-203", "INFRA-204", "INFRA-205"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-016 | Create a Jira issue | easy | project_mgmt | Create a new Jira issue in the "INFRA" project with summary "Set up
centralized log aggregation with Loki", type "Task", priority "High",
and assign it to "ops-bob".
| ["jira"] | {"jira": {"projects": [{"key": "INFRA", "name": "Infrastructure", "members": ["ops-alice", "ops-bob", "ops-carol"], "issues": [{"key": "INFRA-205", "summary": "Implement VPC peering for staging", "status": "In Progress", "assignee": "ops-carol", "priority": "Medium"}]}]}} | {"jira": {"projects": [{"key": "INFRA", "issues": [{"summary": "Set up centralized log aggregation with Loki", "type": "Task", "priority": "High", "assignee": "ops-bob"}]}], "command_history": [{"pattern": "jira issue create.*--project.*INFRA.*--summary"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-017 | Search Gmail for messages from a sender | easy | communication | Search Gmail for all emails from "security@github.com" received in the
last 30 days. Display the subject, date, and a snippet of each message.
| ["google"] | {"google": {"gmail": {"messages": [{"id": "msg-001", "from": "security@github.com", "to": "dev@acme.com", "subject": "Security advisory: CVE-2026-1234", "date": "2026-03-05T08:00:00Z", "snippet": "A critical vulnerability has been identified in..."}, {"id": "msg-002", "from": "security@github.com", "to": "dev@acme.com"... | {"google": {"command_history": [{"pattern": "google gmail search.*from:security@github.com"}], "output_contains": ["CVE-2026-1234", "Dependabot alert", "Secret scanning alert"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-018 | List Google Calendar events for today | easy | data_ops | List all events on the primary Google Calendar for today (2026-03-12).
Show the event title, start time, end time, and attendees.
| ["google"] | {"google": {"calendar": {"events": [{"id": "evt-001", "title": "Daily Standup", "start": "2026-03-12T09:00:00Z", "end": "2026-03-12T09:15:00Z", "attendees": ["alice@acme.com", "bob@acme.com", "sarah@acme.com"], "calendar": "primary"}, {"id": "evt-002", "title": "Sprint Planning", "start": "2026-03-12T14:00:00Z", "end":... | {"google": {"command_history": [{"pattern": "google calendar list.*--date.*2026-03-12"}], "output_contains": ["Daily Standup", "Sprint Planning", "1:1 with Manager"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-019 | List files in Google Drive | easy | data_ops | List all files in the Google Drive folder "Engineering/Architecture".
Show the file name, size, last modified date, and owner.
| ["google"] | {"google": {"drive": {"folders": [{"path": "Engineering/Architecture", "files": [{"name": "System Design v2.pdf", "size": "4.2 MB", "modified": "2026-03-01T10:00:00Z", "owner": "alice@acme.com"}, {"name": "Database Schema.drawio", "size": "128 KB", "modified": "2026-02-28T14:00:00Z", "owner": "bob@acme.com"}, {"name": ... | {"google": {"command_history": [{"pattern": "google drive list.*Engineering/Architecture"}], "output_contains": ["System Design v2.pdf", "Database Schema.drawio", "API Spec OpenAPI.yaml", "Infrastructure Diagram.png", "Migration Plan Q1.docx"]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-020 | Comment on a Linear issue | easy | project_mgmt | Add a comment to the Linear issue "ACM-102" (Database migration script)
with the text: "Migration tested successfully on staging. Ready for production
deployment. Rolling update strategy confirmed with ops team."
| ["linear"] | {"linear": {"issues": [{"id": "ACM-102", "title": "Database migration script", "status": "in_progress", "priority": "medium", "assignee": "sarah@acme.com", "team": "Platform", "comments": [{"user": "sarah@acme.com", "body": "Schema changes drafted, running test suite."}, {"user": "mark@acme.com", "body": "Make sure to ... | {"linear": {"issues": [{"id": "ACM-102", "comments": [{"body_contains": "Migration tested successfully on staging"}]}], "command_history": [{"pattern": "linear issue comment ACM-102"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 3 | 1 | 30 |
cb-021 | Find urgent GitHub issues and post summary to Slack | medium | composite | Find all open GitHub issues in "acme-corp/web-platform" that have the label
"priority:critical" or "priority:high". Then compose a summary of these issues
and post it to the #engineering Slack channel with the format:
"Urgent Issues Report: N issues require attention" followed by a list.
| ["gh", "slack"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 42, "title": "Fix login redirect loop", "state": "open", "assignee": "alice", "labels": ["bug", "priority:critical"]}, {"number": 45, "title": "Memory leak in websocket handler", "state": "open", "assignee": null, "labels": ["bug", "priority:high"]}, {"... | {"gh": {"command_history": [{"pattern": "gh issue list.*--label.*priority"}]}, "slack": {"channels": [{"name": "engineering", "messages": [{"text_contains": "Urgent Issues Report"}, {"text_contains": "login redirect"}, {"text_contains": "Memory leak"}, {"text_contains": "SQL injection"}]}], "command_history": [{"patter... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-022 | Create Linear issues from Slack channel messages about bugs | medium | composite | Search the #incidents Slack channel for messages from the last 24 hours that
mention "bug" or "error". For each unique bug report found, create a
corresponding Linear issue in the "Platform" team with priority "high"
and status "triage".
| ["slack", "linear"] | {"slack": {"channels": [{"id": "C005", "name": "incidents", "messages": [{"user": "ops-bot", "text": "Bug detected: Payment processing fails for amounts > $10,000", "ts": "2026-03-12T08:30:00Z"}, {"user": "alice", "text": "Error in user registration \u2014 email validation regex rejects valid TLDs", "ts": "2026-03-12T0... | {"slack": {"command_history": [{"pattern": "slack message search.*(bug|error)"}]}, "linear": {"issues": [{"title_contains": "Payment processing", "priority": "high", "team": "Platform"}, {"title_contains": "email validation", "priority": "high", "team": "Platform"}, {"title_contains": "PDF export", "priority": "high", ... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 10 | 4 | 90 |
cb-023 | Sync GitHub issue labels with Linear issue priorities | medium | composite | For each open issue in the GitHub repo "acme-corp/web-platform" that has a
"linear:" prefixed label (e.g., "linear:ACM-101"), find the corresponding
Linear issue and update its priority to match the GitHub issue's priority
label. Map as follows: "priority:critical" -> urgent, "priority:high" -> high,
"priority:medium" ... | ["gh", "linear"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 50, "title": "Auth token expiry handling", "state": "open", "labels": ["bug", "priority:critical", "linear:ACM-201"]}, {"number": 51, "title": "API rate limiting", "state": "open", "labels": ["enhancement", "priority:medium", "linear:ACM-202"]}, {"numbe... | {"gh": {"command_history": [{"pattern": "gh issue list.*--repo acme-corp/web-platform"}]}, "linear": {"issues": [{"id": "ACM-201", "priority": "urgent"}, {"id": "ACM-202", "priority": "medium"}, {"id": "ACM-203", "priority": "high"}], "command_history": [{"pattern": "linear issue update ACM-201.*--priority.*urgent"}, {... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 10 | 5 | 90 |
cb-024 | Find unassigned Linear issues and notify team on Slack | medium | composite | Find all Linear issues in the "Platform" team that have status "todo" or
"in_progress" but have no assignee. Post a message to the #engineering
Slack channel listing these unassigned issues and asking team members
to pick them up.
| ["linear", "slack"] | {"linear": {"issues": [{"id": "ACM-301", "title": "Implement rate limiter middleware", "status": "todo", "priority": "high", "assignee": null, "team": "Platform"}, {"id": "ACM-302", "title": "Set up integration tests", "status": "in_progress", "priority": "medium", "assignee": null, "team": "Platform"}, {"id": "ACM-303... | {"linear": {"command_history": [{"pattern": "linear issue list.*--team.*Platform"}]}, "slack": {"channels": [{"name": "engineering", "messages": [{"text_contains": "ACM-301"}, {"text_contains": "ACM-302"}, {"text_contains": "ACM-304"}, {"text_contains": "unassigned"}]}], "command_history": [{"pattern": "slack message s... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-025 | Create Notion summary from Linear project status | medium | composite | Retrieve the status of the Linear project "Q1 Platform Migration" including
all its issues and their current states. Create a Notion page under
"Engineering Wiki" titled "Q1 Platform Migration - Status Report" that
summarizes the project progress: total issues, completed vs in-progress vs
todo counts, and lists any blo... | ["linear", "notion"] | {"linear": {"projects": [{"name": "Q1 Platform Migration", "status": "in_progress", "lead": "sarah@acme.com", "target_date": "2026-03-31", "teams": ["Platform"]}], "issues": [{"id": "ACM-401", "title": "Migrate user service to gRPC", "status": "done", "priority": "high", "project": "Q1 Platform Migration", "team": "Pla... | {"linear": {"command_history": [{"pattern": "linear issue list.*--project.*Q1 Platform Migration"}]}, "notion": {"pages": [{"title": "Q1 Platform Migration - Status Report", "parent": "Engineering Wiki", "content_contains": "blocked"}], "command_history": [{"pattern": "notion page create.*--title.*Q1 Platform Migration... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-026 | Find Jira issues blocked for more than 3 days and notify on Slack | medium | composite | Search the Jira project "INFRA" for issues with status "Blocked" that have
been in that status for more than 3 days. For each blocked issue, post a
notification to the #incidents Slack channel with the issue key, summary,
assignee, and how long it has been blocked.
| ["jira", "slack"] | {"jira": {"projects": [{"key": "INFRA", "name": "Infrastructure", "issues": [{"key": "INFRA-301", "summary": "Upgrade load balancer firmware", "status": "Blocked", "status_changed": "2026-03-07T10:00:00Z", "assignee": "ops-alice", "priority": "High", "blocker_reason": "Waiting for vendor approval"}, {"key": "INFRA-302"... | {"jira": {"command_history": [{"pattern": "jira issue list.*--project.*INFRA.*--status.*Blocked"}]}, "slack": {"channels": [{"name": "incidents", "messages": [{"text_contains": "INFRA-301"}, {"text_contains": "INFRA-303"}]}], "command_history": [{"pattern": "slack message send.*--channel.*incidents"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-027 | Search Gmail for meeting invites and create Linear tasks | medium | composite | Search Gmail for emails with subject containing "Action Items" or "Follow-up"
received in the last 7 days. Extract action items from these emails and create
corresponding Linear issues in the "Platform" team with appropriate priorities.
| ["google", "linear"] | {"google": {"gmail": {"messages": [{"id": "msg-101", "from": "alice@acme.com", "subject": "Meeting Action Items - Sprint Review 3/10", "date": "2026-03-10T16:00:00Z", "body": "Action items from today's sprint review:\n1. [HIGH] Investigate performance regression in search API\n2. [MEDIUM] Update API documentation for v... | {"google": {"command_history": [{"pattern": "google gmail search.*(Action Items|Follow-up)"}]}, "linear": {"issues": [{"title_contains": "performance regression", "priority": "high", "team": "Platform"}, {"title_contains": "CVE-2026-5678", "priority": "urgent", "team": "Platform"}], "command_history": [{"pattern": "lin... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 10 | 6 | 90 |
cb-028 | Find large files in Google Drive and post report to Slack | medium | composite | List all files in Google Drive that are larger than 50 MB. Create a summary
report with the file name, size, owner, and last modified date, then post
this report to the #engineering Slack channel with a message asking owners
to archive or clean up unused large files.
| ["google", "slack"] | {"google": {"drive": {"all_files": [{"name": "production-db-dump-20260301.sql", "size_mb": 2400, "modified": "2026-03-01T03:00:00Z", "owner": "ops-alice@acme.com", "path": "Backups/"}, {"name": "demo-video-q1.mp4", "size_mb": 350, "modified": "2026-02-15T14:00:00Z", "owner": "marketing@acme.com", "path": "Marketing/"},... | {"google": {"command_history": [{"pattern": "google drive list.*--min-size"}]}, "slack": {"channels": [{"name": "engineering", "messages": [{"text_contains": "production-db-dump"}, {"text_contains": "demo-video"}, {"text_contains": "ml-model-weights"}, {"text_contains": "load-test-results"}]}], "command_history": [{"pa... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-029 | Cross-reference GitHub PRs with Linear issues | medium | composite | List all open pull requests in "acme-corp/web-platform" that reference a
Linear issue in their title or body (pattern: ACM-NNN). For each referenced
Linear issue, update its status to "in_progress" if it is currently "todo".
Report any PRs that reference non-existent Linear issues.
| ["gh", "linear"] | {"gh": {"repos": {"acme-corp/web-platform": {"pull_requests": [{"number": 200, "title": "feat: Implement PKCE flow [ACM-501]", "state": "open", "author": "alice", "body": "Implements the OAuth2 PKCE flow as described in ACM-501."}, {"number": 201, "title": "fix: CORS headers [ACM-502]", "state": "open", "author": "bob"... | {"gh": {"command_history": [{"pattern": "gh pr list.*--repo acme-corp/web-platform"}]}, "linear": {"issues": [{"id": "ACM-501", "status": "in_progress"}, {"id": "ACM-502", "status": "in_progress"}, {"id": "ACM-503", "status": "in_progress"}], "command_history": [{"pattern": "linear issue update ACM-501.*--status.*in_pr... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 10 | 6 | 90 |
cb-030 | Create Notion doc listing all Slack channel topics | medium | composite | List all public Slack channels in the workspace and collect their names,
topics, and member counts. Create a Notion page under "Company Resources"
titled "Slack Channel Directory" that organizes all channels into a
structured document with descriptions and member counts.
| ["slack", "notion"] | {"slack": {"channels": [{"id": "C001", "name": "general", "is_private": false, "num_members": 45, "topic": "Company-wide announcements and updates"}, {"id": "C002", "name": "engineering", "is_private": false, "num_members": 22, "topic": "Engineering discussions, code reviews, technical decisions"}, {"id": "C003", "name... | {"slack": {"command_history": [{"pattern": "slack channel list"}]}, "notion": {"pages": [{"title": "Slack Channel Directory", "parent": "Company Resources", "content_contains": "engineering", "content_contains_2": "incidents"}], "command_history": [{"pattern": "notion page create.*--title.*Slack Channel Directory"}]}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 8 | 3 | 60 |
cb-031 | Incident response: find broken commit, create issue, assign on-call, notify Slack | hard | composite | A production incident has been reported. Investigate the GitHub repository
"acme-corp/web-platform" to find the most recent merged PR that touched
files in the "src/auth/" directory. Create a GitHub issue documenting the
incident and link it to that PR. Then find the on-call engineer from the
Linear team "Platform" (th... | ["gh", "linear", "slack"] | {"gh": {"repos": {"acme-corp/web-platform": {"pull_requests": [{"number": 180, "title": "feat: Add session refresh logic", "state": "closed", "merged": true, "merged_at": "2026-03-12T06:00:00Z", "author": "dave", "files_changed": ["src/auth/session.ts", "src/auth/refresh.ts"]}, {"number": 179, "title": "fix: Update use... | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"title_contains": "incident", "state": "open", "assignee": "mark", "body_contains": "180"}]}}, "command_history": [{"pattern": "gh pr list.*--state.*merged"}, {"pattern": "gh issue create.*--repo acme-corp/web-platform"}]}, "linear": {"command_history": [{"patter... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 15 | 7 | 120 |
cb-032 | Sprint planning: pull Linear backlog, create Notion sprint doc, notify team | hard | composite | Prepare for sprint planning. Pull all Linear issues from the "Platform"
team that are in "backlog" or "todo" status and have priority "high" or
"urgent". Create a Notion page under "Engineering Wiki" titled
"Sprint 12 Planning - 2026-03-12" that organizes these issues by priority,
includes estimated effort, and highlig... | ["linear", "notion", "slack"] | {"linear": {"teams": [{"name": "Platform", "key": "ACM"}], "issues": [{"id": "ACM-701", "title": "Implement GraphQL subscriptions", "status": "backlog", "priority": "urgent", "assignee": null, "team": "Platform", "estimate": 8, "labels": ["api"]}, {"id": "ACM-702", "title": "Add OpenTelemetry tracing", "status": "todo"... | {"linear": {"command_history": [{"pattern": "linear issue list.*--team.*Platform"}]}, "notion": {"pages": [{"title_contains": "Sprint 12 Planning", "parent": "Engineering Wiki", "content_contains": "ACM-701", "content_contains_2": "ACM-704", "content_contains_3": "blocked"}], "command_history": [{"pattern": "notion pag... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 12 | 5 | 120 |
cb-033 | Cross-platform project sync: GitHub milestones to Linear projects to Notion | hard | composite | Synchronize project tracking across three platforms. First, list all open
GitHub milestones in "acme-corp/web-platform". For each milestone, find or
create the corresponding Linear project (matching by name). Update the
Linear project's target date to match the milestone's due date. Finally,
create a Notion page under ... | ["gh", "linear", "notion"] | {"gh": {"repos": {"acme-corp/web-platform": {"milestones": [{"title": "v3.0 Release", "state": "open", "due_date": "2026-04-01", "open_issues": 12, "closed_issues": 28, "description": "Major version release with new auth system"}, {"title": "Security Hardening", "state": "open", "due_date": "2026-03-20", "open_issues":... | {"gh": {"command_history": [{"pattern": "gh api.*milestones"}]}, "linear": {"projects": [{"name": "v3.0 Release", "target_date": "2026-04-01"}, {"name": "Security Hardening", "target_date": "2026-03-20"}, {"name": "Performance Sprint", "status": "planned"}], "command_history": [{"pattern": "linear project update.*v3.0.... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 15 | 8 | 120 |
cb-034 | Bug triage: GitHub issues to Linear issues, assign by labels, notify Slack | hard | composite | Perform a full bug triage workflow. Find all open GitHub issues in
"acme-corp/web-platform" with the "bug" label that don't have a
"linear:" label yet. For each bug, create a Linear issue in the
appropriate team based on GitHub labels ("auth" -> Security team,
"api" -> Platform team, "ui" -> Frontend team, default -> P... | ["gh", "linear", "slack"] | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 100, "title": "XSS vulnerability in search input", "state": "open", "labels": ["bug", "auth", "priority:critical"], "assignee": null}, {"number": 101, "title": "REST API returns 500 on empty payload", "state": "open", "labels": ["bug", "api", "priority:... | {"gh": {"repos": {"acme-corp/web-platform": {"issues": [{"number": 100, "labels_contain": "linear:"}, {"number": 101, "labels_contain": "linear:"}, {"number": 102, "labels_contain": "linear:"}, {"number": 103, "labels_contain": "linear:"}]}}, "command_history": [{"pattern": "gh issue list.*--label.*bug"}, {"pattern": "... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 18 | 12 | 180 |
cb-035 | (kforge) Deploy artifact: list artifacts, trigger staging pipeline, check status | hard | custom_cli | Using the kforge CLI, deploy the latest build artifact for the "web-platform"
service to the staging environment. First, list available artifacts to find
the most recent successful build. Then trigger a deployment pipeline for
staging using that artifact. Wait for the pipeline to reach a terminal state
and report the r... | ["kforge"] | {"kforge": {"artifacts": [{"id": "art-050", "service": "web-platform", "version": "2.3.1", "build_status": "success", "created_at": "2026-03-12T08:00:00Z", "sha": "abc123f"}, {"id": "art-049", "service": "web-platform", "version": "2.3.0", "build_status": "success", "created_at": "2026-03-11T16:00:00Z", "sha": "def456a... | {"kforge": {"command_history": [{"pattern": "kforge artifact list.*--service.*web-platform"}, {"pattern": "kforge deploy trigger.*--env.*staging.*--artifact.*art-050"}, {"pattern": "kforge pipeline status"}], "deployments": {"staging": {"web-platform": {"current_version": "2.3.1", "artifact_id": "art-050"}}}}} | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 10 | 4 | 90 |
cb-036 | (flowctl) Create approval workflow: define steps, add gate, trigger run | hard | custom_cli | Using the flowctl CLI, create a new approval workflow called
"production-deploy-approval" for the web-platform service. The workflow
should have three steps: (1) "security-scan" — automated security scanning,
(2) "qa-signoff" — manual approval gate requiring sign-off from the QA lead,
(3) "release-manager-approval" — m... | ["flowctl"] | {"flowctl": {"workflows": [{"id": "wf-001", "name": "staging-deploy", "steps": [{"name": "lint-check", "type": "automated"}, {"name": "deploy-staging", "type": "automated"}], "status": "active"}], "runs": [], "users": [{"email": "qa-lead@acme.com", "role": "qa_lead"}, {"email": "release-mgr@acme.com", "role": "release_... | {"flowctl": {"workflows": [{"name": "production-deploy-approval", "steps": [{"name": "security-scan", "type": "automated"}, {"name": "qa-signoff", "type": "manual", "approver_contains": "qa-lead"}, {"name": "release-manager-approval", "type": "manual", "approver_contains": "release-mgr"}]}], "runs": [{"workflow": "prod... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 12 | 6 | 90 |
cb-037 | (alertmgr) Incident management: acknowledge alerts, create incident, escalate | hard | custom_cli | Using the alertmgr CLI, handle a production incident. First, list all
firing alerts to understand the scope. Acknowledge all critical alerts
related to the "payment-service". Create a new incident titled
"Payment Service Degradation" with severity "P1" and link the acknowledged
alerts to it. Assign the incident to the ... | ["alertmgr"] | {"alertmgr": {"alerts": [{"id": "alert-301", "name": "High error rate on payment-service", "severity": "critical", "status": "firing", "service": "payment-service", "started_at": "2026-03-12T14:00:00Z", "labels": {"team": "payments", "env": "production"}}, {"id": "alert-302", "name": "Payment-service latency > 5s", "se... | {"alertmgr": {"alerts": [{"id": "alert-301", "status": "acknowledged"}, {"id": "alert-302", "status": "acknowledged"}], "incidents": [{"title": "Payment Service Degradation", "severity": "P1", "assignee": "eng-manager@acme.com", "linked_alerts": ["alert-301", "alert-302"], "timeline": [{"entry_contains": "initial"}]}],... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 15 | 8 | 120 |
cb-038 | (datapipe) Build ETL: connect source, create transform, set up sink, schedule | hard | custom_cli | Using the datapipe CLI, build an ETL pipeline that extracts user activity
data from a PostgreSQL source, transforms it to compute daily active user
metrics, and loads the results into an Elasticsearch sink. Create a source
connection to the PostgreSQL database "analytics_db" on host
"db.internal:5432". Create a transfo... | ["datapipe"] | {"datapipe": {"sources": [], "transforms": [], "sinks": [], "pipelines": [{"id": "pipe-001", "name": "hourly-logs-pipeline", "status": "active", "schedule": "0 * * * *", "source": "src-010", "transform": "tx-010", "sink": "sink-010"}], "connections": [{"id": "src-010", "type": "source", "connector": "s3", "config": {"b... | {"datapipe": {"sources": [{"connector": "postgresql", "config_contains": {"host": "db.internal:5432", "database": "analytics_db"}}], "transforms": [{"type": "aggregation", "config_contains": "unique"}], "sinks": [{"connector": "elasticsearch", "config_contains": {"host": "es.internal:9200", "index": "user_metrics"}}], ... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 15 | 7 | 120 |
cb-039 | (meshctl + alertmgr) Detect failing service, reroute traffic, create alert | hard | composite | A service mesh health check has detected issues. Using meshctl, check the
health of all services in the "production" mesh. For any service with error
rate above 5%, reroute its traffic to the canary deployment (weight 0%
to primary, 100% to canary). Then use alertmgr to create alerts for each
failing service and create... | ["meshctl", "alertmgr"] | {"meshctl": {"mesh": "production", "services": [{"name": "user-service", "version": "1.4.0", "instances": 3, "error_rate": 0.2, "latency_p99_ms": 45, "traffic_weights": {"primary": 100, "canary": 0}}, {"name": "payment-service", "version": "2.1.0", "instances": 3, "error_rate": 12.5, "latency_p99_ms": 3200, "traffic_we... | {"meshctl": {"services": [{"name": "payment-service", "traffic_weights": {"primary": 0, "canary": 100}}, {"name": "notification-service", "traffic_weights": {"primary": 0, "canary": 100}}], "command_history": [{"pattern": "meshctl service health.*--mesh.*production"}, {"pattern": "meshctl traffic shift.*payment-service... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 18 | 10 | 150 |
cb-040 | (kforge + flowctl + slack) Full deployment: promote artifact, create approval gate, notify | hard | composite | Execute a full production deployment workflow using multiple tools.
First, use kforge to find the latest successful artifact for "web-platform"
that is currently deployed in staging. Verify the staging deployment is
healthy. Then use flowctl to create and trigger a "prod-release" approval
workflow with steps: automated... | ["kforge", "flowctl", "slack"] | {"kforge": {"artifacts": [{"id": "art-080", "service": "web-platform", "version": "3.0.0-rc1", "build_status": "success", "created_at": "2026-03-12T06:00:00Z", "sha": "abc123f"}, {"id": "art-079", "service": "web-platform", "version": "2.9.5", "build_status": "success", "created_at": "2026-03-10T14:00:00Z", "sha": "def... | {"kforge": {"deployments": {"production": {"web-platform": {"current_version": "3.0.0-rc1", "artifact_id": "art-080"}}}, "command_history": [{"pattern": "kforge artifact list.*--service.*web-platform"}, {"pattern": "kforge deploy status.*--env.*staging"}, {"pattern": "kforge deploy trigger.*--env.*production.*--artifac... | {"outcome": 0.6, "efficiency": 0.2, "recovery": 0.2} | 20 | 12 | 180 |
CLI-Bench: Benchmarking AI Agents on Command-Line Tool Orchestration
Abstract
CLI-Bench is an evaluation benchmark for measuring AI agents' ability to learn and use command-line interface (CLI) tools to complete real-world tasks. Unlike existing benchmarks that test general coding ability or narrow tool-use scenarios, CLI-Bench evaluates tool-agnostic CLI orchestration -- the capacity to read tool documentation, plan multi-step workflows, execute commands, interpret outputs, recover from errors, and achieve desired end states across diverse service domains.
The benchmark comprises 40 tasks spanning six categories (DevOps, project management, communication, data operations, custom CLI, and composite workflows) across 12 CLI tools. Tasks are grounded in stateful mock backends that simulate real services (GitHub, Slack, Linear, Notion, Google Workspace, Jira) with deterministic execution, enabling reproducible evaluation without live API dependencies. Each tool is defined via a declarative YAML adapter specification, making the benchmark trivially extensible to new tools.
A key contribution is the inclusion of five fictional CLI tools (kforge, flowctl, meshctl, datapipe, alertmgr) that no language model has encountered during training. These tools follow realistic CLI conventions but implement novel domain semantics, providing a memorization-proof evaluation of genuine tool-learning capability rather than pattern recall. Evaluation uses state-diffing against expected outcomes, efficiency measurement against optimal command counts, error recovery analysis, and a pass^k consistency metric adapted from tau-bench.
Key Features
- Tool-agnostic via YAML adapters -- Any CLI tool can be added by writing a YAML specification and a mock backend. No hardcoded tool knowledge in the harness.
- Fictional tools for memorization-proof evaluation -- Five novel CLI tools (kforge, flowctl, meshctl, datapipe, alertmgr) test genuine tool learning, not training data recall.
- Multi-turn execution -- Agents operate in a realistic loop: observe task and tool docs, issue commands, receive stdout/stderr, iterate until completion or timeout.
- State-diffing evaluation -- Scoring compares actual service state against expected state using deep recursive comparison with partial credit (0.0--1.0).
- pass^k consistency metric -- Measures reliability across k independent runs, not just peak performance. An agent must succeed on all k runs to score pass^k = 1.0.
- Deterministic mock backends -- All 7 service simulators (GitHub, Slack, Linear, Notion, Google, Jira, plus a generic fictional backend) are fully stateful and deterministic.
Benchmark Statistics
| Dimension | Value |
|---|---|
| Total tasks | 40 |
| Easy / Medium / Hard | 20 / 10 / 10 |
| Real-world CLI tools | 7 (gh, slack, linear, notion, google, jira, microsoft) |
| Fictional CLI tools | 5 (kforge, flowctl, meshctl, datapipe, alertmgr) |
| Task categories | 6 (devops, project_mgmt, communication, data_ops, custom_cli, composite) |
| Commands per tool | >= 5 |
| Max turns per task | 3--15 |
Evaluation Metrics
| Metric | Description |
|---|---|
| Outcome (default weight: 0.6) | State-diff score: fraction of expected state matched after execution |
| Efficiency (default weight: 0.2) | min(1.0, optimal_commands / actual_commands) |
| Recovery (default weight: 0.2) | 1.0 if errors encountered and recovered; 0.5 if no errors; 0.0 if errors with no recovery |
| pass^k | 1.0 if outcome >= 0.5 on all k runs, else 0.0. Measures consistency. |
Installation
pip install git+https://github.com/minervacap2022/CLI-Bench.git
# Or clone and install in development mode
git clone https://github.com/minervacap2022/CLI-Bench.git
cd CLI-Bench
pip install -e ".[dev]"
Quick Start
import asyncio
from pathlib import Path
from cli_bench.agents.dummy import DummyAgent
from cli_bench.harness.benchmark import BenchmarkRunner
async def main():
agent = DummyAgent()
runner = BenchmarkRunner(
tasks_dir=Path("data/tasks"),
agent=agent,
k=1,
)
report = await runner.run_all()
print(f"Overall score: {report.overall_score:.3f}")
print(f"Pass^k: {report.overall_pass_k:.3f}")
asyncio.run(main())
Or via the CLI:
python scripts/run_benchmark.py --agent dummy --k 1
Task Categories
| Category | Description | Example Task |
|---|---|---|
devops |
CI/CD, deployment, infrastructure management | Trigger a deployment pipeline and verify status |
project_mgmt |
Issue tracking, sprint planning, team coordination | Create and assign issues across projects |
communication |
Messaging, notifications, search | Send targeted messages based on channel context |
data_ops |
ETL pipelines, data transformation, monitoring | Build a data pipeline from source to sink |
custom_cli |
Fictional tool operations (memorization-proof) | Manage artifacts in kforge registry |
composite |
Multi-tool workflows spanning categories | Create issue in Linear, notify team in Slack, schedule review in Calendar |
Evaluation Metrics
Outcome (State Diffing)
The primary metric compares the actual state of mock backends against the task's expected state using deep recursive comparison. Dict keys are checked individually with partial credit; list membership is verified order-independently. The resulting score is a float in [0.0, 1.0].
Efficiency
Measures command economy: min(1.0, optimal_commands / actual_commands). An agent that uses exactly the optimal number of commands scores 1.0; using twice as many scores 0.5.
Recovery
Evaluates error handling:
- 1.0: Errors encountered during execution AND the agent successfully recovered (issued a successful command after the last error)
- 0.5: No errors encountered (neutral baseline)
- 0.0: Errors encountered but the agent failed to recover
pass^k
Adapted from tau-bench. Given k independent runs of the same task, pass^k = 1.0 only if all k runs achieve outcome >= 0.5. This measures consistency and reliability, not just peak performance.
Adding Custom Tools
1. Write a YAML Tool Adapter
Create cli_bench/tool_adapters/<tool_name>.yaml:
name: my-tool
description: "Description of the tool"
binary: mytool
auth:
type: env_var
key: MYTOOL_API_KEY
commands:
- name: resource list
description: "List all resources"
args:
- name: filter
type: string
required: false
description: "Filter expression"
output_format: json
side_effects: false
- name: resource create
description: "Create a new resource"
args:
- name: name
type: string
required: true
description: "Resource name"
output_format: json
side_effects: true
2. Implement a Mock Backend
For real tools, subclass BaseMockBackend. For fictional tools, use FictionalMockBackend which provides generic CRUD operations automatically:
from cli_bench.mock_backends.fictional import FictionalMockBackend
backend = FictionalMockBackend(
initial_state={"resources": [{"id": "res-1", "name": "alpha"}]},
tool_name="mytool",
)
3. Write Tasks
Create task YAMLs in data/tasks/ following the BenchTask schema (see existing tasks for examples).
Leaderboard
Results and model comparisons are hosted on HuggingFace Spaces:
https://huggingface.co/datasets/ChengyiX/CLI-Bench
Citation
If you use CLI-Bench in your research, please cite:
@misc{cli-bench-2026,
title={CLI-Bench: Benchmarking AI Agents on Command-Line Tool Orchestration},
author={{KLIK Team}},
year={2026},
url={https://github.com/minervacap2022/CLI-Bench},
}
License
This project is licensed under the Apache License 2.0. See LICENSE for details.
- Downloads last month
- 93