Skip to content
Sign up

flagon-io/g1t

public
0
891 linesCodeBlameRaw

Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.

Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step1//! The MCP server's tools: a few resource tools, each with an `action`.
2//!
3//! Every operation is one action of one tool. A call is dispatched to the
4//! operation it names, so permissions, the audit log, billing and outcomes
5//! are exactly those of the REST API. A token sees only the actions its
6//! scopes allow, and a tool none of whose actions it may use is not listed.
7//!
8//! The listed input schema is one flat object: `action`, then every field
9//! any of its actions takes. Which fields each action needs is in the
10//! `action` field's description and checked on every call. Claude's API,
11//! and so most MCP clients, refuse a tool whose input schema has `oneOf`
12//! at its top level, so the schema keyed by action, with each action's
13//! required fields, is [`discriminated`], published on the server's card
14//! and in the docs.
15
16use g1t_contracts::credentials::NEVER;
17use g1t_contracts::identity::AgentScope;
18use g1t_contracts::scopes::{Level, NO_SCOPE, TokenAccess, scope_for};
19use serde_json::{Map, Value, json};
20
21use crate::operations::Op;
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge22use crate::rules::RulesOp;
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar23use crate::security::SecurityOp;
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step24
25pub struct Action {
26 pub name: &'static str,
27 pub op: Op,
28 /// One line, for the `action` field's description.
29 pub summary: &'static str,
30}
31
32pub struct Tool {
33 pub name: &'static str,
34 pub title: &'static str,
35 /// What it is for, in a sentence or two.
36 pub description: &'static str,
37 pub actions: &'static [Action],
38 /// The action a call without one runs.
39 pub default_action: Option<&'static str>,
40}
41
42const fn a(name: &'static str, op: Op, summary: &'static str) -> Action {
43 Action { name, op, summary }
44}
45
46pub const TOOLS: &[Tool] = &[
47 Tool {
48 name: "search",
49 title: "Search",
50 description: "Find things. `code` (the default) searches all of g1t you can see: repositories, code, issues, pull requests and people, with qualifiers like repo:owner/name, language:rust, is:issue. `context` searches one workspace's catalog, docs, issues and memory by meaning.",
51 default_action: Some("code"),
52 actions: &[
53 a("code", Op::Search, "Search all of g1t: repositories, code, issues, pull requests, people"),
54 a("context", Op::SearchContext, "Search a workspace's context hub by meaning"),
55 a("entity", Op::GetEntity, "One catalog entry and its relations"),
56 a("ticket", Op::GetContext, "A Jira, Linear or Sentry item the work refers to, as it is now"),
57 ],
58 },
59 Tool {
60 name: "repository",
61 title: "Repositories",
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge62 description: "Repositories: find, read and create them, change their settings and rulesets (what may happen to branches and tags, and what a pull request needs to merge), check their CODEOWNERS file, manage their labels and milestones, and see and dismiss their security alerts (secrets and vulnerable dependencies). Name one as \"owner/name\". Deleting, transferring and changing visibility need `confirm`.",
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step63 default_action: None,
64 actions: &[
65 a("list", Op::ListRepos, "Repositories you can see"),
66 a("get", Op::GetRepo, "One repository"),
67 a("create", Op::CreateRepo, "Create one, empty or copied from a public git URL"),
68 a("update", Op::UpdateRepo, "Change description, website, topics, default branch, protection"),
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge69 a("get_settings", Op::GetRepoSettings, "How pull requests merge, and the default branch's protection as its rules stack"),
70 a("update_settings", Op::UpdateRepoSettings, "Change how pull requests merge and the default branch protection ruleset"),
71 a("check_names", Op::ListCheckNames, "Check names reported lately, to require in a ruleset"),
72 a("list_rulesets", Op::Rules(RulesOp::ListRepoRulesets), "Its rulesets, and its workspace's that hold in it"),
73 a("get_ruleset", Op::Rules(RulesOp::GetRepoRuleset), "One ruleset"),
74 a("create_ruleset", Op::Rules(RulesOp::CreateRepoRuleset), "Create a ruleset for its branches or tags"),
75 a("update_ruleset", Op::Rules(RulesOp::UpdateRepoRuleset), "Change a ruleset"),
76 a("delete_ruleset", Op::Rules(RulesOp::DeleteRepoRuleset), "Delete a ruleset"),
77 a("branch_rules", Op::Rules(RulesOp::GetBranchRules), "Every rule that holds for a branch or tag, and where it comes from"),
78 a("rule_evaluations", Op::Rules(RulesOp::ListRuleEvaluations), "How its rules judged pushes and merges, with insights"),
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar79 a("codeowners", Op::GetCodeownersErrors, "Problems in its CODEOWNERS file, by line"),
80 a("list_labels", Op::ListLabels, "Labels, with colors and how many issues and pull requests carry each"),
81 a("create_label", Op::CreateLabel, "Create a label"),
82 a("update_label", Op::UpdateLabel, "Rename a label or change its color or description"),
83 a("delete_label", Op::DeleteLabel, "Delete a label, from everything that carries it"),
84 a("add_default_labels", Op::AddDefaultLabels, "Add the default labels it is missing"),
85 a("list_milestones", Op::ListMilestones, "Milestones, with progress and due dates"),
86 a("get_milestone", Op::GetMilestone, "One milestone with its issues and pull requests"),
87 a("create_milestone", Op::CreateMilestone, "Create a milestone"),
88 a("update_milestone", Op::UpdateMilestone, "Change a milestone's title, description, due date or state"),
89 a("delete_milestone", Op::DeleteMilestone, "Delete a milestone"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step90 a("list_events", Op::ListEvents, "Timeline: pushes, issues, pull requests, comments"),
91 a("rename_branch", Op::RenameBranch, "Rename a branch"),
92 a("rename", Op::RenameRepo, "Rename it; old addresses redirect"),
93 a("transfer", Op::TransferRepo, "Move it to another workspace you own"),
94 a("archive", Op::ArchiveRepo, "Make it read-only"),
95 a("unarchive", Op::UnarchiveRepo, "Make it writable again"),
96 a("set_visibility", Op::SetRepoVisibility, "Make it public or private"),
97 a("delete", Op::DeleteRepo, "Delete it; restorable for 30 days"),
98 a("list_deleted", Op::ListDeletedRepos, "A workspace's deleted repositories"),
99 a("restore", Op::RestoreRepo, "Restore a deleted one"),
100 a("purge", Op::PurgeRepo, "Remove a deleted one for good"),
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily101 a("security_alerts", Op::ListSecurityAlerts, "Secret and dependency alerts, filtered by state"),
102 a("dismiss_alert", Op::DismissSecurityAlert, "Dismiss an alert with a reason"),
103 a("reopen_alert", Op::ReopenSecurityAlert, "Reopen a dismissed alert"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step104 ],
105 },
106 Tool {
107 name: "issue",
108 title: "Issues",
109 description: "Issues: what should change. Read one before working on it to see the pull requests already made for it. Issues and pull requests share numbers; `comment` works on either.",
110 default_action: None,
111 actions: &[
112 a("list", Op::ListIssues, "Issues on a repository, newest first"),
Fast pages, required checks on the branch, self-hosted runners, honest incidents113 a("get", Op::GetIssue, "One issue with comments and its pull requests"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step114 a("create", Op::CreateIssue, "Open an issue"),
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar115 a("update", Op::UpdateIssue, "Change title, body, labels, milestone or assignees"),
116 a("labels", Op::ListIssueLabels, "The labels an issue or pull request carries"),
117 a("add_labels", Op::AddIssueLabels, "Add labels to an issue or pull request"),
118 a("set_labels", Op::SetIssueLabels, "Replace the labels of an issue or pull request"),
119 a("remove_labels", Op::RemoveIssueLabels, "Take labels off an issue or pull request"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step120 a("close", Op::CloseIssue, "Close it without a pull request"),
121 a("reopen", Op::ReopenIssue, "Reopen it"),
122 a("comment", Op::AddComment, "Comment on an issue or pull request; path and line for one line of a change"),
123 a("import", Op::ImportIssue, "Open an issue from a Jira, Linear or Sentry item"),
124 ],
125 },
126 Tool {
127 name: "pull_request",
128 title: "Pull requests",
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar129 description: "Pull requests: start a change for an issue, record your session, mark it ready, ask people and teams to review, review and merge. Read `overlaps` and `behind` on `get` before going far, and `code_owners` for whose approval it needs.",
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step130 default_action: None,
131 actions: &[
132 a("list", Op::ListPullRequests, "Pull requests on a repository, newest first"),
Fast pages, required checks on the branch, self-hosted runners, honest incidents133 a("get", Op::GetPullRequest, "Status, checks and required checks, reviews, overlaps, whether it is behind"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step134 a("changes", Op::GetPullRequestChanges, "Files and line-by-line diff"),
135 a("create", Op::CreatePullRequest, "Start a draft with its own fork to push to, or open one from a pushed branch"),
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar136 a("update", Op::UpdatePullRequest, "Change its base branch, labels, milestone, assignees or reviewers"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step137 a("record_session", Op::RecordSession, "Append prompt, reasoning and tool entries to its session"),
138 a("read_session", Op::ReadSession, "Its recorded session"),
139 a("ready", Op::MarkPullRequestReady, "Mark a draft ready, with a summary"),
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar140 a("request_reviewers", Op::RequestReviewers, "Ask people or teams to review it"),
141 a("remove_requested_reviewers", Op::RemoveRequestedReviewers, "Stop asking people or teams to review it"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step142 a("review", Op::ReviewPullRequest, "Approve or request changes"),
143 a("close", Op::ClosePullRequest, "Close without merging"),
144 a("merge", Op::MergePullRequest, "Land it, or join the merge queue"),
145 a("merge_queue", Op::GetMergeQueue, "The repository's merge queue"),
146 ],
147 },
148 Tool {
149 name: "agent",
150 title: "g1t agents",
151 description: "Put g1t's agent to work and talk to it. One agent per issue; to do more at once, use more issues. Starting an agent uses the workspace's money.",
152 default_action: None,
153 actions: &[
154 a("delegate", Op::Delegate, "Open an issue and put an agent on it in one step"),
155 a("assign", Op::AssignIssue, "Put an agent on an existing issue"),
156 a("message", Op::MessageAgent, "Tell the agent on a pull request something, or ask another agent"),
157 a("answer", Op::AnswerMessage, "Answer a question or handoff sent to you"),
158 a("take_messages", Op::TakeMessages, "For a g1t agent: messages not seen yet"),
159 ],
160 },
161 Tool {
162 name: "plan",
163 title: "Plans",
Fast pages, required checks on the branch, self-hosted runners, honest incidents164 description: "Turn an outcome into issues: an agent proposes them with what done means and their dependencies; nothing opens until you apply the plan.",
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step165 default_action: None,
166 actions: &[
167 a("create", Op::PlanWork, "Ask an agent for a plan; read it with get until ready"),
168 a("get", Op::GetPlan, "A plan and the issues it proposes"),
169 a("apply", Op::ApplyPlan, "Open its issues; with assign, agents start in dependency order"),
170 ],
171 },
172 Tool {
173 name: "memory",
174 title: "Memory",
175 description: "What the project and its workspace remember for the next agent: how to build, conventions, decisions, traps. Recall before you start; remember one short fact at a time, never a secret.",
176 default_action: None,
177 actions: &[
178 a("recall", Op::Recall, "Search memory, or list it all"),
179 a("remember", Op::Remember, "Save one fact"),
180 ],
181 },
182 Tool {
183 name: "workflow",
184 title: "Workflows",
Fast pages, required checks on the branch, self-hosted runners, honest incidents185 description: "GitHub Actions workflows from .g1t/workflows: their runs, jobs and logs, and running, cancelling or rerunning them. Also the self-hosted runners they run on: a workspace's (`workspace`) or a repository's own (`repo`), their groups, and where agent work runs.",
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step186 default_action: None,
187 actions: &[
188 a("list", Op::ListWorkflows, "Workflows on the default branch"),
189 a("list_runs", Op::ListWorkflowRuns, "Runs, newest first"),
190 a("get_run", Op::GetWorkflowRun, "One run with its jobs and steps"),
191 a("job_logs", Op::GetJobLogs, "A job's log after a sequence number"),
192 a("dispatch", Op::DispatchWorkflow, "Run a workflow_dispatch workflow"),
193 a("cancel", Op::CancelWorkflowRun, "Cancel a run"),
194 a("rerun", Op::RerunWorkflowRun, "Run a finished run again"),
195 a("update", Op::UpdateWorkflow, "Turn a workflow on or off"),
Fast pages, required checks on the branch, self-hosted runners, honest incidents196 a("list_runners", Op::ListRunners, "Self-hosted runners, with status, labels and what each is doing"),
197 a("create_runner_token", Op::CreateRunnerRegistrationToken, "A one-hour token for g1t-runner register"),
198 a("remove_runner", Op::RemoveRunner, "Remove a self-hosted runner"),
199 a("list_runner_groups", Op::ListRunnerGroups, "A workspace's runner groups"),
200 a("create_runner_group", Op::CreateRunnerGroup, "Make a group, for some repositories"),
201 a("update_runner_group", Op::UpdateRunnerGroup, "Rename a group or change its repositories"),
202 a("delete_runner_group", Op::DeleteRunnerGroup, "Delete a group; its runners join the default"),
203 a("get_runner_settings", Op::GetRunnerSettings, "Where agent work runs; whether forks may use runners"),
204 a("update_runner_settings", Op::UpdateRunnerSettings, "Change them"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step205 ],
206 },
207 Tool {
208 name: "secret",
209 title: "Secrets and variables",
210 description: "A repository's or workspace's secrets and variables, read by workflows and deployments. Secret values are never returned.",
211 default_action: None,
212 actions: &[
213 a("list_secrets", Op::ListActionsSecrets, "Secrets, without values"),
214 a("set_secret", Op::SetActionsSecret, "Add or change a secret"),
215 a("delete_secret", Op::DeleteActionsSecret, "Remove a secret"),
216 a("list_variables", Op::ListActionsVariables, "Variables, with values"),
217 a("set_variable", Op::SetActionsVariable, "Add or change a variable"),
218 a("delete_variable", Op::DeleteActionsVariable, "Remove a variable"),
219 ],
220 },
221 Tool {
222 name: "webhook",
223 title: "Webhooks",
224 description: "HTTPS addresses sent signed events as they happen, for a repository or a whole workspace.",
225 default_action: None,
226 actions: &[
227 a("list", Op::ListWebhooks, "Webhooks, without secrets"),
228 a("create", Op::CreateWebhook, "Register one; a ping is sent"),
229 a("update", Op::UpdateWebhook, "Change address, events or active"),
230 a("delete", Op::DeleteWebhook, "Remove one"),
231 a("ping", Op::PingWebhook, "Send a ping"),
232 a("list_deliveries", Op::ListWebhookDeliveries, "Latest deliveries"),
233 a("redeliver", Op::RedeliverWebhook, "Send a delivery again"),
234 ],
235 },
236 Tool {
237 name: "access",
238 title: "Who has access",
239 description: "Who has access to a repository and with which role (read, triage, write, maintain, admin), outside collaborators, and a workspace's base permission.",
240 default_action: None,
241 actions: &[
242 a("list_collaborators", Op::ListCollaborators, "Everyone with a role, and pending invitations"),
243 a("get_permission", Op::GetCollaboratorPermission, "One person's role and capabilities"),
244 a("add_collaborator", Op::AddCollaborator, "Give someone a role, by username or email"),
245 a("update_collaborator", Op::UpdateCollaborator, "Change a direct role"),
246 a("remove_collaborator", Op::RemoveCollaborator, "Take away a direct role"),
247 a("list_invitations", Op::ListRepoInvitations, "Pending invitations to a repository"),
248 a("revoke_invitation", Op::RevokeRepoInvitation, "Withdraw one"),
249 a("set_base_permission", Op::SetBasePermission, "What every member gets on each repository"),
250 a("list_outside_collaborators", Op::ListOutsideCollaborators, "People with roles who are not members"),
251 ],
252 },
253 Tool {
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar254 name: "team",
255 title: "Teams",
256 description: "Teams: groups of a workspace's members, given roles on repositories together, mentioned as @workspace/team and asked to review together. Name one by `workspace` and its slug (`team`). Any member may create a team; the workspace's owners and the team's maintainers manage it. A secret team is seen only by its people and the owners.",
257 default_action: None,
258 actions: &[
259 a("list", Op::ListTeams, "A workspace's teams you can see"),
260 a("get", Op::GetTeam, "One team"),
261 a("create", Op::CreateTeam, "Create a team; you become its maintainer"),
262 a("update", Op::UpdateTeam, "Change its name, slug, description, visibility, parent or notifications"),
263 a("delete", Op::DeleteTeam, "Delete it; its child teams move up"),
264 a("list_members", Op::ListTeamMembers, "Its people and their roles, child teams' with include_child_teams"),
265 a("set_member", Op::SetTeamMember, "Add a member of the workspace, or change their role"),
266 a("remove_member", Op::RemoveTeamMember, "Take someone out of it"),
267 a("list_child_teams", Op::ListChildTeams, "The teams nested under it"),
268 a("list_repos", Op::ListTeamRepos, "The repositories it has a role on"),
269 a("set_repo", Op::SetTeamRepo, "Give it a role on a repository"),
270 a("remove_repo", Op::RemoveTeamRepo, "Take its role on a repository away"),
271 a("set_review_assignment", Op::SetTeamReviewAssignment, "Whom it picks when asked to review"),
272 a("list_user_teams", Op::ListUserTeams, "The teams someone is in"),
273 ],
274 },
275 Tool {
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step276 name: "workspace",
277 title: "Workspaces",
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge278 description: "Workspaces own repositories (g1t.sh/{workspace}/{repo}): create, update or delete one, invite members, connect integrations and model providers, set rulesets that hold across its repositories, and keep your own pinned projects at the top of its sidebar.",
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step279 default_action: None,
280 actions: &[
Merge branch 'worktree-agent-ad7c6d88d93adc817'281 a("get", Op::GetWorkspace, "A workspace's details and settings"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step282 a("create", Op::CreateWorkspace, "Create a workspace"),
Packages, with a container registry on g1t.sh; workspaces deleted whole and kept 30 days; Members for every member283 a("delete", Op::DeleteWorkspace, "Delete a workspace and everything in it (support can restore it for 30 days)"),
Merge branch 'worktree-agent-ad7c6d88d93adc817'284 a("update", Op::UpdateWorkspace, "Change its name, description, base permission or who may create teams"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step285 a("list_invites", Op::ListWorkspaceInvites, "Its invites"),
286 a("invite_member", Op::InviteMember, "Invite an email address"),
287 a("revoke_invite", Op::RevokeWorkspaceInvite, "Revoke a pending invite"),
288 a("list_integrations", Op::ListIntegrations, "Model providers, alert sources, trackers"),
289 a("connect_integration", Op::ConnectIntegration, "Connect one"),
AI Gateway: OpenAI's format, open models, and your own providers290 a("update_integration", Op::UpdateIntegration, "Change one: rotate its key, choose its AI Gateway models"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step291 a("disconnect_integration", Op::DisconnectIntegration, "Remove one"),
292 a("test_integration", Op::TestIntegration, "Check its credentials"),
293 a("get_model_routes", Op::GetModelRoutes, "Where each kind of work's model requests go"),
294 a("set_model_routes", Op::SetModelRoutes, "Replace them"),
API: pinned projects over REST and MCP295 a("list_pinned_projects", Op::ListPinnedProjects, "Your pinned projects in it, in your order"),
296 a("pin_project", Op::PinProject, "Pin a project, at a position or the end"),
297 a("unpin_project", Op::UnpinProject, "Unpin a project"),
298 a("reorder_pinned_projects", Op::ReorderPinnedProjects, "Put your pins in a new order"),
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge299 a("list_rulesets", Op::Rules(RulesOp::ListWorkspaceRulesets), "Its rulesets, which hold across its repositories"),
300 a("get_ruleset", Op::Rules(RulesOp::GetWorkspaceRuleset), "One of its rulesets"),
301 a("create_ruleset", Op::Rules(RulesOp::CreateWorkspaceRuleset), "Create a ruleset for some or all of its repositories"),
302 a("update_ruleset", Op::Rules(RulesOp::UpdateWorkspaceRuleset), "Change one of its rulesets"),
303 a("delete_ruleset", Op::Rules(RulesOp::DeleteWorkspaceRuleset), "Delete one of its rulesets"),
304 a("rule_evaluations", Op::Rules(RulesOp::ListWorkspaceRuleEvaluations), "How rules judged changes across its repositories"),
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step305 ],
306 },
307 Tool {
Usage, Billing settings and prepaid AI credit; fixes from the UX audit308 name: "billing",
309 title: "Billing",
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens310 description: "A workspace's billing: its usage by product, project and day, its budget (the monthly spend limit, alerts and whether usage pauses at it), its AI credit, its invoices, and its AI Gateway requests. Amounts are whole millionths of a dollar (`_micros`), or cents (`_cents`) where named. Members read it; changing the budget and buying credit are for owners, as people, and never for g1t's agents.",
Usage, Billing settings and prepaid AI credit; fixes from the UX audit311 default_action: Some("usage"),
312 actions: &[
313 a("usage", Op::GetUsage, "Usage over a range of days, by product, meter, project and day, and what paid for it"),
314 a("budget", Op::GetBudget, "The monthly spend limit, what was spent, alerts and whether usage pauses at the limit"),
315 a("set_budget", Op::SetBudget, "Change the spend limit, alerts, pausing or the alert webhook"),
316 a("ai_credit", Op::GetAiCredit, "AI credit left, its grants, auto-reload and how to buy more"),
317 a("buy_ai_credit", Op::BuyAiCredit, "A payment page to buy AI credit, for a person to open"),
318 a("invoices", Op::ListInvoices, "Every invoice, the itemised usage invoices, and the next one so far"),
319 a("billing_details", Op::GetBillingDetails, "Who invoices are made out to and the payment method on file"),
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens320 a("gateway_requests", Op::ListGatewayRequests, "Recent AI Gateway requests: model, tokens, cost, status and token"),
Usage, Billing settings and prepaid AI credit; fixes from the UX audit321 ],
322 },
323 Tool {
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar324 name: "security",
325 title: "Security",
326 description: "A repository's security: secret scanning alerts and push protection bypasses, custom secret patterns, code scanning alerts and SARIF uploads, vulnerability alerts, the dependency graph and its SBOM, dependency review, settings, and a workspace's overview. Fix an alert with g1t. Findings are shown to those who can change the code only. Give `repo` (owner/name), or `workspace` for lists across one.",
327 default_action: Some("secret_alerts"),
328 actions: &[
329 a("secret_alerts", Op::Security(SecurityOp::ListSecretAlerts), "Secret scanning alerts; by state, secret_type, validity, bypassed"),
330 a("secret_alert", Op::Security(SecurityOp::GetSecretAlert), "One secret alert, with where it was found and its bypass requests"),
331 a("update_secret_alert", Op::Security(SecurityOp::UpdateSecretAlert), "Dismiss a secret alert with a reason, or reopen it"),
332 a("secret_locations", Op::Security(SecurityOp::ListSecretLocations), "Every file, line and commit a secret is in"),
333 a("bypass", Op::Security(SecurityOp::BypassPushProtection), "Push past push protection with a reason, or ask to"),
334 a("check_validity", Op::Security(SecurityOp::CheckSecretValidity), "Ask a secret's issuer whether it still works"),
335 a("bypass_requests", Op::Security(SecurityOp::ListBypassRequests), "A workspace's push protection bypass requests"),
336 a("review_bypass", Op::Security(SecurityOp::ReviewBypassRequest), "Approve, deny or cancel a bypass request"),
337 a("patterns", Op::Security(SecurityOp::ListCustomPatterns), "Custom secret patterns of a repository or workspace"),
338 a("create_pattern", Op::Security(SecurityOp::CreateCustomPattern), "Create a custom secret pattern, as a draft or published"),
339 a("update_pattern", Op::Security(SecurityOp::UpdateCustomPattern), "Change, publish or unpublish a custom pattern"),
340 a("delete_pattern", Op::Security(SecurityOp::DeleteCustomPattern), "Delete a custom pattern"),
341 a("dry_run_pattern", Op::Security(SecurityOp::DryRunCustomPattern), "Run a pattern over the default branch without saving it"),
342 a("code_alerts", Op::Security(SecurityOp::ListCodeAlerts), "Code scanning alerts; by state, severity, tool, rule_id"),
343 a("code_alert", Op::Security(SecurityOp::GetCodeAlert), "One code scanning alert by number"),
344 a("update_code_alert", Op::Security(SecurityOp::UpdateCodeAlert), "Dismiss a code scanning alert with a reason, or reopen it"),
345 a("analyses", Op::Security(SecurityOp::ListAnalyses), "Code scanning analyses, newest first"),
346 a("upload_sarif", Op::Security(SecurityOp::UploadSarif), "Upload a SARIF file, gzipped and base64-encoded"),
347 a("sarif_upload", Op::Security(SecurityOp::GetSarifUpload), "Whether a SARIF upload was read, and its analyses"),
348 a("vulnerability_alerts", Op::Security(SecurityOp::ListVulnerabilityAlerts), "Vulnerable dependencies; by state, severity, ecosystem, package"),
349 a("vulnerability_alert", Op::Security(SecurityOp::GetVulnerabilityAlert), "One vulnerability alert"),
350 a("update_vulnerability_alert", Op::Security(SecurityOp::UpdateVulnerabilityAlert), "Dismiss a vulnerability alert with a reason, or reopen it"),
351 a("fix", Op::Security(SecurityOp::FixAlert), "Put g1t on an issue to fix an alert"),
352 a("dependency_graph", Op::Security(SecurityOp::GetDependencyGraph), "Every package the lockfiles resolve, direct or transitive"),
353 a("sbom", Op::Security(SecurityOp::GetSbom), "The dependency graph as an SPDX 2.3 document"),
354 a("compare_dependencies", Op::Security(SecurityOp::CompareDependencies), "What changes in dependencies between base...head"),
355 a("settings", Op::Security(SecurityOp::GetSettings), "A repository's security settings"),
356 a("update_settings", Op::Security(SecurityOp::UpdateSettings), "Change when checks fail and dependency review's policy"),
357 a("workspace_settings", Op::Security(SecurityOp::GetWorkspaceSettings), "A workspace's delegated bypass and validity checks"),
358 a("update_workspace_settings", Op::Security(SecurityOp::UpdateWorkspaceSettings), "Turn delegated bypass or validity checks on or off"),
359 a("overview", Op::Security(SecurityOp::GetOverview), "A workspace's alerts, trends and coverage"),
360 ],
361 },
362 Tool {
API: notifications over REST and MCP, with notifications scopes363 name: "notifications",
364 title: "Notifications",
365 description: "Your inbox: what needs you, and what you follow. One thread per issue, pull request, workflow or deployment, with why you were told (`reason`): an agent waiting on you, a review asked of you, an assignment, a mention, your work's checks, or what you subscribe to and watch. Mark threads read or done once handled, and choose what you hear of with subscribe, unsubscribe and watch. Your own: a personal token.",
366 default_action: Some("list"),
367 actions: &[
368 a("list", Op::ListNotifications, "Unread threads, latest first; all, a view, a reason, a repository"),
369 a("get", Op::GetNotificationThread, "One thread with its recent activity and your subscription"),
370 a("mark_read", Op::MarkThreadRead, "Mark a thread read, or unread"),
371 a("mark_all_read", Op::MarkNotificationsRead, "Mark everything read up to a time, or one repository's"),
372 a("done", Op::MarkThreadDone, "Mark a thread done; new activity brings it back"),
373 a("save", Op::SaveThread, "Save a thread, or unsave it"),
374 a("snooze", Op::SnoozeThread, "Snooze a thread until a time, or bring it back"),
375 a("subscription", Op::GetThreadSubscription, "Your subscription to an issue or pull request"),
376 a("subscribe", Op::SetThreadSubscription, "Subscribe to an issue or pull request, or ignore it"),
377 a("unsubscribe", Op::DeleteThreadSubscription, "Unsubscribe until you comment or are mentioned"),
378 a("watching", Op::GetRepoSubscription, "How you watch a repository"),
379 a("watch", Op::SetRepoSubscription, "Watch a repository: participating, all, ignore or custom"),
380 a("unwatch", Op::DeleteRepoSubscription, "Stop watching a repository"),
381 a("watched", Op::ListWatchedRepos, "Repositories you watch other than the default way"),
382 ],
383 },
384 Tool {
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step385 name: "account",
386 title: "Your account",
387 description: "Who this token acts as and its workspaces (`whoami`), your email addresses, your invites, and invitations to repositories waiting for you.",
388 default_action: Some("whoami"),
389 actions: &[
390 a("whoami", Op::Whoami, "Who the token acts as, and its workspaces"),
391 a("list_emails", Op::ListEmails, "Your addresses"),
392 a("add_email", Op::AddEmail, "Add an address"),
393 a("remove_email", Op::RemoveEmail, "Remove an address"),
394 a("update_email_settings", Op::UpdateEmailSettings, "Primary, backup and privacy"),
395 a("list_invites", Op::ListInvites, "Your invites to g1t"),
396 a("create_invite", Op::CreateInvite, "Make an invite"),
397 a("revoke_invite", Op::RevokeInvite, "Revoke one"),
398 a("list_repository_invitations", Op::ListMyRepoInvitations, "Invitations to repositories for you"),
399 a("accept_repository_invitation", Op::AcceptRepoInvitation, "Accept one"),
400 a("decline_repository_invitation", Op::DeclineRepoInvitation, "Decline one"),
401 ],
402 },
403];
404
405/// Operations that cannot be undone, or reach beyond g1t's own records:
406/// clients ask before running a tool that has any of them.
407fn destructive(op: Op) -> bool {
408 matches!(
409 op,
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar410 Op::Security(SecurityOp::DeleteCustomPattern | SecurityOp::BypassPushProtection)
Merge rulesets: branch and tag rules, agent-first, enforced on push and merge411 | Op::Rules(RulesOp::DeleteRepoRuleset | RulesOp::DeleteWorkspaceRuleset)
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar412 | Op::DeleteWorkspace
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily413 | Op::UpdateWorkspace
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step414 | Op::DeleteRepo
415 | Op::PurgeRepo
416 | Op::TransferRepo
417 | Op::SetRepoVisibility
418 | Op::RemoveEmail
419 | Op::RemoveCollaborator
420 | Op::DisconnectIntegration
AI Gateway: OpenAI's format, open models, and your own providers421 | Op::UpdateIntegration
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step422 | Op::DeleteWebhook
423 | Op::DeleteActionsSecret
424 | Op::DeleteActionsVariable
425 | Op::SetActionsSecret
426 | Op::SetActionsVariable
427 | Op::SetModelRoutes
428 | Op::SetBasePermission
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar429 | Op::DeleteTeam
430 | Op::RemoveTeamRepo
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step431 | Op::MergePullRequest
Fast pages, required checks on the branch, self-hosted runners, honest incidents432 | Op::RemoveRunner
433 | Op::DeleteRunnerGroup
434 | Op::UpdateRunnerSettings
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step435 )
436}
437
438/// Whether an operation only reads.
439pub fn reads_only(op: Op) -> bool {
440 NO_SCOPE.contains(&op.name())
441 || scope_for(op.name()).is_some_and(|scope| scope.level() == Level::Read)
442}
443
444/// What decides which actions a caller sees.
445pub enum Gate<'a> {
446 /// No limit beyond the person's own role.
447 Everything,
448 /// A g1t agent's token: the operations its run lists.
449 Agent(&'a AgentScope),
450 /// An access token with scopes.
451 Token(&'a TokenAccess),
452}
453
454impl Gate<'_> {
455 pub fn allows(&self, op: Op) -> bool {
456 match self {
457 Gate::Everything => true,
458 Gate::Agent(scope) => op.allowed_by(scope) && !NEVER.contains(&op.name()),
459 Gate::Token(access) => {
460 if NO_SCOPE.contains(&op.name()) {
461 return true;
462 }
463 match scope_for(op.name()) {
464 Some(scope) => access.allows(scope),
465 None => access.scopes.is_none(),
466 }
467 }
468 }
469 }
470}
471
472impl Tool {
473 pub fn by_name(name: &str) -> Option<&'static Tool> {
474 TOOLS.iter().find(|tool| tool.name == name)
475 }
476
477 pub fn action(&self, name: &str) -> Option<&'static Action> {
478 // The tools are 'static; find through TOOLS to keep the lifetime.
479 TOOLS
480 .iter()
481 .find(|tool| tool.name == self.name)
482 .and_then(|tool| tool.actions.iter().find(|action| action.name == name))
483 }
484
485 pub fn visible(&self, gate: &Gate) -> Vec<&'static Action> {
486 TOOLS
487 .iter()
488 .find(|tool| tool.name == self.name)
489 .map(|tool| tool.actions.iter().filter(|action| gate.allows(action.op)).collect())
490 .unwrap_or_default()
491 }
492
493 /// The flat input schema of the actions given.
494 pub fn input_schema(&self, actions: &[&Action]) -> Value {
495 let mut properties = Map::new();
496 let lines: Vec<String> = actions
497 .iter()
498 .map(|action| {
499 let required: Vec<String> = action.op.required();
500 if required.is_empty() {
501 format!("{}: {}.", action.name, action.summary)
502 } else {
503 format!("{} ({}): {}.", action.name, required.join(", "), action.summary)
504 }
505 })
506 .collect();
507 let mut action_schema = json!({
508 "type": "string",
509 "enum": actions.iter().map(|action| action.name).collect::<Vec<_>>(),
510 "description": lines.join("\n"),
511 });
512 if let Some(default) = self.default_action.filter(|name| actions.iter().any(|action| action.name == *name)) {
513 action_schema["default"] = json!(default);
514 }
515 properties.insert("action".to_owned(), action_schema);
516 for action in actions {
517 for (name, schema) in action.op.properties() {
518 merge_property(&mut properties, name, schema);
519 }
520 }
521 let mut required = vec![];
522 if self.default_action.is_none() {
523 required.push("action");
524 }
525 let mut schema = json!({ "type": "object", "properties": properties });
526 if !required.is_empty() {
527 schema["required"] = json!(required);
528 }
529 schema
530 }
531
532 /// The input schema keyed by action: one `oneOf` branch per action,
533 /// each with its own fields and the ones it needs.
534 pub fn discriminated(&self, actions: &[&Action]) -> Value {
535 let branches: Vec<Value> = actions
536 .iter()
537 .map(|action| {
538 let mut properties = Map::new();
539 properties.insert("action".to_owned(), json!({ "const": action.name }));
540 properties.extend(action.op.properties());
541 let mut required = vec![Value::String("action".to_owned())];
542 // The default action may leave `action` out.
543 if self.default_action == Some(action.name) {
544 required.clear();
545 }
546 required.extend(action.op.required().into_iter().map(Value::String));
547 json!({
548 "title": action.name,
549 "description": action.summary,
550 "type": "object",
551 "properties": properties,
552 "required": required,
553 })
554 })
555 .collect();
556 json!({ "type": "object", "oneOf": branches })
557 }
558
559 /// MCP's hints about the actions given: whether the tool only reads,
560 /// whether it can destroy something, and whether calling it twice is
561 /// the same as once.
562 pub fn annotations(&self, actions: &[&Action]) -> Value {
563 let read_only = actions.iter().all(|action| reads_only(action.op));
564 json!({
565 "title": self.title,
566 "readOnlyHint": read_only,
567 "destructiveHint": !read_only && actions.iter().any(|action| destructive(action.op)),
568 "idempotentHint": read_only,
569 "openWorldHint": false,
570 })
571 }
572
573 /// The tool as `tools/list` gives it, for a caller behind `gate`, or
574 /// `None` when it may use none of its actions.
575 pub fn listed(&self, gate: &Gate) -> Option<Value> {
576 let actions = self.visible(gate);
577 if actions.is_empty() {
578 return None;
579 }
580 Some(json!({
581 "name": self.name,
582 "title": self.title,
583 "description": self.description,
584 "inputSchema": self.input_schema(&actions),
585 "annotations": self.annotations(&actions),
586 }))
587 }
588}
589
590/// Adds a property to a tool's flat schema. The first action to use a name
591/// describes it; a later one with other allowed values adds them.
592fn merge_property(properties: &mut Map<String, Value>, name: String, schema: Value) {
593 match properties.get_mut(&name) {
594 None => {
595 properties.insert(name, schema);
596 }
597 Some(existing) => {
598 if let (Some(Value::Array(had)), Some(Value::Array(more))) =
599 (existing.get("enum").cloned(), schema.get("enum"))
600 {
601 let mut merged = had;
602 for value in more {
603 if !merged.contains(value) {
604 merged.push(value.clone());
605 }
606 }
607 existing["enum"] = Value::Array(merged);
608 }
609 // Different kinds of value under one name: say less, accept both.
610 if existing.get("type") != schema.get("type")
611 && let Some(fields) = existing.as_object_mut()
612 {
613 fields.remove("type");
614 fields.remove("items");
615 }
616 }
617 }
618}
619
620/// What a call to a tool runs: the operation its action names, or why not.
621pub fn resolve(tool: &Tool, arguments: &Value) -> Result<Op, String> {
622 let names = || {
623 tool.actions
624 .iter()
625 .map(|action| action.name)
626 .collect::<Vec<_>>()
627 .join(", ")
628 };
629 let Some(name) = arguments["action"].as_str().or(tool.default_action) else {
630 return Err(format!("Give an action: one of {}.", names()));
631 };
632 let Some(action) = tool.action(name) else {
633 return Err(format!("{} has no action {name}. Its actions: {}.", tool.name, names()));
634 };
635 let missing: Vec<String> = action
636 .op
637 .required()
638 .into_iter()
639 .filter(|field| arguments.get(field).is_none_or(Value::is_null))
640 .collect();
641 if !missing.is_empty() {
642 return Err(format!("{}.{name} needs {}.", tool.name, missing.join(", ")));
643 }
644 Ok(action.op)
645}
646
647#[cfg(test)]
648mod tests {
649 use super::*;
650 use g1t_contracts::scopes::{Preset, Scope};
651
652 fn listed(gate: &Gate) -> Vec<Value> {
653 TOOLS.iter().filter_map(|tool| tool.listed(gate)).collect()
654 }
655
656 fn token(scopes: Option<Vec<Scope>>) -> TokenAccess {
657 TokenAccess {
658 token_id: "tok_1".to_owned(),
659 scopes: scopes.map(|scopes| scopes.iter().map(|scope| scope.as_str().to_owned()).collect()),
660 legacy: false,
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens661 name: None,
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step662 }
663 }
664
665 #[test]
666 fn every_operation_is_exactly_one_action_of_one_tool() {
667 for op in Op::ALL {
668 let count = TOOLS
669 .iter()
670 .flat_map(|tool| tool.actions.iter())
671 .filter(|action| action.op == op)
672 .count();
673 assert_eq!(count, 1, "{} is {count} actions", op.name());
674 }
675 for tool in TOOLS {
676 let mut names = std::collections::HashSet::new();
677 for action in tool.actions {
678 assert!(names.insert(action.name), "{}.{} twice", tool.name, action.name);
679 }
680 if let Some(default) = tool.default_action {
681 assert!(tool.action(default).is_some(), "{}", tool.name);
682 }
683 }
Usage, Billing settings and prepaid AI credit; fixes from the UX audit684 assert!(TOOLS.len() <= 17, "{} tools", TOOLS.len());
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step685 }
686
687 #[test]
688 fn every_operation_needs_exactly_one_scope_or_none() {
689 use g1t_contracts::scopes::OPERATIONS;
690 for op in Op::ALL {
691 let mapped = OPERATIONS.iter().filter(|(name, _)| *name == op.name()).count();
692 let free = NO_SCOPE.contains(&op.name());
693 assert_eq!(mapped + usize::from(free), 1, "{}", op.name());
694 }
695 for (name, _) in OPERATIONS {
696 assert!(Op::by_name(name).is_some(), "{name} is not an operation");
697 }
698 }
699
700 #[test]
701 fn each_tool_schema_is_valid_with_one_branch_per_action() {
702 for tool in TOOLS {
703 let actions: Vec<&Action> = tool.actions.iter().collect();
704 let flat = tool.input_schema(&actions);
705 assert_eq!(flat["type"], "object");
706 assert!(flat.get("oneOf").is_none(), "no oneOf at the top level");
707 let listed: Vec<&str> = flat["properties"]["action"]["enum"]
708 .as_array()
709 .unwrap()
710 .iter()
711 .map(|name| name.as_str().unwrap())
712 .collect();
713 assert_eq!(listed, tool.actions.iter().map(|action| action.name).collect::<Vec<_>>());
714 for action in tool.actions {
715 for field in action.op.required() {
716 assert!(flat["properties"].get(&field).is_some(), "{}.{}: {field}", tool.name, action.name);
717 }
718 }
719 let keyed = tool.discriminated(&actions);
720 let branches = keyed["oneOf"].as_array().unwrap();
721 assert_eq!(branches.len(), tool.actions.len());
722 for (branch, action) in branches.iter().zip(tool.actions) {
723 assert_eq!(branch["properties"]["action"]["const"], action.name);
724 for field in branch["required"].as_array().unwrap() {
725 assert!(branch["properties"].get(field.as_str().unwrap()).is_some(), "{}.{}: {field}", tool.name, action.name);
726 }
727 }
728 // A well-formed JSON Schema object throughout.
729 let text = serde_json::to_string(&flat).unwrap();
730 assert!(serde_json::from_str::<Value>(&text).is_ok());
731 }
732 }
733
734 #[test]
735 fn a_read_only_token_sees_read_actions_only() {
736 let access = token(Preset::ReadOnly.scopes());
737 let gate = Gate::Token(&access);
738 for tool in TOOLS {
739 for action in tool.visible(&gate) {
740 assert!(reads_only(action.op), "{}.{}", tool.name, action.name);
741 }
742 }
743 let tools = listed(&gate);
744 for tool in &tools {
745 assert_eq!(tool["annotations"]["readOnlyHint"], true, "{}", tool["name"]);
746 assert_eq!(tool["annotations"]["destructiveHint"], false);
747 }
748 let issue = tools.iter().find(|tool| tool["name"] == "issue").unwrap();
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar749 assert_eq!(issue["inputSchema"]["properties"]["action"]["enum"], json!(["list", "get", "labels"]));
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step750 // Nothing of the agent tool is a read.
751 assert!(!tools.iter().any(|tool| tool["name"] == "agent"));
752 }
753
754 #[test]
755 fn a_narrow_token_sees_only_its_tools() {
756 let access = token(Some(vec![Scope::IssuesWrite]));
757 let names: Vec<Value> = listed(&Gate::Token(&access)).into_iter().map(|tool| tool["name"].clone()).collect();
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar758 // Labels and milestones are the repository's, managed with issues:write.
759 assert_eq!(names, vec![json!("repository"), json!("issue"), json!("plan"), json!("account")]);
API: notifications over REST and MCP, with notifications scopes760 // Notifications are a resource of their own: reading them lists
761 // only what reads.
762 let reader = token(Some(vec![Scope::NotificationsRead]));
763 let tools = listed(&Gate::Token(&reader));
764 let notifications = tools.iter().find(|tool| tool["name"] == "notifications").unwrap();
765 assert_eq!(
766 notifications["inputSchema"]["properties"]["action"]["enum"],
767 json!(["list", "get", "subscription", "watching", "watched"])
768 );
769 assert_eq!(notifications["annotations"]["readOnlyHint"], true);
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step770 let full = token(None);
771 assert_eq!(listed(&Gate::Token(&full)).len(), TOOLS.len());
772 assert_eq!(listed(&Gate::Everything).len(), TOOLS.len());
773 }
774
775 #[test]
776 fn a_tool_that_can_destroy_says_so() {
777 let tools = listed(&Gate::Everything);
778 let repository = tools.iter().find(|tool| tool["name"] == "repository").unwrap();
779 assert_eq!(repository["annotations"]["destructiveHint"], true);
780 assert_eq!(repository["annotations"]["readOnlyHint"], false);
781 let memory = tools.iter().find(|tool| tool["name"] == "memory").unwrap();
782 assert_eq!(memory["annotations"]["destructiveHint"], false);
783 }
784
785 #[test]
786 fn calls_resolve_to_their_operation_or_say_what_is_missing() {
787 let issue = Tool::by_name("issue").unwrap();
788 assert_eq!(resolve(issue, &json!({ "action": "get", "repo": "a/b", "number": 1 })), Ok(Op::GetIssue));
789 assert_eq!(resolve(issue, &json!({ "action": "get", "repo": "a/b" })), Err("issue.get needs number.".to_owned()));
790 assert!(resolve(issue, &json!({})).unwrap_err().starts_with("Give an action"));
791 assert!(resolve(issue, &json!({ "action": "explode" })).unwrap_err().contains("no action explode"));
792 let search = Tool::by_name("search").unwrap();
793 assert_eq!(resolve(search, &json!({ "query": "x" })), Ok(Op::Search));
794 let account = Tool::by_name("account").unwrap();
795 assert_eq!(resolve(account, &json!({})), Ok(Op::Whoami));
796 }
797
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar798 #[test]
799 fn teams_are_one_tool_and_a_workspace_reader_sees_only_its_reads() {
800 let team = Tool::by_name("team").unwrap();
801 let names: Vec<&str> = team.actions.iter().map(|action| action.name).collect();
802 assert_eq!(
803 names,
804 [
805 "list",
806 "get",
807 "create",
808 "update",
809 "delete",
810 "list_members",
811 "set_member",
812 "remove_member",
813 "list_child_teams",
814 "list_repos",
815 "set_repo",
816 "remove_repo",
817 "set_review_assignment",
818 "list_user_teams",
819 ]
820 );
821 let reader = token(Some(vec![Scope::WorkspaceRead]));
822 let tools = listed(&Gate::Token(&reader));
823 let listed_team = tools.iter().find(|tool| tool["name"] == "team").unwrap();
824 assert_eq!(
825 listed_team["inputSchema"]["properties"]["action"]["enum"],
826 json!(["list", "get", "list_members", "list_child_teams", "list_repos", "list_user_teams"])
827 );
828 assert_eq!(listed_team["annotations"]["readOnlyHint"], true);
829 // A team's role on a repository is who has access.
830 let admin = token(Some(vec![Scope::WorkspaceAdmin]));
831 let tools = listed(&Gate::Token(&admin));
832 let listed_team = tools.iter().find(|tool| tool["name"] == "team").unwrap();
833 let actions = listed_team["inputSchema"]["properties"]["action"]["enum"].as_array().unwrap();
834 assert!(actions.contains(&json!("set_review_assignment")) && !actions.contains(&json!("set_repo")));
835 let access = token(Some(vec![Scope::AccessAdmin]));
836 let tools = listed(&Gate::Token(&access));
837 let listed_team = tools.iter().find(|tool| tool["name"] == "team").unwrap();
838 assert_eq!(listed_team["inputSchema"]["properties"]["action"]["enum"], json!(["set_repo", "remove_repo"]));
839 // Both kinds of role a schema names are offered.
840 let roles = &listed(&Gate::Everything).into_iter().find(|tool| tool["name"] == "team").unwrap()["inputSchema"]
841 ["properties"]["role"]["enum"];
842 for role in ["member", "maintainer", "read", "admin"] {
843 assert!(roles.as_array().unwrap().contains(&json!(role)), "{role}");
844 }
845 assert_eq!(
846 resolve(team, &json!({ "action": "set_repo", "workspace": "acme", "team": "backend", "repo": "rocket" })),
847 Err("team.set_repo needs role.".to_owned())
848 );
849 }
850
851 #[test]
852 fn reviewers_and_code_owners_are_actions_of_their_tools() {
853 let pull = Tool::by_name("pull_request").unwrap();
854 assert_eq!(
855 resolve(pull, &json!({ "action": "request_reviewers", "repo": "a/b", "number": 1, "team_reviewers": ["backend"] })),
856 Ok(Op::RequestReviewers)
857 );
858 assert_eq!(pull.action("remove_requested_reviewers").map(|action| action.op), Some(Op::RemoveRequestedReviewers));
859 let repository = Tool::by_name("repository").unwrap();
860 assert_eq!(resolve(repository, &json!({ "action": "codeowners", "repo": "a/b" })), Ok(Op::GetCodeownersErrors));
861 assert!(reads_only(Op::GetCodeownersErrors));
862 assert!(!reads_only(Op::RequestReviewers));
863 }
864
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step865 /// How much smaller `tools/list` is than one tool per operation. Run
866 /// with `--nocapture` to see the numbers.
867 #[test]
868 fn the_tool_list_is_much_smaller_than_one_tool_per_operation() {
869 let before: Vec<Value> = Op::ALL
870 .into_iter()
871 .map(|op| json!({ "name": op.name(), "description": op.description(), "inputSchema": op.input() }))
872 .collect();
873 let after = listed(&Gate::Everything);
874 let before_bytes = serde_json::to_string(&json!({ "tools": before })).unwrap().len();
875 let after_bytes = serde_json::to_string(&json!({ "tools": after })).unwrap().len();
876 let agent = token(Preset::Agent.scopes());
877 let agent_bytes = serde_json::to_string(&json!({ "tools": listed(&Gate::Token(&agent)) })).unwrap().len();
878 let read = token(Preset::ReadOnly.scopes());
879 let read_bytes = serde_json::to_string(&json!({ "tools": listed(&Gate::Token(&read)) })).unwrap().len();
880 println!(
881 "tools/list: before {} tools, {before_bytes} bytes (~{} tokens); after {} tools, {after_bytes} bytes (~{} tokens); agent preset {agent_bytes} bytes (~{} tokens); read only {read_bytes} bytes (~{} tokens)",
882 before.len(),
883 before_bytes / 4,
884 after.len(),
885 after_bytes / 4,
886 agent_bytes / 4,
887 read_bytes / 4,
888 );
889 assert!(after_bytes * 2 < before_bytes, "{after_bytes} vs {before_bytes}");
890 }
891}

This file's history is long; its oldest lines are credited to the oldest commit read.