Skip to content
Sign up

flagon-io/g1t

public
00

Where people and agents ship software together. The open-source git platform for the whole job: issues, agents, checks and deploys to the edge.

g1t.sh
gitagentscloudflarerustforgemcp

Commit

Sessions g1t starts on Claude Managed Agents can search and read the web: the agent's toolset turned web_search and web_fetch off, and its sandbox reaches only g1t, so on flagon-io/demo #1 the first session couldn't check Cloudflare Connect's date on Cloudflare's page as the issue asked, and stopped. Both tools are on now, with every other tool, under the same auto permission policy; the token stays safe, since the sandbox holds only a placeholder swapped in on the way to g1t. Agents already on accounts take it at their next session (config_hash). The sessions guide says so. Checked: cargo test -p g1t-integrations (76).

G1t-Session:
ses_01m4z09g22e7ysqpg2ha423cws
syntaqxcommitted Parentcf8330cBrowse files
2 files+5−60/2 viewed
+2−1
290290 commits yet, where there is nothing to open one against: it pushes the
291291 first version straight to the default branch.
292292 g1t's MCP server is attached too, for anything the CLI has no command
293− for. Both reach g1t with a token made for you
293+ for. It can search and read the web (web search and web fetch), for a
294+ fact the work needs from outside g1t. Both reach g1t with a token made for you
294295 and that repository only, which lasts the session and is revoked when it
295296 ends. The token is kept in a vault on the Anthropic account: the MCP
296297 server's credential, and `G1T_TOKEN`, which the sandbox holds only as a
+3−5
137137 "tools": [
138138 {
139139 "type": "agent_toolset_20260401",
140+ // Every tool, web_search and web_fetch too: work often needs a
141+ // fact from outside g1t (a date, an API's docs).
140142 "default_config": { "permission_policy": { "type": "auto" } },
141− "configs": [
142− { "name": "web_fetch", "enabled": false },
143− { "name": "web_search", "enabled": false },
144− ],
145143 },
146144 { "type": "mcp_toolset", "mcp_server_name": "g1t", "default_config": { "permission_policy": { "type": "always_allow" } } },
147145 ],
15511549 assert_eq!(body["mcp_servers"][0], json!({ "type": "url", "name": "g1t", "url": "https://mcp.g1t.sh" }));
15521550 let tools = body["tools"].as_array().unwrap();
15531551 assert_eq!(tools[0]["type"], "agent_toolset_20260401");
1554− assert!(tools[0]["configs"].as_array().unwrap().iter().all(|config| config["enabled"] == false));
1552+ assert!(tools[0].get("configs").is_none(), "web_search and web_fetch are on, as every tool is");
15551553 assert_eq!(tools[1], json!({ "type": "mcp_toolset", "mcp_server_name": "g1t", "default_config": { "permission_policy": { "type": "always_allow" } } }));
15561554 assert_eq!(body["skills"], json!(skills));
15571555 assert_eq!(body["multiagent"]["agents"][0]["type"], "self");

Comments

Sign in to comment.