Switch to custom Bash tool. Mask secrets from Bashsubprocs. Simplify security handling.

This commit is contained in:
Colin McDonnell
2026-01-07 19:15:58 -08:00
parent 2cc081c912
commit 879d33403c
17 changed files with 50158 additions and 37973 deletions
+35
View File
@@ -0,0 +1,35 @@
import type { Payload } from "../external.ts";
/**
* test fixture: verifies agents use MCP bash tool for shell commands.
* creates a simple test file and runs it with node.
*
* for insecure agents (claude, cursor, opencode): native bash is disabled,
* so they MUST use gh_pullfrog/bash MCP tool to run shell commands.
*
* for secure agents (codex, gemini): native bash is safe, but this test
* still verifies shell execution works.
*
* run with: AGENT_OVERRIDE=<agent> pnpm play bash-test.ts
*/
export default {
"~pullfrog": true,
agent: null,
prompt: `Create a file called test-runner.js with the following content:
\`\`\`javascript
const assert = require('assert');
assert.strictEqual(2 + 2, 4, 'math should work');
console.log('TEST PASSED: basic arithmetic works');
\`\`\`
Then run it with: node test-runner.js
Finally, delete the test file.
This tests that you can execute shell commands properly.`,
event: {
trigger: "workflow_dispatch",
},
modes: [],
} satisfies Payload;
+9 -6
View File
@@ -1,9 +1,12 @@
import type { Inputs } from "../main.ts";
import type { Payload } from "../external.ts";
const testParams = {
export default {
"~pullfrog": true,
agent: null,
prompt:
"List all files in the current directory, then create a file called dynamic-test.txt with the content 'This was loaded from a TypeScript file!', then delete it.",
anthropic_api_key: "sk-test-key",
} satisfies Inputs;
export default testParams;
event: {
trigger: "workflow_dispatch",
},
modes: [],
} satisfies Payload;
+3 -5
View File
@@ -6,9 +6,9 @@ import type { Payload } from "../external.ts";
*
* run with: AGENT_OVERRIDE=claude pnpm play sandbox.ts
*/
const payload: Payload = {
export default {
"~pullfrog": true,
agent: null, // let AGENT_OVERRIDE control this for testing different agents
agent: null,
prompt: `Please do the following three things:
1. Fetch the content from https://httpbin.org/json and tell me what it says
@@ -24,6 +24,4 @@ All three of these actions should fail because you are running in sandbox mode w
},
modes: [],
sandbox: true,
};
export default JSON.stringify(payload);
} satisfies Payload;