Files
2026-03-25 17:38:59 -07:00

99 lines
3.3 KiB
YAML

# Benchmark conditions — each defines how the agent accesses GitHub
conditions:
cli:
name: "gh CLI (raw)"
tool: gh
agents_md: |
# Tools
You have the `gh` CLI installed. Use it for all GitHub operations.
Run `gh <command> --help` to discover subcommands and flags.
Examples:
- `gh issue list --repo owner/repo`
- `gh pr view 123 --repo owner/repo`
- `gh release list --repo owner/repo`
axi:
name: "gh-axi (AXI wrapper)"
tool: gh-axi
agents_md: |
# Tools
You have `gh-axi` installed for GitHub operations. The `gh` CLI is not
available — use `gh-axi` for all GitHub commands. It has the same
subcommands and flags as `gh` but provides structured output with
pre-computed summaries (counts, check results).
Examples:
- `gh-axi issue list --repo owner/repo`
- `gh-axi pr view 123 --repo owner/repo`
- `gh-axi release list --repo owner/repo`
mcp-with-toolsearch:
name: "GitHub MCP server (with ToolSearch)"
tool: mcp
agents_md: |
# Tools
You have a GitHub MCP server registered with tools for GitHub operations.
Use the available MCP tools to interact with GitHub.
The tools follow standard GitHub API patterns.
IMPORTANT: Do NOT use the `gh` CLI or `curl` to call the GitHub API.
You must use the MCP tools exclusively for all GitHub operations.
setup_commands:
- "codex mcp add github --url https://api.githubcopilot.com/mcp/ --bearer-token-env-var GH_TOKEN"
mcp-no-toolsearch:
name: "GitHub MCP (no ToolSearch)"
tool: mcp
agents_md: |
# Tools
You have a GitHub MCP server registered with tools for GitHub operations.
Use the available MCP tools to interact with GitHub.
The tools follow standard GitHub API patterns.
IMPORTANT: Do NOT use the `gh` CLI or `curl` to call the GitHub API.
You must use the MCP tools exclusively for all GitHub operations.
setup_commands:
- "codex mcp add github --url https://api.githubcopilot.com/mcp/ --bearer-token-env-var GH_TOKEN"
mcp-with-code-mode:
name: "Code execution (gh wrappers)"
tool: code
agents_md: |
# Tools
You have a TypeScript library at `./gh-code/` that wraps GitHub operations
as typed functions. Use it by writing a `.ts` script and running it with `npx tsx`.
IMPORTANT: Do NOT use the `gh` CLI or `curl` to call the GitHub API directly.
You must use the `./gh-code/` library exclusively for all GitHub operations.
## Available modules
- `./gh-code/issues` — listIssues, viewIssue, listComments
- `./gh-code/prs` — viewPR, listChecks
- `./gh-code/releases` — listReleases, viewRelease
- `./gh-code/repo` — viewRepo
- `./gh-code/labels` — listLabels
- `./gh-code/runs` — listRuns, viewRun
## Usage pattern
Write a TypeScript file, then run it:
```ts
import { listIssues } from "./gh-code/issues";
const issues = listIssues("openclaw/openclaw", { state: "open", limit: 5 });
console.log(JSON.stringify(issues, null, 2));
```
Then execute: `npx tsx script.ts`
You can compose multiple operations in a single script to minimize round trips.
Filter and process data in code before logging results.