You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: AGENTS.md
+45Lines changed: 45 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -285,6 +285,49 @@ schedule:
285
285
- release/*
286
286
```
287
287
288
+
### Engine Configuration
289
+
290
+
The `engine` field specifies which AI model to use and optional execution parameters. It accepts both a simple string format (model name only) and an object format with additional options.
291
+
292
+
```yaml
293
+
# Simple string format (just a model name)
294
+
engine: claude-opus-4.5
295
+
296
+
# Object format with additional options
297
+
engine:
298
+
model: claude-opus-4.5
299
+
max-turns: 50
300
+
timeout-minutes: 30
301
+
```
302
+
303
+
#### Fields
304
+
305
+
| Field | Type | Default | Description |
306
+
|-------|------|---------|-------------|
307
+
| `model` | string | `claude-opus-4.5` | AI model to use. Options include `claude-sonnet-4.5`, `gpt-5.2-codex`, `gemini-3-pro-preview`, etc. |
308
+
| `max-turns` | integer | *(none)* | Maximum number of agentic turns (tool-use iterations) the model is allowed per run. Maps to the `--max-turns` Copilot CLI argument. Use this to cap compute and prevent runaway loops. |
309
+
| `timeout-minutes` | integer | *(none)* | Maximum time in minutes the agent workflow is allowed to run. Maps to the `--max-timeout` Copilot CLI argument. Use this to cap long-running agent sessions. |
310
+
311
+
#### `max-turns`
312
+
313
+
Each "turn" is one iteration of the model calling a tool and receiving its output. Setting `max-turns` places an upper bound on how many such iterations the agent can perform in a single pipeline run. This is useful for:
314
+
315
+
- **Cost control** — limiting expensive model invocations.
316
+
- **Safety** — preventing infinite loops where the agent repeatedly calls tools without converging on a result.
317
+
- **Predictability** — ensuring the pipeline completes within a reasonable time frame.
318
+
319
+
When omitted, the Copilot CLI uses its built-in default. When set, the compiler emits `--max-turns <value>` in the generated pipeline's copilot params.
320
+
321
+
#### `timeout-minutes`
322
+
323
+
The `timeout-minutes` field sets a wall-clock limit (in minutes) for the entire agent session. It maps to the `--max-timeout` Copilot CLI argument. This is useful for:
324
+
325
+
- **Budget enforcement** — hard-capping the total runtime of an agent to control compute costs.
326
+
- **Pipeline hygiene** — preventing agents from occupying a runner indefinitely if they stall or enter long retry loops.
327
+
- **SLA compliance** — ensuring scheduled agents complete within a known window.
328
+
329
+
When omitted, the Copilot CLI uses its built-in default. When set, the compiler emits `--max-timeout <value>` in the generated pipeline's copilot params.
330
+
288
331
### Tools Configuration
289
332
290
333
The `tools` field controls which tools are available to the agent. Both sub-fields are optional and have sensible defaults.
@@ -439,6 +482,8 @@ Should be replaced with the human-readable name from the front matter (e.g., "Da
439
482
440
483
Additional params provided to agency CLI. The compiler generates:
441
484
- `--model <model>` - AI model from `engine` front matter field (default: claude-opus-4.5)
485
+
- `--max-turns <n>`- Maximum agentic turns from `engine.max-turns` (omitted when not set)
486
+
- `--max-timeout <n>`- Workflow timeout in minutes from `engine.timeout-minutes` (omitted when not set)
442
487
- `--disable-builtin-mcps`- Disables all built-in MCPs initially
0 commit comments