feat: document pi harness session info and token extraction in AGENTS
This commit is contained in:
2 files changed
+9
-1
No files matched your search
@@ -49,6 +49,10 @@ The `./task` file is a bash script that provides all functions necessary to work
|
||||
|
||||
Always try to run a command with the `./task` file script.
|
||||
|
||||
### Session Info (pi harness)
|
||||
|
||||
Inside the **pi** harness, the model ID is in `$PI_MODEL` (e.g. `moonshotai/Kimi-K2.6`), and the session log is at `$PI_SESSION_FILE`. That JSONL file contains token usage in the `message.usage` fields (`input` and `output`). When completing a prompt file, extract the model and cumulative token counts from these sources.
|
||||
|
||||
### Jupyter Notebooks
|
||||
|
||||
**Run with timeout**: Run the notebook with `timeout 10 task run-jupyter-notebook <path> [env]`.
|
||||
|
||||
+5
-1
@@ -26,7 +26,11 @@ The `./task` file is a bash script that provides all functions necessary to work
|
||||
|
||||
Always try to run a command with the `./task` file script.
|
||||
|
||||
### New Tools
|
||||
### Session Info (pi harness)
|
||||
|
||||
Inside the **pi** harness, the model ID is in `$PI_MODEL` (e.g. `moonshotai/Kimi-K2.6`), and the session log is at `$PI_SESSION_FILE`. That JSONL file contains token usage in the `message.usage` fields (`input` and `output`). When completing a prompt file, extract the model and cumulative token counts from these sources.
|
||||
|
||||
### New tools
|
||||
|
||||
Do not automatically install new software packages. Always ask for permission.
|
||||
|
||||
|
||||
Reference in new issue
Block a user