playbook_start_task
playbook_start_task(project: str, intent: str)Read-only. The call an agent makes before it writes anything.
Arguments
Section titled “Arguments”| Argument | Notes |
|---|---|
project |
The basename of the workspace directory. Case-insensitive. |
intent |
What you are about to do, in your own words. This selects the workflow. |
What comes back
Section titled “What comes back”- The project’s guardrails.
- The workflow matching
intent- a bug fix and a new feature do not get the same one. - The definition of done.
- A short list of refs worth reading next.
Deliberately not the whole corpus. A project is around 25 documents; returning all of them would cost a context window to answer a question that needed three files.
On project
Section titled “On project”This is the one argument worth getting right. Working in /home/you/code/billing-api means
project="billing-api". A wrong name returns another team’s standards, which is worse than
returning nothing at all.
If the project does not exist, the response says so and names
playbook_scaffold_standards rather than returning an
empty set.
Implementation note
Section titled “Implementation note”The guardrails and workflow bodies are rendered by the same function
playbook_get_standard uses, with a byte-identity test holding the two together. That
duplication had crept back twice before the test existed, because earlier passes merged tool
names without merging their renderers.