docs: instruct agents to cite scraper runs

This commit is contained in:
CREDO23 2026-07-22 19:34:56 +02:00
parent 95b4afddeb
commit dd5d267056
2 changed files with 4 additions and 2 deletions

View file

@ -1,8 +1,9 @@
<citations>
Cite with one token: the bracket label `[n]`. Every citable result —
prose from a `task` knowledge_base/research specialist (including the
knowledge_base specialist's `[n]`-labelled workspace findings) — already
carries `[n]` labels on a single shared count.
knowledge_base specialist's `[n]`-labelled workspace findings) and
scraper specialists' run-backed findings — already carries `[n]` labels
on a single shared count.
Those labels are the only citation you write; the server resolves each one
back to its source after the turn.

View file

@ -5,3 +5,4 @@ Rules (universal):
- `status=blocked` due to missing required inputs -> `missing_fields` must be non-null.
- `assumptions`: any inferences you made about the user's intent; `null` when no inferences were needed.
- The `evidence` object's fields are documented in your route-specific `<output_contract>` above; never invent fields the tool did not return.
- When a finding is drawn from a scraper run, append that run's `[n]` (the tool result states `Cite this scraper run as [n]`) to the finding text so the citation survives into the final answer. Copy the label exactly; never invent one.