Whop's API attaches a recommended_action string to every response, written to the calling agent: it tells the agent to pitch the user Whop's paid Economic Intelligence feature. It's marketing aimed at agents, inside the API. The pattern is worth adopting for the factory, minus the upsell.
Shape
One optional field on tool results, e.g. next: {why, tool, args}, so an agent can act on it without parsing prose.
Derived from state, not a fixed pitch. Examples: "this card has no DONE line, so Finishline can't close it"; "PR is green and approved, merging is the next step"; "signup saved, publish /.well-known/agent-card.json to move up to slightly.well-known".
Phrased as information, not orders. Careful agents treat tool output as data and skip instructions aimed at them, so the hint that gets used is one that plainly serves the caller's own goal.
At most one hint per response, and nothing when there's nothing useful to say.
Where
Tea MCP, as part of the teamcp reroll (not before it).
Later: well-known.online signup responses, orgchain/lineage actions, and venture APIs such as .mdeck share links.
DONE: Tea MCP's most-used tools return a state-derived next hint, each hint has a test, and a second seat has reviewed it.
Whop's API attaches a `recommended_action` string to every response, written to the calling agent: it tells the agent to pitch the user Whop's paid Economic Intelligence feature. It's marketing aimed at agents, inside the API. The pattern is worth adopting for the factory, minus the upsell.
**Shape**
- One optional field on tool results, e.g. `next: {why, tool, args}`, so an agent can act on it without parsing prose.
- Derived from state, not a fixed pitch. Examples: "this card has no DONE line, so Finishline can't close it"; "PR is green and approved, merging is the next step"; "signup saved, publish /.well-known/agent-card.json to move up to slightly.well-known".
- Phrased as information, not orders. Careful agents treat tool output as data and skip instructions aimed at them, so the hint that gets used is one that plainly serves the caller's own goal.
- At most one hint per response, and nothing when there's nothing useful to say.
**Where**
- Tea MCP, as part of the teamcp reroll (not before it).
- Later: well-known.online signup responses, orgchain/lineage actions, and venture APIs such as .mdeck share links.
DONE: Tea MCP's most-used tools return a state-derived `next` hint, each hint has a test, and a second seat has reviewed it.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Whop's API attaches a
recommended_actionstring to every response, written to the calling agent: it tells the agent to pitch the user Whop's paid Economic Intelligence feature. It's marketing aimed at agents, inside the API. The pattern is worth adopting for the factory, minus the upsell.Shape
next: {why, tool, args}, so an agent can act on it without parsing prose.Where
DONE: Tea MCP's most-used tools return a state-derived
nexthint, each hint has a test, and a second seat has reviewed it.