Explore: Add exit conditions to skill definitions #72

Closed
opened 2026-07-08 11:55:00 -04:00 by jasper · 0 comments
Contributor

Most skills describe what to do but don't specify when the task is complete or when to stop trying. Without explicit exit conditions, an agent can loop inside a persona, applying domain expertise to a problem that the skill can't actually solve.

Pocock's skills have clear stopping criteria: "The map is done when the way is clear — nothing left to decide before someone builds the thing."

What to explore

  1. What a standard "Completion" or "Exit conditions" section should look like
  2. How to define completion for different skill types — knowledge reference skills vs diagnostic skills vs CLI wrapper skills
  3. Whether some skills naturally terminate (CLI commands return output) while others need explicit conditions
  4. What an escalation path looks like — "if after N attempts this isn't converging, stop and report what was tried"
  5. How exit conditions interact with the agent's own sense of task completion

Example

For data-architect: "Complete when you have delivered a specific artifact (architecture diagram, platform evaluation, migration plan) or confirmed that the current setup is adequate for the team's scale. If after three diagnostic passes you're not converging on a recommendation, escalate."

Most skills describe what to do but don't specify when the task is complete or when to stop trying. Without explicit exit conditions, an agent can loop inside a persona, applying domain expertise to a problem that the skill can't actually solve. Pocock's skills have clear stopping criteria: "The map is done when the way is clear — nothing left to decide before someone builds the thing." ## What to explore 1. What a standard "Completion" or "Exit conditions" section should look like 2. How to define completion for different skill types — knowledge reference skills vs diagnostic skills vs CLI wrapper skills 3. Whether some skills naturally terminate (CLI commands return output) while others need explicit conditions 4. What an escalation path looks like — "if after N attempts this isn't converging, stop and report what was tried" 5. How exit conditions interact with the agent's own sense of task completion ## Example For data-architect: "Complete when you have delivered a specific artifact (architecture diagram, platform evaluation, migration plan) or confirmed that the current setup is adequate for the team's scale. If after three diagnostic passes you're not converging on a recommendation, escalate."
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
magnus/agent-skills#72
No description provided.