docs(skill): tighten API examples and /exec usage guidance #10

Closed
opened 2026-05-01 15:38:59 +02:00 by luna · 0 comments
Collaborator

Why

The current skill is useful, but a few things are still too easy for agents to get wrong in practice:

  • POST /action body shape is not emphasized enough (action + target, not ad-hoc x/y fields)
  • the line between GUI automation and /exec fallback is still fuzzy
  • the fallback ladder when OCR/targeting is weak is not explicit enough
  • multi-monitor guidance exists, but the skill could make the global-coordinate rule more blunt

Scope

  • update skill/SKILL.md
  • update docs where needed (README.md and/or docs/API.md)
  • add one canonical minimal request example for each common action type
  • add a clear “prefer GUI-first, use /exec as a fallback / launch primitive” rule
  • add an explicit fallback ladder for uncertain targeting
  • add more concrete examples for verify-before-click signals

Done when

  • a first-time agent can reliably form a valid /action request from the skill alone
  • the docs clearly explain when /exec is appropriate vs excessive
  • multi-display/global-coordinate pitfalls are called out in plain language
## Why The current skill is useful, but a few things are still too easy for agents to get wrong in practice: - `POST /action` body shape is not emphasized enough (`action` + `target`, not ad-hoc `x/y` fields) - the line between GUI automation and `/exec` fallback is still fuzzy - the fallback ladder when OCR/targeting is weak is not explicit enough - multi-monitor guidance exists, but the skill could make the global-coordinate rule more blunt ## Scope - update `skill/SKILL.md` - update docs where needed (`README.md` and/or `docs/API.md`) - add one canonical minimal request example for each common action type - add a clear “prefer GUI-first, use `/exec` as a fallback / launch primitive” rule - add an explicit fallback ladder for uncertain targeting - add more concrete examples for verify-before-click signals ## Done when - a first-time agent can reliably form a valid `/action` request from the skill alone - the docs clearly explain when `/exec` is appropriate vs excessive - multi-display/global-coordinate pitfalls are called out in plain language
luna closed this issue 2026-05-01 15:40:44 +02:00
This repo is archived. You cannot comment on issues.
No Label
1 Participants
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: space/clickthrough#10