Kiro Product
CLI
Feature Description
When invoking a skill via slash command (e.g. /autoresearch, /browser-use, /find-skills), Kiro currently injects the skill's SKILL.md content as the user's message and sends it immediately. The agent receives the skill reference doc as the entire message body with no actual task or instructions attached.
This means the agent has no idea what the user wants to do — it just sees documentation dumped into the chat.
Use Case
Users should be able to invoke a skill AND provide instructions in the same message, e.g.:
/autoresearch fix all lint errors/browser-use open https://example.com and take a screenshot/find-skills playwright
Alternatively, if no instructions are provided after the slash command, Kiro should either:
- Prompt the user for instructions before sending, or
- Send the skill context but also signal to the agent that the user invoked the skill intentionally (so the agent can ask what they want to do with it)
Currently, repeatedly invoking skills just floods the chat with reference docs and the agent keeps asking "what do you want me to do?"
Additional Context
Observed with multiple skills: autoresearch, find-skills, browser-use. Each was invoked via slash command multiple times, and each time only the SKILL.md content was sent as the message with no user intent.
Kiro Product
CLI
Feature Description
When invoking a skill via slash command (e.g.
/autoresearch,/browser-use,/find-skills), Kiro currently injects the skill's SKILL.md content as the user's message and sends it immediately. The agent receives the skill reference doc as the entire message body with no actual task or instructions attached.This means the agent has no idea what the user wants to do — it just sees documentation dumped into the chat.
Use Case
Users should be able to invoke a skill AND provide instructions in the same message, e.g.:
/autoresearch fix all lint errors/browser-use open https://example.com and take a screenshot/find-skills playwrightAlternatively, if no instructions are provided after the slash command, Kiro should either:
Currently, repeatedly invoking skills just floods the chat with reference docs and the agent keeps asking "what do you want me to do?"
Additional Context
Observed with multiple skills: autoresearch, find-skills, browser-use. Each was invoked via slash command multiple times, and each time only the SKILL.md content was sent as the message with no user intent.