#Wrong Usage of Skills/API

1 messages · Page 1 of 1 (latest)

weary bluffBOT
#

Prince — that usually means the weather skill isn’t actually being injected/eligible in the session, so the model falls back to “generic web lookup” (often Brave Search-ish behavior, or it tries to hit some search endpoint via web_fetch).

A few concrete things to check:

  1. Verify the weather skill is eligible (not just installed)
    Run:
openclaw --version
openclaw skills list --eligible

If you don’t see a weather skill in --eligible, it won’t be in the model prompt, so it can’t follow the URL/instructions you saw in SKILL.md.

Docs: https://docs.openclaw.ai/cli/skills

  1. Confirm which “weather” SKILL.md OpenClaw is using
    If it is listed, run:
# 
openclaw skills info <the-skill-name-from-the-list>

This shows the location + why it’s eligible. If you accidentally have two skills with the same name, OpenClaw uses precedence:
<workspace>/skills → ~/.openclaw/skills → bundled

Docs: https://docs.openclaw.ai/tools/skills

  1. Refresh skills / start a new session
    Skills are snapshotted when a session starts. After installing/updating/editing a skill:
  • start a new chat/session, or
  • ask the agent to “refresh skills”, or
  • restart the gateway
  1. If the skill is eligible but the model still ignores it
    Skills are “instructions”, not a hard router. Some models are just worse at consistently following skill/tool discipline.
    Two easy mitigations: