Bug type

Behavior bug (incorrect output/state without crash)

Summary

Description

Summary

Existing agent sessions keep using an outdated skill list (skillsSnapshot) even after new skills are installed and available globally.

Expected behavior

When a new skill is installed and becomes eligible, existing sessions should either:

• refresh their skill snapshot automatically (as documented for watcher-based refresh), or
• provide an official CLI command to refresh/rebuild one session snapshot.

Impact

• Agents cannot reliably discover newly installed skills in ongoing sessions.
• Operational confusion: global skills and per-session visible skills diverge.
• Forces session rotation/manual workarounds.

Steps to reproduce

Reproduction steps

1. Start an agent session (e.g. agent:asterix:main).
2. Install a new skill globally (example: kanban-tasks).
3. Verify installation with openclaw skills (new skill is visible).
4. Run agent again in the same existing session.
5. Observe the session/system skill list is still stale (new skill missing).
6. Run openclaw sessions cleanup.
7. Run agent again; snapshot remains stale.

Expected behavior

Requested fix

• Invalidate/rebuild session skillsSnapshot when skill inventory changes (new/removed/updated skills), not only on selected file change cases.
• Or provide an explicit official CLI command to refresh one session snapshot safely.

Actual behavior

Actual behavior

• openclaw skills shows the new skill (e.g. kanban-tasks) is installed.
• Agents in existing sessions still see only the old skill list in their system context.
• openclaw sessions cleanup does not refresh/rebuild the stale session snapshot.
• No documented openclaw sessions refresh / rebuild command exists.

OpenClaw version

2026.3.12 (6472949)

Operating system

Linux

Install method

npm

Model

gpt5.4

Provider / routing chain

Provider / routing chain — Effective request path through gateways, proxies, providers, or model routers.

Config file / key location

./openclaw/openclaw.json

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response