Issue Station
Sign In
Tools hereToolbox

openclaw - 💡(How to fix) Fix [Bug] External skills are ready but not injected into multi-agent specialist runtime in v2026.3.24 [3 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
openclaw/openclaw#55479•Fetched 2026-04-08 01:39:07
View on GitHub
Comments
3
Participants
2
Timeline
5
Reactions
0
Author
IanYang868IanYang868
Participants
EronFanEronFanIanYang868IanYang868
Timeline (top)
commented ×3labeled ×2

In v2026.3.24, external skills marked ready are not injected into specialist agents’ runtime prompt in multi-agent setups.

Root Cause

Fresh session IDs were also tested, so this does not appear to be caused by session reuse.

Code Example

Evidence 1: `openclaw skills list`
The external skills were recognized as ready by `openclaw skills list`, including:
- summarize
- web-content-fetcher
- web-search

Evidence 2: specialist runtime JSON
In specialist agent JSON output, `systemPromptReport.skills.entries` only contained bundled skills:
- clawhub
- coding-agent
- gh-issues
- github
- healthcheck
- mcporter
- node-connect
- skill-creator
- weather

The external skills did not appear:
- engram
- summarize
- web-content-fetcher
- web-search

Evidence 3: tested both shared and agent-local paths
I installed the same external skills under:
- `~/.openclaw/skills`
- and each agent’s own `<workspace>/skills`

Both were recognized on disk / by `openclaw skills list`, but specialist runtime still only showed bundled skills.
RAW_BUFFERClick to expand / collapse

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

In v2026.3.24, external skills marked ready are not injected into specialist agents’ runtime prompt in multi-agent setups.

Steps to reproduce

  1. Install OpenClaw v2026.3.24 and create a multi-agent setup with main plus at least one specialist agent (for example investor).
  2. Install external skills such as engram, summarize, web-content-fetcher, and web-search under ~/.openclaw/skills.
  3. Also install the same skills under the specialist agent workspace, e.g. ~/.openclaw/workspace-investor/skills.
  4. Run openclaw skills list and confirm those external skills are shown as ready.
  5. Run a fresh-session local test such as: openclaw agent --agent investor --local --session-id repro-skill-investor-1 --message "Only list the currently visible skill names." --json
  6. Inspect systemPromptReport.skills.entries in the JSON output.

Expected behavior

The external skills should appear in systemPromptReport.skills.entries for the specialist agent when they are installed under either ~/.openclaw/skills or the agent’s own <workspace>/skills.

Actual behavior

openclaw skills list shows the external skills as ready, but the specialist agent runtime only receives bundled skills in systemPromptReport.skills.entries. The external skills (engram, summarize, web-content-fetcher, web-search) do not appear.

OpenClaw version

2026.3.24

Operating system

macOS 15.4

Install method

npm global

Model

openai/gpt-5.4

Provider / routing chain

openclaw -> PackyAPI -> openai/gpt-5.4

Additional provider/model setup details

Model/auth was working normally through PackyAPI. main and specialist agents could successfully call openai/gpt-5.4, so this does not appear to be an auth, provider, or network failure. The problem only affected external skill injection into specialist runtime.

Logs, screenshots, and evidence

Evidence 1: `openclaw skills list`
The external skills were recognized as ready by `openclaw skills list`, including:
- summarize
- web-content-fetcher
- web-search

Evidence 2: specialist runtime JSON
In specialist agent JSON output, `systemPromptReport.skills.entries` only contained bundled skills:
- clawhub
- coding-agent
- gh-issues
- github
- healthcheck
- mcporter
- node-connect
- skill-creator
- weather

The external skills did not appear:
- engram
- summarize
- web-content-fetcher
- web-search

Evidence 3: tested both shared and agent-local paths
I installed the same external skills under:
- `~/.openclaw/skills`
- and each agent’s own `<workspace>/skills`

Both were recognized on disk / by `openclaw skills list`, but specialist runtime still only showed bundled skills.

Impact and severity

Affected: OpenClaw users running multi-agent setups that rely on external skills for specialist agents. Severity: High (blocks workflows that depend on external skills) Frequency: Reproduced in all tested specialist agents and fresh sessions Consequence: Specialist agents cannot access installed external skills even though those skills are recognized as ready by openclaw skills list

Additional information

External skills were tested in both locations documented for skills loading:

  • ~/.openclaw/skills
  • each agent's own <workspace>/skills

In both cases, openclaw skills list recognized the skills as ready, but specialist runtime still only showed bundled skills.

Fresh session IDs were also tested, so this does not appear to be caused by session reuse.

extent analysis

Fix Plan

To resolve the issue of external skills not being injected into specialist agents' runtime prompt in multi-agent setups, follow these steps:

  • Update the openclaw configuration to include the external skills directory for each specialist agent.
  • Modify the openclaw code to load external skills from both the shared ~/.openclaw/skills directory and the agent's own <workspace>/skills directory.

Example code changes:

# In openclaw/config.py
EXTERNAL_SKILLS_DIRS = [
    '~/.openclaw/skills',
    '<workspace>/skills'  # Add this line
]

# In openclaw/skills.py
def load_external_skills(agent_workspace):
    skills = []
    for dir in EXTERNAL_SKILLS_DIRS:
        skills_dir = os.path.join(agent_workspace, dir)
        if os.path.exists(skills_dir):
            skills.extend(load_skills_from_dir(skills_dir))
    return skills
  • Update the openclaw agent command to pass the agent's workspace to the load_external_skills function.
# In openclaw/agent.py
def run_agent(agent_name, workspace, **kwargs):
    # ...
    external_skills = load_external_skills(workspace)
    # ...

Verification

To verify the fix, run the following steps:

  • Install the external skills in both ~/.openclaw/skills and the specialist agent's own <workspace>/skills directory.
  • Run openclaw skills list to confirm the external skills are recognized as ready.
  • Run a fresh-session local test using the openclaw agent command, and inspect the systemPromptReport.skills.entries in the JSON output to ensure the external skills are included.

Extra Tips

  • Make sure to update the openclaw configuration and code to reflect the changes.
  • Test the fix with multiple specialist agents and external skills to ensure the issue is fully resolved.
  • Consider adding logging or debugging statements to the load_external_skills function to help diagnose any future issues.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

The external skills should appear in systemPromptReport.skills.entries for the specialist agent when they are installed under either ~/.openclaw/skills or the agent’s own <workspace>/skills.

#api#installation#tensor shape#autograd error#agent setup

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING