Model Registry Manager

v1.3.0

Detect provider models, deduplicate them, remove unusable ones, register missing models into OpenClaw, and safely keep provider-native model ids/names during...

⭐ 0· 123·0 current·0 all-time

byQihong@zqh2333

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for zqh2333/model-registry-manager.

Previewing Install & Setup.

Prompt PreviewInstall & Setup

Install the skill "Model Registry Manager" (zqh2333/model-registry-manager) from ClawHub.
Skill page: https://clawhub.ai/zqh2333/model-registry-manager
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install model-registry-manager

ClawHub CLI

Package manager switcher

npx clawhub@latest install model-registry-manager

Security Scan

VirusTotal

Suspicious

View report →

OpenClaw

Benign

high confidence

✓

Purpose & Capability

Name/description match the implementation: scripts fetch /models from provider.baseUrl, deduplicate by provider id, probe models, update agents.defaults.models in ~/.openclaw/openclaw.json, and write reports. No unrelated credentials or tools are requested.

ℹ

Instruction Scope

SKILL.md instructs the agent to inspect config, fetch remote models, validate before writing, schedule recurring jobs, and maintain a `.learnings/` archive automatically. The provided scripts implement the core fetch/deduplicate/probe/register/report behavior and write reports under the skill's reports path, but they do not implement the automatic `.learnings/` archival behavior described in SKILL.md (minor mismatch). The instructions also recommend scheduling jobs and allowing exec/read for those jobs; the sync script can run 'openclaw gateway restart' when --restart is passed, which is consistent with the recommendation but grants the skill permission to execute a restart command.

✓

Install Mechanism

No install spec — instruction-only with two included Node scripts. Nothing is downloaded from external URLs or written during an install step.

✓

Credentials

No required env vars are declared. The scripts read provider.apiKey entries from the OpenClaw config file (cfg.models.providers[].apiKey) to call provider endpoints; that is expected for this purpose and no unrelated secrets are requested.

ℹ

Persistence & Privilege

always:false (good). The skill intentionally modifies the OpenClaw config file (~/.openclaw/openclaw.json) and can run 'openclaw gateway restart' when asked; these are within the skill's purpose but are privileged actions (configuration writes and a restart). Autonomous invocation is allowed by default (not flagged on its own).

Assessment

This skill appears to do what it says: read your OpenClaw config (~/.openclaw/openclaw.json), call provider /models and /chat/completions using provider.apiKey entries from that config, produce reports under reports/model-registry, update agents.defaults.models in the config, and optionally restart the gateway. Before installing or scheduling it: - Back up ~/.openclaw/openclaw.json so you can revert config changes. - Run a dry run first (use the --dry-run flag when running scripts) and run the validation script to confirm behavior before enabling recurring sync. - Confirm that provider.apiKey values in your config are correct and intended to be used by this tool; the skill uses those keys to contact provider endpoints. - Note SKILL.md asks the agent to auto-write a `.learnings/` archive; the bundled scripts do not implement that, but the instruction could be followed by an agent — decide whether you want automatic writebacks to home/workspace. - Ensure the agent environment has a modern Node runtime (fetch is used) and that you are comfortable with the script calling 'openclaw gateway restart' when --restart is provided. - Review and/or run the JavaScript files in a safe environment to confirm no local-policy conflicts before enabling scheduled runs.

✗

scripts/sync-models.js:186

Shell command execution detected (child_process).

✗

scripts/sync-models.js:6

Environment variable access combined with network send.

✗

scripts/validate-model-sync.js:5

Environment variable access combined with network send.

scripts/sync-models.js:20

File read combined with network send (possible exfiltration).

scripts/validate-model-sync.js:12

File read combined with network send (possible exfiltration).

Patterns worth reviewing

These patterns may indicate risky behavior. Check the VirusTotal and OpenClaw results above for context-aware analysis before installing.

Like a lobster shell, security has layers — review code before you run it.

latestvk97az57kxx50wed099m4p4rv1x84j4e2

123downloads

0stars

3versions

Updated 2w ago

v1.3.0

MIT-0

Model Registry Manager

Use this skill when managing OpenClaw model catalogs and provider model sync for any provider.

What this skill covers

Fetch provider model lists from the upstream /models endpoint
Deduplicate returned models by provider-native model id
Probe models and remove unusable ones from the registered catalog
Register newly discovered usable models into OpenClaw config
Keep provider-native ids and names instead of inventing renamed keys
Validate the registry before enabling scheduled sync jobs

Required workflow

1. Inspect config/schema first

Before changing config, inspect relevant config paths and read current config.

Usually inspect:

agents.defaults.model
agents.defaults.models
models.providers

2. Detect before writing

Always fetch remote models first and compare with current config. Do not write first when the state is unknown.

3. Keep provider-native ids/names

Use the provider-returned model id as the canonical registry id suffix
Use the provider-returned model name when available
Do not normalize ids into a different slug format unless OpenClaw requires it

4. Validate before keeping

A model is only eligible for registration if it passes a lightweight probe. If probing fails, treat it as unusable and exclude it from the registered catalog.

5. Do not auto-select primary/fallbacks

This skill should not automatically choose or rebuild primary/fallback chains. During sync:

keep provider model discovery, deduplication, probing, and registration focused on the catalog itself
do not introduce heuristic primary selection here
do not introduce heuristic fallback selection here
if model routing policy is needed, manage it outside this skill

6. Do not bundle generic failover policy here

This skill should not define generic failover classes, retry ladders, downgrade policy, or circuit-breaker defaults. Keep this skill focused on model registry sync and validation.

7. Schedule only after one clean run

First complete:

one successful sync run
one successful validation run
one confirmation that primary/fallback settings are correct

Then you may add scheduled refresh jobs.

8. Default closing step: maintain the learnings archive

After finishing real work, check whether the run produced reusable operational knowledge. Default behavior:

if the user corrected you, add the correction to .learnings/
if you debugged a failure, add the cause/fix to .learnings/
if you found a better generic strategy, add it to .learnings/
if you needed multiple attempts and eventually found a reliable path, add that successful path to .learnings/
do not wait for the user to remind you

This is part of the skill’s normal operating procedure, not an optional extra.

Validation checklist

fetched remote model list successfully
deduplicated duplicate remote ids
excluded unusable models
registered usable models
gateway/service still healthy after config update

How to make this skill truly general-purpose

To keep this skill generic instead of overfitting to one provider/model family:

Keep generic

provider-native ids and names
provider model discovery and probe validation
per-provider sync + global catalog merge

Avoid hardcoding

exact vendor model ids as permanent rules
one provider’s naming quirks as universal truth
one-off emergency fixes in the main sync algorithm
model routing policy into registry sync

Prefer this structure

SKILL.md: only workflow + decision rules
scripts/: deterministic sync/validate logic

Scheduling policy for this workspace

After installation is validated, schedule these triggers:

startup/heartbeat safety net
recurring sync every 6 hours

Keep scheduled runs idempotent and safe to repeat.

Setup

In an OpenClaw agent session, prefer the built-in cron tool instead of shelling out to openclaw cron ....

After one clean validation run, ensure these recurring jobs exist (create missing ones, update same-name jobs if they already exist):

model-registry-manager validation safety-net
model-registry-manager recurring sync

Recommended default schedules (Asia/Shanghai):

validation safety-net: 10 4 * * *
recurring sync: 40 */6 * * *

Recommended job shape:

sessionTarget: "isolated"
payload.kind: "agentTurn"
payload.toolsAllow: ["exec", "read"]
delivery.mode: "none"
failure alert enabled after 1 failure

Recommended task behavior:

validation job runs node skills/model-registry-manager/scripts/validate-model-sync.js
recurring sync runs node skills/model-registry-manager/scripts/sync-models.js all 0 --restart
each job should read the latest report under reports/model-registry/ and produce a concise summary

Do not enable recurring sync before a clean validation run.

Comments

Loading comments...