Kilocode Model Sync

Sync the Kilocode provider model list in openclaw.json with the live Kilo AI API. Use when: running the weekly model sync job, checking for new/removed/updated Kilocode models, applying approved model patches to openclaw.json, or reporting sync results to mission control and Telegram. Requires KILOCODE_API_KEY in ~/.openclaw/.env.

----- THIS SKILL IS DEPRECATED - DON'T RUN -----

----- KILOCODE GATEWAY IS NOW BUILT-IN IN OPENCLAW -----

Kilocode Model Sync

Workspace: ~/.openclaw/workspace-steve/ Scripts: ~/.openclaw/workspace/skills/kilocode-model-sync/scripts/ Snapshots/diffs/patches: ~/.openclaw/workspace/kilocode-models/ Telegram delivery: Use the message tool with channel=telegram, target Guillaume's chat.

---

Step 1 — Run the sync script

bash

source ~/.openclaw/.env && python3 ~/.openclaw/workspace/skills/kilocode-model-sync/scripts/sync_models.py

Parse the JSON result from stdout. It contains:

Field	Meaning
status	"changed" | "no_change" | "error"
total_models	Total models fetched from API
diff.added	New models not in previous snapshot
diff.removed	Models that disappeared
diff.updated	Models with changed fields
snapshot_path	Where today's snapshot was saved
patch_path	Path to the .patch.json file (only if changed)
diff_path	Path to the .diff.json file (only if changed)

---

Step 2 — If status == "no_change"

Write a brief memory note and stop. No notification needed.

text

# memory/YYYY-MM-DD.md entry:
## Kilocode Model Sync — YYYY-MM-DD
- Status: no_change
- Total models: X
- No action required.

---

Step 3 — If status == "changed", notify Guillaume via Telegram

Use the message tool to send to Guillaume's Telegram:

text

🤖 *Kilocode Model Sync — {date}*

Found changes in the Kilocode model list:
✅ Added: {N} model(s)
❌ Removed: {N} model(s)
🔄 Updated: {N} model(s)

*New models:*
{for each added: • `id` — Name (ctx: Xk tokens, cost: $Y input/$Z output per 1k)}

*Removed models:*
{for each removed: • `id` — Name}

Reply to Grog: `@steve approve` to apply, or `@steve skip` to ignore.

Format costs as $0.001/1k (multiply the raw per-token value × 1000). Use free for models with 0 cost on both input and output.

---

Step 4 — Wait for approval signal from Grog

Grog will relay a sessions_send message to your session containing either:

• approve — proceed to apply
• skip — log and stop

---

Step 5 — If approved, apply the patch

bash

python3 ~/.openclaw/workspace/skills/kilocode-model-sync/scripts/apply_patch.py <patch_path>

The script:

1. Backs up openclaw.json with a timestamp
2. Replaces models.providers.kilocode.models with the new list
3. Runs openclaw gateway restart
4. Polls openclaw gateway status until RPC probe: ok (up to 30s)

Result fields:

• status: "ok" | "applied_restart_failed"
• backup_path: where the backup was saved
• gateway.ok: true/false
• gateway.error: error message if failed

---

Step 6 — Write mission control entry

Append to ~/.openclaw/workspace/memory/YYYY-MM-DD.md:

markdown

## Kilocode Model Sync — YYYY-MM-DD
- **Status:** Applied ✅
- **Added:** N models — list ids
- **Removed:** N models — list ids
- **Updated:** N models — list ids
- **Backup:** backup_path
- **Gateway restart:** ok / failed (error message)
- **Approved by:** Guillaume (via Telegram)

---

Step 7 — Send Telegram confirmation

text

✅ *Kilocode Model Sync — Applied*

{N} new models added to openclaw.json.
{N} models removed.
Gateway restarted successfully. ✅

Backup saved at: {backup_path}

If gateway restart failed:

text

⚠️ *Kilocode Model Sync — Patch Applied, Gateway Restart Failed*

Models were written to openclaw.json but the gateway did not restart cleanly.
Error: {error}
Please restart manually: `openclaw gateway restart`

---

Step 8 — Report summary to Grog (sessions_send to main session)

After completion, send a concise summary to the main agent session so Grog can relay it to Guillaume if needed. Use sessions_send with label=main or sessionKey=agent:main:main.

---

Error handling

• If sync_models.py exits with status "error": notify Guillaume via Telegram with the error, stop.
• If apply_patch.py fails to write config: the script auto-restores from backup; report failure.
• If gateway doesn't come back: report in Telegram, tell Guillaume to restart manually.
• Always write a memory note even on failure.