zero-yx/static_flow / skills/approving-llm-gateway-account-batches
approving-llm-gateway-account-batches
VerifiedUse when pending LLM Gateway account-contribution requests need to be preflighted, validated, issued, patched, and usage-refreshed in bulk through the admin API, including deterministic proxy assignment and per-account concurrency settings.
Approving LLM Gateway Account Batches
Use this skill after public account-contribution requests are already queued as
pending. It automates the admin side:
- fetch pending requests,
- preflight each request by attempting refresh and probing Codex models through
the local
http://127.0.0.1:11116proxy, - validate each request,
- approve and issue each account,
- patch the imported account with proxy, request settings, and
auto_refresh_enabled, - attempt a best-effort usage refresh for the imported account.
This skill does not bypass the existing review flow. It only drives the existing admin API in batch form.
When To Use
- The public batch submit step already created
pendingrequests. - You want to process one batch end to end through admin APIs.
- Imported Codex accounts should receive standard settings after issue.
- Some imported Codex auth bundles may have usable access tokens but broken or intentionally short-lived refresh credentials.
- Imported Codex accounts should still be issuable when direct models access works even if OAuth refresh is broken.
Admin API Assumptions
- Admin base examples:
http://127.0.0.1:19182http://127.0.0.1:19082
- Required routes:
GET /admin/llm-gateway/account-contribution-requestsPOST /admin/llm-gateway/account-contribution-requests/{id}/validatePOST /admin/llm-gateway/account-contribution-requests/{id}/approve-and-issuePATCH /admin/llm-gateway/accounts/{name}POST /admin/llm-gateway/accounts/{name}/refresh-usageGET /admin/llm-gateway/accountsGET /admin/llm-gateway/proxy-configs
Preferred Workflow
- Confirm the batch prefix and expected request count.
- Run a dry run first.
- Process only matching
pendingrequests. - Preflight refresh and direct models access before validate.
- Treat successful models access as the issue gate.
- Skip validate and issue when the models probe fails.
- If refresh preflight fails but models succeed, still issue the account and
patch
auto_refresh_enabled = false. - Patch imported accounts immediately after issue.
- Attempt post-issue usage refresh through the dedicated usage endpoint.
- If usage refresh fails, keep the issued account and record the failure, but do not roll back or mark the whole batch item as failed.
- Verify all matching requests moved out of
pending.
Proxy Assignment Rule
- Only consider
activeproxy configs. - Count current usage from
activeCodex accounts withproxy_config_id. - Choose the proxy with the lowest current count.
- Break ties by proxy name for deterministic output.
- Update the in-memory count after each assignment so one batch is balanced.
Standard Patch Settings
auto_refresh_enabled = refresh_preflight_okrequest_max_concurrency = 3request_min_start_interval_ms = random[100, 1000]proxy_mode = fixedproxy_config_id = selected least-used proxy- post-issue usage refresh is best-effort and does not block a models-usable account from being considered successfully issued
Helper Script
Run:
python3 skills/approving-llm-gateway-account-batches/scripts/approve_account_contribution_batch.py --help
Dry run:
python3 skills/approving-llm-gateway-account-batches/scripts/approve_account_contribution_batch.py \
--admin-base-url "http://127.0.0.1:19182" \
--account-prefix "pickup7_" \
--expected-count 20 \
--admin-note "batch validate and issue"
Real run:
python3 skills/approving-llm-gateway-account-batches/scripts/approve_account_contribution_batch.py \
--admin-base-url "http://127.0.0.1:19182" \
--account-prefix "pickup7_" \
--expected-count 20 \
--admin-note "batch validate and issue" \
--apply
Real run with explicit refresh tuning:
python3 skills/approving-llm-gateway-account-batches/scripts/approve_account_contribution_batch.py \
--admin-base-url "http://127.0.0.1:19182" \
--account-prefix "pickup7_" \
--expected-count 20 \
--admin-note "batch validate and issue" \
--codex-proxy-url "http://127.0.0.1:11116" \
--refresh-max-attempts 6 \
--refresh-retry-delay-seconds 2 \
--apply
Notes
- Prefer local pb-mapper admin access on
127.0.0.1:19182when available. - Default Codex preflight proxy is
http://127.0.0.1:11116. - The script prints per-account progress with immediate flush, so long-running preflight/validate/issue/usage-refresh steps stay observable.
- The script does not write raw request tokens into the result JSON. It records
only preflight conclusions such as models success, refresh success, and the
chosen
auto_refresh_enabledvalue. - The script stops neither validation, issue, nor usage refresh routing logic inside the backend; it only sequences admin calls.
- The script records one JSON result file under
/tmp/.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |