← ClaudeAtlas

gpt-controllisted

Use when the user asks to get an independent model review, cross-check code or a plan, continue a GPT-Control conversation, inspect a model run, or generate or iterate on an image through Chrome Bridge. Not for facts the working tree or a local test can answer directly.
wolfiesch/gpt-control · ★ 0 · AI & Automation · score 71
Install: claude install-skill wolfiesch/gpt-control
# GPT-Control GPT-Control is a cross-model review broker, not a generic browser wrapper. ## Choose a transport - `chrome_bridge`: signed-in ChatGPT in an inactive tab. If the bridge is leased or unavailable, retry; GPT-Control will not open another browser. - `codex`: official Codex SDK in a read-only sandbox. Default when Chrome Bridge is not installed. - `responses`: official Responses API. Paid; requires `api_confirmed=true`. - `oracle_browser`: explicit legacy fallback. It can take focus and requires `allow_focus_steal=true`. - `oracle_api`: explicit paid legacy fallback. ## IDs and tools Every submission has two IDs: - `conversation_id`: provider lineage. Pass it to `gpt_chat` or `gpt_consult` for a follow-up. - `run_id`: one exact submission. Pass it to `gpt_run` for status, wait, or result. Tools: - `gpt_consult`: structured review with findings, manifest, and receipt. - `gpt_chat`: plain conversation turn. - `gpt_run`: read one exact run. - `gpt_run_cancel`: cancel an active in-process run. - `gpt_conversation_close`: local cleanup. Provider-side data remains. - `gpt_image`: Chrome Bridge image iteration. - `gpt_diagnose`: transport state without starting work. ## Attachment boundary Attachments are realpath-resolved, regular files under the workspace by default. The service caps count and aggregate bytes, hashes every file, and returns the manifest in the result. Outside-workspace and sensitive-file overrides require explicit flags. Never infer either flag