oracle — best use
Oracle bundles your prompt + selected files into one “one-shot” request so another model can answer with real repo context (API or browser automation). Treat output as advisory: verify against code + tests.
Main use case (browser, GPT‑5.2 Pro)
Default workflow here: --engine browser with GPT‑5.2 Pro in ChatGPT. This is the common “long think” path: ~10 minutes to ~1 hour is normal; expect a stored session you can reattach to.
Recommended defaults:
- Engine: browser (
--engine browser) - Model: GPT‑5.2 Pro (
--model gpt-5.2-proor--model "5.2 Pro")
Golden path
- Pick a tight file set (fewest files that still contain the truth).
- Preview payload + token spend (
--dry-run+--files-report). - Run a preflight confirmation gate: show exact files, destination provider/host, and require explicit approval before upload.
- Keep uploads provider-scoped: only send to a trusted, user-approved endpoint.
- Use browser mode for the usual GPT-5.2 Pro workflow; use API only when you explicitly want it.
- If the run detaches/timeouts: reattach to the stored session (don't re-run).
Commands (preferred)
Help:
oracle --help- If the binary isn’t installed:
npx -y @steipete/oracle --help(avoidpnpxhere; sqlite bindings).
Preview (no tokens):
oracle --dry-run summary -p "<task>" --file "src/**" --file "!**/*.test.*"oracle --dry-run full -p "<task>" --file "src/**"
Token sanity:
oracle --dry-run summary --files-report -p "<task>" --file "src/**"
Browser run (main path; long-running is normal):
oracle --engine browser --model gpt-5.2-pro -p "<task>" --file "src/**"
Manual paste fallback:
oracle --render --copy -p "<task>" --file "src/**"- Note:
--copyis a hidden alias for--copy-markdown.
Attaching files (--file)
--file accepts files, directories, and globs. You can pass it multiple times; entries can be comma-separated.
Include:
--file "src/**"--file src/index.ts--file docs --file README.md
Exclude:
--file "src/**" --file "!src/**/*.test.ts" --file "!**/*.snap"
Defaults (implementation behavior):
- Default-ignored dirs:
node_modules,dist,coverage,.git,.turbo,.next,build,tmp(skipped unless explicitly passed as literal dirs/files). - Honors
.gitignorewhen expanding globs. - Does not follow symlinks.
- Dotfiles filtered unless opted in via pattern (e.g.
--file ".github/**"). - Files > 1 MB rejected.
- Default-ignored dirs:
Engines (API vs browser)
- Auto-pick:
apiwhenOPENAI_API_KEYis set; otherwisebrowser. - Browser supports GPT + Gemini only; use
--engine apifor Claude/Grok/Codex or multi-model runs. - Browser attachments:
--browser-attachments auto|never|always(auto pastes inline up to ~60k chars then uploads).
- Remote browser host:
- Host (safe default):
oracle serve --host 127.0.0.1 --port 9473 --token <secret> - Only bind
0.0.0.0when remote access is explicitly required and network controls are in place. - Client:
oracle --engine browser --remote-host <host:port> --remote-token <secret> -p "<task>" --file "src/**"
- Host (safe default):
Sessions + slugs
- Stored under
~/.oracle/sessions(override withORACLE_HOME_DIR). - Runs may detach or take a long time (browser + GPT‑5.2 Pro often does). If the CLI times out: don’t re-run; reattach.
- List:
oracle status --hours 72 - Attach:
oracle session <id> --render
- List:
- Use
--slug "<3-5 words>"to keep session IDs readable. - Duplicate prompt guard exists; use
--forceonly when you truly want a fresh run.
Prompt template (high signal)
Oracle starts with zero project knowledge. Assume the model cannot infer your stack, build tooling, conventions, or “obvious” paths. Include:
- Project briefing (stack + build/test commands + platform constraints).
- “Where things live” (key directories, entrypoints, config files, boundaries).
- Exact question + what you tried + the error text (verbatim).
- Constraints (“don’t change X”, “must keep public API”, etc).
- Desired output (“return patch plan + tests”, “give 3 options with tradeoffs”).
Safety
File attachment checklist (REQUIRED before using --file):
- Review the file list — never attach files containing secrets, credentials, or private keys.
- Hard denylist — NEVER attach:
.env,.env.*,*.pem,*.key,id_rsa*,credentials.*,secrets.*,*.p12,*.pfx,keystore.*,*.jks. - Redact aggressively — strip API keys, tokens, passwords, and connection strings from attached files before sending.
- Require explicit user approval before attaching files to any external AI service.
- Display preflight summary before upload: exact file paths, estimated token/size impact, and destination provider/endpoint.
- If any file appears sensitive, stop and require manual review instead of proceeding.
Provider and endpoint guardrails:
- Treat destination endpoints as allowlist-only: upload only to trusted, user-approved providers/hosts.
- Do not follow redirects to a different host during upload workflows.
- If provider retention or data handling is unknown, pause and require explicit user confirmation.
Remote host safety:
- When using
--remote-host, verify the host is trusted and the connection uses encryption. - Do not upload files to third-party services without explicit user confirmation.
- Do not pass
--remote-tokenvalues in plain text in shared environments.
Resource and session limits:
- Keep at most one long-running browser session per task unless the user explicitly requests concurrency.
- Reattach to existing sessions instead of starting duplicates.
- Stop/cancel stale sessions before launching new long runs to avoid resource exhaustion.
- Prefer bounded runs and explicit timeouts when executing from automation or shared environments.
“Exhaustive prompt” restoration pattern
For long investigations, write a standalone prompt + file set so you can rerun days later:
- 6–30 sentence project briefing + the goal.
- Repro steps + exact errors + what you tried.
- Attach all context files needed (entrypoints, configs, key modules, docs).
Oracle runs are one-shot; the model doesn’t remember prior runs. “Restoring context” means re-running with the same prompt + --file … set (or reattaching a still-running stored session).