| name | oracle-pro-debugger |
| description | Use when a difficult debugging, architecture, migration, or hypothesis-validation problem needs an external ChatGPT Pro or Oracle CLI second opinion after local inspection, always with dry-run first, minimal file selection, local safety scan, and explicit user approval before any real external run. |
Oracle Pro Debugger
Use this skill when a local investigation leaves a hard question that benefits
from a second opinion through the Oracle CLI:
- difficult or intermittent failures;
- non-reproducible bugs;
- deep debugging after local inspection;
- architecture review;
- cross-checking competing hypotheses;
- analysis of sanitized logs;
- explicit user requests to consult ChatGPT Pro, GPT Pro, or Oracle.
Do not use it for:
- trivial work;
- replacing the initial local investigation;
- unsanitized sensitive context;
- full repository uploads;
- broad globs, local runtime state, certificates, wallets, cookies, local-only
configs, or production-sensitive material.
Required Flow
- Inspect locally, reproduce when possible, and reduce the question.
- Select the smallest useful file set with
--file.
- Use
--print-command or ORACLE_NO_EXEC=1 to validate without touching the
package executor or Oracle.
- Run a dry-run first and review the summary plus files report.
- Review the local content scan and included/excluded file list.
- Ask the user for explicit approval before any real external consultation.
- Run the real consultation only with
ORACLE_APPROVED=1 and --real-run.
- Treat Oracle output as advice, not truth.
- Verify recommendations with tests, lint, typecheck, or local reproduction.
Read references/oracle-security-policy.md before any real run. Use
references/oracle-usage.md for examples.
Required Wrapper
Do not call Oracle directly. Always use:
.agents/skills/oracle-pro-debugger/scripts/oracle_safe_consult.sh \
--print-command \
--dry-run \
--prompt "Diagnose this TypeScript failure and suggest locally testable hypotheses." \
--file "src/**/*.ts" \
--file "tsconfig.json"
Approved real-run example:
ORACLE_APPROVED=1 .agents/skills/oracle-pro-debugger/scripts/oracle_safe_consult.sh \
--real-run \
--engine browser \
--model "gpt-5-pro" \
--prompt "Review these hypotheses and identify the most likely one with local verification steps." \
--file "src/**/*.ts" \
--file "tests/**/*.ts"
The wrapper rejects broad file sets, scans local content before external use,
runs in dry-run mode by default, prints the sanitized command, and requires
approval for any external send.
