Install
$ agentstack add skill-trancong12102-agentskills-oracle ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Oracle
Delegate deep analysis to Codex CLI — launch it with a clear question, wait for it to finish, then present the results. Codex runs in a read-only sandbox with full codebase access, so it gathers its own context.
Prerequisites
- Codex CLI (required): Install with
npm i -g @openai/codex, authenticate withcodex login
If Codex CLI is not installed, stop and tell the user to install it.
Workflow
Do not read script source code. Run scripts directly and use --help for usage.
Step 1: Formulate Question
Codex CLI has full read access to the codebase and can explore files, grep code, and web search on its own. Your job is to craft a clear, specific question — not to gather context for it.
- Understand the user's question and what they need analyzed.
- Formulate a clear, specific question that captures the user's intent — include relevant file paths, function names, or architectural areas to point Codex in the right direction.
- Optionally use
--context-filefor truly external context that Codex cannot access on its own (e.g., user-provided files outside the repo, paste content).
Step 2: Run Codex
Scripts are in scripts/ relative to this skill's directory. Run --help for full usage.
Launch scripts/codex-oracle.py as a background Bash task (run_in_background: true). Codex CLI may take up to 30 minutes.
# New session
python3 scripts/codex-oracle.py --question "..." [--context-file path] [--focus text] [--dry-run]
# Resume a previous session for follow-up
python3 scripts/codex-oracle.py --session-id --question "follow-up question..." [--context-file path] [--focus text]
After launching the background task, end your response immediately and wait. Do not poll, read output files, or check process status. You will be notified automatically when Codex completes.
Step 3: Review Response
- Read the Codex output from the background task completion notification.
- Capture the
oracle-session-idfrom the last line — store it internally for follow-ups. - Review the response yourself. Decide whether it fully answers the user's question or needs clarification/deeper analysis.
Step 4: Follow-up Loop (if needed)
If the Codex response is incomplete, ambiguous, or you need it to drill deeper — send follow-ups before presenting anything to the user. Repeat as many times as needed.
- Use the stored
oracle-session-idwith--session-idto resume the session. Codex retains the full conversation history. - Only send the new follow-up question — do not repeat prior questions or the system prompt.
- Launch as a background task, wait for completion, and review the new response.
- Loop back to decide: sufficient, or another follow-up needed?
The session accumulates context with each round, making subsequent answers more informed. Start a new session (Step 1) only when the topic changes entirely.
Step 5: Present Results
Once you have a complete, clear answer from the oracle (after one or more rounds):
- Synthesize all Codex responses into a single coherent answer for the user.
- Use your own judgment on formatting and what to highlight — you do not need to echo every detail from every round.
Rules
- Codex is the analyst — your role is to formulate, launch, and present. Do not start your own parallel analysis while Codex runs.
- Organize findings by theme, not severity. Group related insights together. Structure adapts to question type (architecture → components/trade-offs, bug → root cause hypotheses, security → threat model, etc.).
- Read background-task output via the
Readtool on theoutput-filepath from the completion notification.TaskOutputcannot find background Bash task IDs and will fail. - Always use the wrapper script for Codex. The script sets the correct model and read-only mode; calling
codexCLI directly bypasses these.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: trancong12102
- Source: trancong12102/agentskills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.