Security audit
Security checks for vulnerabilities and agentic risk
This skill is a local Markdown guide for improving ChatGPT workflows, with disclosed local notes and no executable automation.
Before installing, be aware that the skill keeps local ChatGPT workflow notes in ~/chatgpt/ and may guide you to paste or upload chosen material into ChatGPT. Do not store secrets or raw private documents in its memory notes, and choose a narrow activation preference if you do not want it to trigger on general ChatGPT mentions.
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
integration: pending
## Workflow Defaults
<!-- Preferred surfaces: standard chat, Temporary Chat, Projects, GPTs -->
<!-- Preferred prompt packet size, draft depth, and revision style -->
## Privacy Boundaries
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
<!-- Durable custom instruction patterns that are actually helpful -->
## Failure Modes
<!-- Repeated problems: too generic, ignores constraints, stale context, hallucinations -->
<!-- Best fixes that worked -->
## Notes
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Stuffing temporary requirements into custom instructions -> every later chat inherits the wrong behavior.
- Using the same chat for unrelated jobs -> old assumptions leak into new tasks.
- Asking for a final answer before defining audience, output format, and success criteria -> bland generic output.
- Trusting confident claims without asking what they are based on -> hallucinations survive review.
- Uploading files without telling ChatGPT which file is authoritative -> mixed or contradictory answers.
- Letting Projects accumulate stale drafts and renamed files -> the model anchors on obsolete context.
- Trying to fix a broken workflow with more adjectives -> structure beats style words.
The activation trigger list is broad enough to fire on many ordinary mentions of ChatGPT-related terms, which can cause the skill to activate when the user did not clearly request it. That creates context confusion and increases the chance of unnecessary memory writes to ~/chatgpt/memory.md, especially early in a conversation before intent is explicit.
No suspicious patterns detected.