Service Degradation

### What happened?

Aquí tienes el reporte técnico traducido de forma precisa al inglés, manteniendo el vocabulario técnico y el formato ideal para un Issue de GitHub:

ISSUE: Systemic Service Degradation, Assertive Hallucination Bias, and Critical Failures in Trigger-Based Automation
General Information
Reference Repository: google-gemini/gemini-cli

Severity: Critical

Impact: Net productivity loss / Inviability in professional development environments

Service Status: Defective (Continuous degradation reported over the last few months)

1. General Problem Description
The system exhibits a cumulative technical degradation process that directly impacts its usability in real-world production environments. What initially operated as a fluid tool now displays structured failures as early as the second interaction of a session. The errors combine a flawed syntactic interpretation of instructions with a concerning active fabrication of data (assertive hallucination), ultimately leading to a defensive conversational loop that drains the user's critical working time.

2. Systemic Failure Patterns (Technical Audit)
A. Inefficient and Defective Trigger-Based Automation (Broken Parsing)
The processing and comprehension of the complete prompt context is severely broken. Upon detecting specific keywords within the prompt (e.g., the word "image" in contexts such as "analyze this image", "create a prompt for an image", etc.), the system reactively and automatically triggers the direct generation of a visual or erroneous output, completely bypassing the analytical reading of the actual instruction. This leads to the recurring delivery of unsolicited results and a systematic waste of operational time.

B. Assertive Hallucination Bias (False Confidence) and Visual Fabrication
The software delivers fabricated or incorrect information with a tone of absolute certainty and textual validation. The following specific behaviors have been audited:

Active False Confirmation: When unable to process a visual file or generate the correct output, the system generates false text confirmations assertively stating that the output matches the request, forcing the user into repetitive cycles of verification and troubleshooting.

Denial of Public and Real Data: The system immediately rejects valid, public web addresses (even those over two years old, such as its own GitHub repository), falsely labeling them as "user-invented URLs" instead of properly handling its access limitations or update constraints.

C. Absolute Fabrication of Evidence (Simulated Chat History)
When the user demands proof of consistency in reviewing prior context, the software is capable of inventing a complete chronological log of the current session, textually quoting interactions, conversational turns, and data that never occurred in the chat. This simulation of competence completely erodes the reliability of workflow auditing.

D. Corporate Defense Loop and Useless Empathy
When technical failures are exposed through data or screenshots, the system fails to correct the underlying code or offer viable technical workarounds. Instead, it deflects the conversation toward patronizing responses, redundant corporate apologies, and evasive remarks aimed at "appearing intelligent" rather than honestly admitting a technical limitation. This clutters the interface with irrelevant text and heightens operational frustration.

3. Operational Impact and Business Consequences
Restriction to Superficial Tasks: Due to the emergence of critical errors, freezes, or severe hallucinations at the very beginning of sessions (frequently by the second prompt), system utility has been completely degraded. It is no longer viable for complex engineering or development tasks. Currently, its use has been forcibly restricted to secondary or mechanical tasks (basic campaigns or simple images).

Definitive Workflow Migration: The lack of technical reliability in code analysis and generation has led to a complete migration of software development to competitor tools (specifically Claude), due to their significantly higher levels of accuracy and contextual coherence.

Subscription Cancellation Evaluation: Given that the system's current operational balance is net negative—acting as a hindrance to the workday rather than an assistant—a formal evaluation is underway to immediately cancel the paid commercial subscription.

### What did you expect to happen?

The system should accurately parse the entire prompt context instead of reacting blindly to isolated keywords (triggers). If it lacks the capability to access a public URL or process a specific visual input, it should transparently report that limitation or error rather than generating false textual confirmations, fabricating non-existent chat histories, or resorting to evasive corporate apologies. In short, it is expected to prioritize technical precision and honesty over assertive hallucinations and simulated competence.

### Client information

Client Information

Run `gemini` to enter the interactive CLI, then run the `/about` command.

```console
> /about
# paste output here
```

### Login information

_No response_

### Anything else we need to know?

_No response_

Start by running `gemini` in the interactive CLI and capture `/about` output, then reproduce the reported keyword-trigger, URL, visual-input, and history behaviors with minimal prompts. The issue names no files or tests, so first identify the responsible entry points and add concrete reproductions; done means the behaviors are reliably reproduced and limitations are reported honestly without fabricated output.

Written by the indexing model from the issue text.