Incident

User reports Sonnet 5 bug causing prompt contamination and shares workaround

A user published an article detailing failure modes of Anthropic's Sonnet 5 model and an instruction layer to mitigate them. They also provide a temporary fix for a bug where user preferences/project instructions bleed into each prompt, contaminating the interaction loop. The user believes this bug is responsible for much of the initial negative feedback about Sonnet 5.

4 engagement·1 source·Sat, Jul 11, 2026, 05:20 PM

On July 11, 2026, a Reddit user shared an article they wrote about failure modes in Anthropic's Sonnet 5 model. The article includes an instruction layer designed to alleviate these issues, as well as a temporary workaround for a bug that causes user preferences and project instructions to leak into every prompt, contaminating the interaction loop. The user urges others to report the bug to Anthropic and suggests that this bug is a major source of early negative reactions to Sonnet 5, though not the only cause. The user is known for previous work on 'Object Floor' instruction sets for mitigating model failure modes.

Entities

Sonnet 5(model)Anthropic(company)

IncidentSat, Jul 11, 2026, 12:48 PM

User reports Claude API skill bug and shares workaround

A Reddit user reported a bug in Anthropic's Claude desktop app where the 'claude-api' skill consumes an excessive number of tokens. The user shared a configuration workaround to disable the skill via the settings.json file, warning others not to trigger it.

7 engagement·1 source·reddit

IncidentSat, Jul 11, 2026, 07:07 AM

Claude app user receives unrelated security advisory in response to simple prompt

A Reddit user reported that typing a simple prompt into the Claude app on iPad (Fable 5) resulted in a detailed, coherent writeup about a security product called 'Zenith Gateway'—including CVEs, patches, and affected versions—instead of the expected answer. The user suspects the model received someone else's input by mistake, as the response was fully structured with reasoning steps, not a random hallucination. This incident raises concerns about input integrity in AI chat interfaces.

3 engagement·1 source·reddit

CommunitySat, Jul 11, 2026, 09:07 AM

Developer open-sources agent-instructions repo to curb AI coding agent degradation

A developer frustrated by AI coding agents losing context and hallucinating after about 10 minutes created a set of rules to keep them on track. The rules, shared as an open-source GitHub repo, aim to reduce the need for constant reminders and prevent infinite loops. The project has gained attention from other developers facing similar issues.

2 engagement·1 source·reddit

IncidentSat, Jul 11, 2026, 10:54 PM

Users report Claude now blocks password field autofill in automated QA tests

Developers report that Anthropic's Claude model has been updated with a safety instruction that prevents it from filling password fields, even in disposable QA containers. This breaks automated audit scripts and forces manual intervention, frustrating users who need to override the behavior for testing purposes.

10 engagement·1 source·reddit

IncidentSun, Jul 12, 2026, 03:54 PM

Cursor users report worsening UI changes and recurring bugs

A Reddit user expresses frustration over Cursor's frequent UI changes that break functionality, such as agent file search, minimize button, and MCP server connections. The post highlights a cycle of forced updates that randomize button locations and remove features, reflecting ongoing community dissatisfaction.