Brain Fog Mode

License: MIT Version Agent Skill Feedback Welcome

Brain Fog Mode is a beta Agent Skill that helps AI assistants reduce cognitive load, simplify complex material and decisions, and keep one clear next step when users are multitasking, stressed, foggy, or overloaded. It works for coding and everyday tasks.

It is an accessibility and work-continuity aid, not a medical tool. It does not diagnose, measure, treat, or explain symptoms.

Why

AI tools can move work fast, but they also create decision overload: too many options, too much background, and too many parallel paths. They create a second problem too — output arrives faster than a human can check it, and confidence in AI-assisted work tends to rise faster than understanding of it. Both get harder when you are multitasking, stressed, foggy, or under high cognitive load.

Brain Fog Mode changes the interaction shape:

  • it keeps the goal and task state visible
  • it separates confirmed facts from assumptions
  • it recommends one clear next action
  • it asks at most one blocking question at a time
  • it creates checkpoints and restart notes when the user stops

The goal is not only to help during brain fog. If the interaction makes decisions faster and reduces avoidable complexity, it can help anyone move through AI-assisted work with less friction.

How

Install from a GitHub repository:

npx skills add beholdrlabs/brain-fog-mode

The CLI lists skills found in the repository; choose brain-fog-mode.

Install from a local checkout:

npx skills add .

Inspect a local checkout without installing:

npx skills add . --list
npx skills use . --skill brain-fog-mode

Activate it simply:

  • "brain fog mode"
  • "low-bandwidth mode"

The skill can also activate when the user explicitly says they are overloaded, foggy, or losing track. Named activation and explicit reduced-bandwidth statements turn on session mode, which stays active until you disable it. Disable it with:

  • "normal mode"
  • "disable brain fog mode"

There is also a lighter level. Asking the agent to make a call for you — "help me decide", "which should I pick", "what would you go with" — applies the decision-reduction behavior to that one answer and then stops. No mode announcement, no checkpoint structure, nothing carried into the next message.

Simplify mode is also one answer only. It activates when the user describes material as complex or too detailed and explicitly asks to simplify it, for example: "This is very complex; please simplify it." It preserves the outcome, constraints, and important caveats while removing unnecessary detail.

It should not activate only because a task is complex, the user made a typo, or the user asked for a concise answer. Asking for a comparison or a list of pros and cons is a request for the option map, not for a decision, and is answered without a recommendation unless the user asks for one.

What It Includes

skills/
`-- brain-fog-mode/
    |-- SKILL.md
    |-- README.md
    |-- references/   # software, general work, health/safety, rationale
    |-- assets/       # checkpoint, restart, and work-plan templates
    `-- evals/        # trigger, behavioral, and safety eval cases

The core behavior lives in skills/brain-fog-mode/SKILL.md. Detailed guidance is loaded from references/ only when needed.

Beta Status

Brain Fog Mode is ready for beta review and real-world feedback. The public package is intentionally small and text-only: Markdown, JSON eval files, and reusable templates.

Current audit status:

  • static security review is in progress
  • source-faithfulness checks passed on 2026-08-04
  • behavioral eval automation is still open
  • human safety review is still open

See docs/security-audit.md and docs/source-faithfulness.md.

What's Next?

Short roadmap:

  • build repeatable evaluations for activation, response shape, safety, source faithfulness, accuracy, and token spend
  • prototype an agent eval harness with paired runs, an LLM-as-judge pass, and human review
  • add a Promptfoo eval config for repeatable local and CI checks
  • gather user feedback to tune the prompt and response shape
  • enable GitHub Discussions and publish the included low-effort feedback form

Reviewers And Feedback

We are actively looking for users to try Brain Fog Mode and review whether it actually reduces task overhead in real work.

We especially appreciate review from psychologists, clinicians, cognitive accessibility specialists, and people with lived experience of brain fog or cognitive overload. Reviews are informal and scoped; they are not medical certification or product endorsement.

With permission, we will give review attribution in the repo by name, role, and reviewed scope. Private feedback is also welcome through a channel agreed with the maintainer.

Public user feedback belongs in the repository's Feedback GitHub Discussion form. The form source is .github/DISCUSSION_TEMPLATE/feedback.yml; enable Discussions and create a category with the slug feedback to activate it. Do not include sensitive health or work information in public submissions.

Private user feedback and clinician reviews should use docs/reviewer-questionnaire.md through a private channel agreed with the maintainer.

Evaluation And Checks

The eval files are plain JSON so they can be used by different agent clients and eval tools. You do not need any specific agent CLI to read or reuse them.

Manual eval flow:

  1. Pick a case from skills/brain-fog-mode/evals/.
  2. Run a single prompt once, or deliver each entry in turns as a separate conversational turn, without the skill and with the skill available for automatic activation.
  3. Compare the outputs against the case assertions.
  4. Record accuracy, source faithfulness, response shape, token usage, and human notes.

Automated eval: TBD. Planned work includes a Promptfoo config plus baseline/generic/skill runs with LLM-as-judge scoring and human review.

Safety

Brain Fog Mode is an interaction and accessibility aid. It is meant to help a user get through a task with less load, not to keep them working while unwell.

If symptoms are severe, worsening, or feel unsafe, the skill should point the user toward a qualified healthcare professional or local emergency services instead of pushing through. It avoids country-specific hotlines, legal thresholds, diagnosis, medication, supplements, and cause claims.

License

MIT License. See LICENSE.