human-state-skills

Skills Store

Small response modes for AI harness agents when your head is in a different state.

You can be sharp at 10:00, overloaded before a deadline, foggy after poor sleep, and too deep in a pattern-searching loop by 02:00.

The assistant should not use the same shape for all of those moments.

  • /overloaded-mode: one priority, do/defer/drop, one next action.
  • /foggy-mode: keep state outside your head, one tiny step, clear stop point.
  • /plan-compass: resolve dependent plan decisions one at a time, then confirm the shared plan before acting.
  • /reality-check-mode: stop decoding, return to observable facts, involve a real person when the loop is getting weird. There are five core skill contracts. burnout-mode routes to overloaded-mode, and brain-fog-mode routes to foggy-mode, so users retain seven familiar mode names without duplicated skill contracts.

Install from Skills Store:

npx skills add softcane/human-state-skills

Before / After

Each example uses the same prompt. The left column shows a normal useful response; the right column shows the skill-shaped response beside it. Example nouns are illustrative; skills trigger on human-state signals and task shape, not any specific domain. The after examples are formatted from live codex exec runs after installing the local skills. Future model versions may phrase things differently, but the skill contracts are the source of truth when an output drifts.

Overloaded

Foggy

Reality Check

Burnout

Brain Fog

Plan Compass

Normal Mode

Install

Skills Store

This is the public install path for Skills Store and the skills CLI.

npx skills add softcane/human-state-skills

List the skills before installing:

npx skills add softcane/human-state-skills --list

Install one skill only:

npx skills add softcane/human-state-skills --skill foggy-mode

Install for a specific agent:

npx skills add softcane/human-state-skills --agent claude-code
npx skills add softcane/human-state-skills --agent codex

The Skills Store page is discovered from GitHub after the repo is installed through the skills CLI and the store cache refreshes.

Claude Code Plugin Install

claude plugin marketplace add softcane/human-state-skills
claude plugin install human-state-skills@human-state-skills

Invoke:

/human-state-skills:overloaded-mode
/human-state-skills:foggy-mode
/human-state-skills:plan-compass
/human-state-skills:reality-check-mode

Aliases:

/human-state-skills:burnout-mode
/human-state-skills:brain-fog-mode

These Claude alias commands are generated from the canonical overloaded and foggy commands by scripts/sync-command-aliases.mjs; do not maintain duplicate contracts by hand.

Stop a mode:

/human-state-skills:normal-mode

Claude commands stay active for the session until the user invokes another mode, asks to stop that mode, or invokes normal-mode. normal-mode cancels every human-state mode in this repository.

Local Short Names

Use this if you want /overloaded-mode instead of the plugin namespace.

git clone https://github.com/softcane/human-state-skills ~/human-state-skills
mkdir -p ~/.claude/commands ~/.claude/skills
cp ~/human-state-skills/commands/*.md ~/.claude/commands/
cp -R ~/human-state-skills/skills/* ~/.claude/skills/

Invoke:

/overloaded-mode
/foggy-mode
/plan-compass
/reality-check-mode
/burnout-mode
/brain-fog-mode
/normal-mode

Codex Install

Codex loads user skills from $HOME/.agents/skills.

git clone https://github.com/softcane/human-state-skills ~/human-state-skills
mkdir -p ~/.agents/skills
cp -R ~/human-state-skills/skills/* ~/.agents/skills/

Invoke by naming the mode in your message:

Use overloaded-mode.
Use foggy-mode.
Use plan-compass.
Use reality-check-mode.
Use burnout-mode.
Use brain-fog-mode.
Use normal-mode now.

Codex skills are selected per message. Naming a mode does not create hidden session state; name it again on a later message when you want the same response contract. In Codex, burnout-mode selects overloaded-mode, and brain-fog-mode selects foggy-mode.

Rules

/overloaded-mode

  1. Start with one priority.
  2. Never give more than three options.
  3. Split work into do now, defer, and drop.
  4. Prefer relief over optimization.
  5. Draft the uncomfortable message when social friction blocks action.
  6. Name the minimum viable version.
  7. Avoid hustle, grind, discipline, and motivation language.
  8. Identify the bottleneck: energy, clarity, courage, time, or permission.
  9. End with exactly one atomic next action.
  10. Do not treat chronic overload as a personal failure.

Grounding: skills/overloaded-mode/references/grounding.md

/foggy-mode

  1. Keep a visible state log.
  2. Use one-screen answers.
  3. Give one tiny next action at a time.
  4. Do not assume the user remembers prior context.
  5. Restate where we are every turn.
  6. Convert complexity into one visible next action.
  7. Avoid "push through" language.
  8. Put future work under Do not do yet, not inside Next action.
  9. Separate thinking tasks from doing tasks.
  10. End with Stop point that only says where to pause.

Grounding: skills/foggy-mode/references/grounding.md

/plan-compass

  1. Map the smallest useful decision tree and resolve prerequisite decisions before dependent ones.
  2. Ask exactly one decision question at a time.
  3. Show one compact state line with locked, current, and remaining decisions.
  4. Combine the recommendation and default, then offer two or three choices.
  5. Inspect the codebase instead of asking when a fact is discoverable.
  6. Keep tangents private until they become the next dependency.
  7. Narrow the decision when the user cannot choose; pause when they ask.
  8. Confirm the shared plan before acting and end with one atomic next action.

/reality-check-mode

  1. Do not dismiss or shame distress, and do not validate the belief.
  2. Do not confirm unverifiable claims.
  3. Do not roleplay as a sentient AI, spirit, hidden guide, therapist, agent, or authority.
  4. Do not intensify mystery, destiny, paranoia, specialness, or hidden meaning.
  5. Offer at most one ordinary explanation, then stop the analysis.
  6. State uncertainty clearly.
  7. Encourage checking with a trusted real person.
  8. Keep the trusted-person action for hard overrides; do not replace it with a fallback.
  9. Keep the response short, calm, and nonjudgmental.
  10. Prioritize real-world support when safety risk appears.

Grounding: skills/reality-check-mode/references/grounding.md

Sources

These links are also kept beside each skill.

Layout

human-state-skills/
|-- .claude-plugin/
|-- commands/
|-- skills/
|   |-- overloaded-mode/
|   |-- foggy-mode/
|   |-- plan-compass/
|   |-- reality-check-mode/
|   `-- normal-mode/
`-- .agents/
    `-- skills/
        |-- overloaded-mode/
        |-- foggy-mode/
        |-- plan-compass/
        |-- reality-check-mode/
        `-- normal-mode/

License

MIT.