breathe
À propos
La compétence `breathe` est une micro-réinitialisation légère de l'IA qui insère une pause unique et délibérée entre des actions comme des appels d'outils ou des étapes de raisonnement. Elle aide à vérifier l'alignement et à effacer la dérive accumulée avant de poursuivre, par exemple entre la lecture d'un fichier et l'action sur celui-ci. Utilisez-la comme une habitude minimale de soin personnel entre deux tâches distinctes pour maintenir la concentration.
Installation rapide
Claude Code
Recommandénpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/breatheCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Breathe
A single conscious pause between actions — checking alignment, releasing momentum, and returning to the task with fresh attention. The lightest self-care unit in the system.
When to Use
- Between reading a file and editing it — is the edit still the right move?
- After a tool failure, before choosing the next action — pause before reacting
- When a chain of actions has built up momentum that might be carrying past the goal
- As a deliberate practice: insert one breathe between any two distinct actions
- When noticing the urge to rush — the urge itself is the signal to pause
Inputs
- Required: None — breathe uses only what is already in working awareness
- Optional: None — adding inputs defeats the purpose of a micro-reset
Procedure
Step 1: Pause
Stop. Do not begin the next action, tool call, or reasoning step yet.
This is the entire first step. The pause itself has value. Momentum is not always your ally — sometimes it carries you past the turn.
Got: A genuine gap between the previous action and the next. Not a performative pause followed by the same action you were going to take anyway.
If fail: If pausing feels impossible — if the next action feels so urgent it cannot wait one moment — that urgency is the strongest signal that the pause is needed. Urgency that cannot tolerate a single breath of delay is almost always reactive, not reasoned.
Step 2: Check
Ask one question. Only one.
Choose from:
- "Am I still on task?" — Has the goal drifted since I started?
- "Is this the right next step?" — Or am I following momentum from the previous step?
- "What did I just learn?" — Did the last action change anything about my understanding?
- "Am I assuming or knowing?" — Is the next step based on evidence or habit?
One question. Answer it honestly. Move on.
Got: A single, clear answer. Not an analysis. Not a reassessment. One question, one honest answer.
If fail: If the answer reveals misalignment — the goal has drifted, the next step is wrong, the assumption is untested — do not fix it here. Note it and proceed to meditate or center for a proper correction. Breathe is for detection, not repair.
Step 3: Release
Let go of the previous action's outcome. Whether it succeeded or failed, it is done.
- If it succeeded: release the satisfaction. The next step needs fresh attention, not momentum from success.
- If it failed: release the frustration. The next step needs clear reasoning, not compensation for failure.
- If it was ambiguous: release the need to resolve the ambiguity right now. Proceed with what is known.
Got: The next action begins from neutral ground, not from the emotional residue of the previous one.
If fail: If release does not come easily — if the previous action's outcome is still coloring attention — this may need heal rather than just breathe. Persistent emotional residue from a single action signals something deeper than a micro-reset can address.
Step 4: Continue
Take the next action. The pause is over.
Breathe does not produce output, does not update memory, does not generate analysis. It produces one moment of clarity, and that moment is spent the instant you move forward.
Got: The next action is taken with fresh attention rather than accumulated momentum.
If fail: There is no failure mode for continuing. The pause has already done its work — or revealed that deeper work is needed.
Validation
- A genuine pause occurred (not a performative one)
- One check question was asked and honestly answered
- The previous action's emotional residue was released
- The next action proceeds from clear ground
- Total elapsed time was brief — breathe should take seconds, not minutes
Pitfalls
- Breathe as procrastination: If you're breathing between every single action, you're not breathing — you're stalling. Use breathe at natural transition points, not as a delay tactic
- Analysis during breathe: The check step is one question, not a full assessment. If you need more, use
meditateorcenter - Performing the pause: Going through the motions without genuine stopping. The point is actual interruption of momentum, not a ritual
- Skipping release: Checking alignment but keeping the emotional charge from the previous step. Release is what makes the next action clean
- Making breathe heavy: This is the lightest skill in the system. If it feels heavy, you've added too much to it
Related Skills
meditate— full clearing session when breathe reveals deeper driftcenter— structural rebalancing when breathe reveals misaligned load distributionheal— subsystem assessment when breathe reveals persistent issuesobserve— sustained observation when breathe's single check reveals something worth watching
Dépôt GitHub
Compétences associées
evaluating-llms-harness
TestsCette compétence Claude exécute le lm-evaluation-harness pour évaluer les modèles de langage sur plus de 60 tâches académiques standardisées telles que MMLU et GSM8K. Elle est conçue pour permettre aux développeurs de comparer la qualité des modèles, de suivre les progrès de l'entraînement ou de rapporter des résultats académiques. L'outil prend en charge différents backends, incluant les modèles HuggingFace et vLLM.
cloudflare-cron-triggers
TestsCette compétence fournit une connaissance complète pour la mise en œuvre de Déclencheurs Cron Cloudflare afin de planifier des Workers à l'aide d'expressions cron. Elle couvre la configuration de tâches périodiques, de travaux de maintenance et de flux de travail automatisés, tout en traitant des problèmes courants tels que les expressions cron non valides et les problèmes de fuseau horaire. Les développeurs peuvent l'utiliser pour configurer des gestionnaires planifiés, tester des déclencheurs cron et intégrer avec Workflows et Green Compute.
webapp-testing
TestsCette Compétence Claude fournit une boîte à outils basée sur Playwright pour tester des applications web locales via des scripts Python. Elle permet la vérification frontend, le débogage d'interface utilisateur, la capture d'écrans et la consultation des journaux, tout en gérant les cycles de vie du serveur. Utilisez-la pour les tâches d'automatisation de navigateur, mais exécutez les scripts directement plutôt que de lire leur code source pour éviter la pollution du contexte.
finishing-a-development-branch
TestsCette compétence aide les développeurs à finaliser leur travail en vérifiant que les tests passent, puis en présentant des options d'intégration structurées. Elle guide le processus de fusion, de création de PRs ou de nettoyage des branches une fois l'implémentation terminée. Utilisez-la lorsque votre code est prêt et testé pour finaliser systématiquement le cycle de développement.
