Notre avis
Audite un espace de travail « second cerveau » en testant réellement la récupération du contexte, l'accès aux sources, l'exécution des workflows et les routines récurrentes, puis rédige un rapport daté.
Points forts
- Vocabulaire de statuts volontairement limité (PASS / PARTIAL / UNKNOWN / FAIL) qui empêche les notations vagues.
- Exige des preuves datées plutôt que des captures de configuration : une source déclarée n'est pas une source accessible.
- Quatre sondes de récupération ancrées dans le travail réel de l'utilisateur, avec route déclarée et recours éventuel à la recherche de secours.
- Se termine par un maximum de trois actions concrètes, chacune assortie d'un test de complétion.
Limites
- Suppose une structure existante (SYSTEM.md, workspace/sources.md, journal de décisions, dossier reviews/) et ne s'applique pas à un dépôt quelconque.
- La vérification reste tributaire de ce que l'agent peut réellement lire et exécuter dans son environnement.
- Rapport qualitatif sans métrique ni score : difficile à comparer d'une exécution à l'autre.
À lancer périodiquement avant de s'appuyer sur son second cerveau ou sur ses automatisations, pour confirmer que tout fonctionne encore.
À éviter sur un projet sans espace de travail documenté de type second cerveau, ou si l'on cherche un score quantitatif plutôt qu'un diagnostic.
Analyse de sécurité
SûrThe skill instructs reading workspace files, running retrieval probes, and writing a dated markdown report. It does not include destructive commands, network exfiltration, credential handling, or obfuscated payloads. No declared tools are required beyond ordinary file access.
Aucun point d'attention détecté
Exemples
Run a healthcheck on my second brain and write the report to reviews/healthcheck-$(date +%Y-%m-%d-%H%M).md.Before I start this week, check whether a fresh session could figure out my current objective, an active project, a past decision, and where an important original file lives — and tell me which of those routes are broken.Go through workspace/sources.md and my scheduled jobs and tell me which sources I haven't successfully read recently and which recurring routines have no evidence of actually producing output.name: healthcheck description: Verify whether the second brain can retrieve current context, reach important sources, execute useful workflows, and support recurring behavior.
Healthcheck
This is a verification review, not a score-chasing exercise.
Read SYSTEM.md, the core workspace files, source registry, decision journal, installed skills, and recent healthcheck reports if present.
Create a new report under:
reviews/healthcheck-YYYY-MM-DD-HHMM.md
Review areas
Assess five areas using only these statuses:
PASS— directly verified with current evidence.PARTIAL— some useful behavior works, but meaningful coverage is missing.UNKNOWN— documented but not verified.FAIL— a required route or behavior is demonstrably broken.
1. Core context
Can a fresh session answer:
- who this is for,
- what matters now,
- who is served,
- important working preferences, without relying on this chat?
2. Retrieval
Run four realistic retrieval probes tied to the user's actual work:
- current objective or commitment,
- an active project or deliverable,
- a prior decision or lesson,
- an important original file or external record.
For each probe, record the declared route, evidence found, and whether fallback search was needed.
3. Source access
For each important source in workspace/sources.md, record:
- what it is authoritative for,
- access method,
- last successful read or dated evidence,
- known limitations.
Configuration alone is not proof of access.
4. Workflow execution
Select up to three workflows that matter to current objectives. Look for:
- clear trigger,
- inputs,
- useful output,
- validation,
- failure handling,
- evidence of an actual run.
5. Recurring behavior
Inspect any schedules, hooks, automations, or explicit human routines. Distinguish:
- documented intent,
- enabled trigger,
- evidence that the expected output actually happened.
Report format
Start with a one-paragraph conclusion.
Then include:
- status table for the five areas,
- four retrieval probes,
- source access table,
- up to three workflow checks,
- concrete issues grouped as:
- broken,
- unverified,
- optional improvement,
- up to three next actions with a clear completion test.
Do not manufacture problems to fill the report.
Priorisation de Tâches
Productivite
Priorise vos tâches avec les frameworks Eisenhower, ICE et RICE.
Generateur de Rapport Hebdomadaire
Productivite
Generez des rapports de statut hebdomadaires structures et concis.
Rapport de Daily Standup
Productivite
Génère des rapports de daily standup structurés et concis.