tentatio

Vault-wide quality audit skill (Tentatio = testing, trial, proof). Detects batch metadata inconsistencies: folder/status mismatches, non-standard status values, registry/folder sync errors, missing READMEs, skill health issues, and knowledge freshness. Complements real-time hooks (which catch single-file issues at write time) by scanning historically accumulated problems across the entire vault. Triggers on '/tentatio', 'vault audit', 'check vault health', 'run quality checks', etc.

Tentatio

Tentatio (Latin): testing, trial, proof

Vault-wide batch quality audit. Catches systemic inconsistencies that real-time hooks cannot detect (root cause: lifecycle skills move folders but don't always batch-update child frontmatter).

Trigger Patterns

TriggerScopeDescription
/tentatiofullFull vault audit
/tentatio 50xprojectSingle project audit
/tentatio --skillsskillsSkill health only
/tentatio --knowledgeknowledgeKnowledge freshness only
/tentatio --knowledge 33knowledgeKnowledge freshness for specific domain folder
/tentatio --sessionssessionsSession follow-up tracker
/tentatio --evolve—Self-evolution analysis

Vault Structure

→ See ../_common/references/vault-structure.md for full structure. → See ../_common/references/frontmatter-standard.md for status taxonomy.

Key Paths:

  • Projects: 50. PROJECTS/51. Active/, 52. Someday/, 53. Pending/, 54. Archive/
  • Knowledge: 30. KNOWLEDGE/, 20. LITERATURE/
  • Registry: 90. SYSTEM/93. Registries/Project-Number-Registry.md
  • Reports: 90. SYSTEM/97. Verification-Reports/
  • Skills: ~/.claude/skills/

Always excluded: 100. SECURE/**, 90. SYSTEM/91. Templates/**

Checks

10 checks via references/check-manifest.md. Load only checks matching current scope (lazy-loading).

Core Workflow (4 Steps)

Step 1: Collect & Scan

  1. Read references/check-manifest.md
  2. Determine scope from trigger → filter manifest by scope
  3. For each matching check: load reference file, execute sequentially
  4. Each check follows: Scope → Detection → Classification → Auto-Fix
  5. Collect issues as: {check_id, severity, file_path, current_value, expected_value}

Step 2: Report

Display Korean summary table, save detailed report to 90. SYSTEM/Verification-Reports/YYYY-MM-DD-tentatio.md.

→ See references/report-template.md for format and section matrix.

Step 3: Fix (optional)

Ask user: "자동수정 가능한 {M}건을 수정할까요?"

Auto-fix patterns (see individual check Section 4 for details):

  • README.md folder/status sync → change status to match folder
  • Deprecated values → 1:1 mapping per frontmatter-standard.md
  • Deprecated types → 1:1 mapping per type-standard.md
  • Knowledge metadata → user-confirmed tier/review_by

Individual notes in archived projects: report only. Method: Edit tool for frontmatter edits.

Step 4: Done

Report saved. If auto-fixes applied, show count and suggest re-running /tentatio.

Step 5: Self-Evolution (--evolve)

Runs independently of standard checks.

  1. Load check-manifest.md → all registered checks and scopes
  2. Run 5 detection rules:
    • Uncovered Folders: vault top-level folders not in any check's Scope.paths
    • Unvalidated Fields: frontmatter fields used in 10+ files without validation
    • Unregistered Checks: check-*.md files not in manifest
    • Manifest Drift: manifest entries with no matching file
    • Structure Drift: vault-structure.md entries vs actual vault folders (find mismatches)
  3. Generate coverage matrix
  4. Present gaps to user via AskUserQuestion → user confirms before new check creation

Report format: → See references/report-template.md Self-Evolution section.

Excluded Checks

Hooks handle: standard markdown links, AI model anachronisms, Unicode checkboxes, verification markers. Purgatio handles stale inbox notes. Broken wikilinks deferred.

Integration

Standalone audit tool. Related: memorio/suspensio/dilatio (root cause of mismatches), purgatio (individual note quality), elevatio (initial project creation). Root cause fix: lifecycle skills now batch-update child frontmatter (see _common/references/child-frontmatter-update.md).

Output Format Rules

Skill files: English (token efficiency) User-facing output: Korean (user preference)