validate

name	validate
description	This skill should be used when the user asks to "validate workspace", "check knowledge integrity", "run validation", "check for broken links", or "verify memories before promoting".

Validate Workspace

Runs a comprehensive set of integrity checks on the workspace knowledge files and reports findings grouped by severity.

When to Use

• Before promoting fleeting memories to permanent
• After a bulk import or reorganization
• Periodically as part of knowledge hygiene
• When search is returning unexpected results

Process

Step 1: Identify Workspace

Use listWorkspaces to find the target workspace. If the user specified one, match by name or slug.

Step 2: Run Validation

Call the validateWorkspace MCP tool:

validateWorkspace({ workspaceId: <id> })

This runs all checks server-side:

• Broken links — every linkedMemories entry and sources[] reference must exist on disk. Note: broken-link checking covers DB-tracked linkedMemories and sources frontmatter only. Inline markdown links in document bodies are not checked.
• Schema compliance — memory files in memory/{subtype}/ must have title and subtype frontmatter
• Duplicate IDs — same filePath cannot appear twice in permanentMemories
• Orphaned content — DB rows with no matching file on disk
• Lifecycle consistency — promoted fleeting memories must have a valid promotedFromId
• Stale memory — permanent memories whose supersededById points to a record that itself has been superseded (multi-hop stale chain)
• Missing sources — permanent memories whose sources[] array references paths that do not exist under memory/sources/ or memory/references/
• Commit-message conformance — commits touching memory/ should follow memory(<op>): convention
• Index status — files with needsEmbedding > 0 are reported as awaiting embedding

Step 3: Present Findings

Format the report by severity:

Workspace Validation: <workspace-name>

ERRORS (must fix)
  [broken-links] memory/facts/001.md → linkedMemory "memory/patterns/missing.md" not found
  [orphaned-content] memory/decisions/old.md — in DB but not on disk

WARNINGS (should fix)
  [schema-compliance] memory/insights/202501010042-insight.md — missing field: subtype
  [lifecycle-consistency] Fleeting memory id=17 promoted but missing promotedFromId

INFO
  [index-status] 3 file(s) awaiting embedding — run /engy:reindex to generate vectors
  [commit-message-conformance] Commit "add memory file" touches memory/ but skips convention

Summary: 2 errors, 2 warnings, 2 infos

Step 4: Recommend Actions

For each severity group, suggest next steps:

• Errors — must be resolved before the knowledge graph is trustworthy. Fix links, re-index, delete orphaned DB rows, or restore missing files.
• Warnings — should be fixed. Add missing frontmatter fields, correct lifecycle references.
• Info — optional. Run /engy:reindex if embedding lag is reported.

If there are no findings: "Workspace is clean — no integrity issues found."

Output Format

Findings grouped by severity (errors first), followed by a one-line summary. Each finding includes: [check-name] <path-if-applicable> — <message>.

Key Principles

• Read-only — validate never modifies state.
• Severity grouping — errors must be fixed; warnings are advisory.

Flow Position

Typical trigger: user wants to verify knowledge graph integrity before promoting memories or after a bulk import.

Next step: address errors first, then run /engy:reindex if index status flagged embedding lag.