Currency Audit

Currency Audit

Full assessment: external freshness + internal consistency + semantic accuracy. Stamp only when all pass.

brain-qa.cjs checks the date mechanically (the lastReviewed field). This skill defines the full review that earns that date. A re-stamp means both "content matches external reality" AND "content is internally consistent."

Critical Thinking Discipline

Every audit item must pass through critical thinking — not just mechanical verification. Apply these disciplines throughout:

The key question at every checklist item: "Am I verifying this is correct, or am I assuming it's correct because it looks plausible?"

When to Use

• brain-qa shows files with expired or missing lastReviewed dates
• Before a release that ships brain files to heirs
• On a scheduled cadence (target: full audit every 90 days)
• After a major external change (new API version, deprecated pattern, VS Code update)

Quick Reference

Audit Process

1. Triage

Run brain-qa and read the stale-file output:

node scripts/brain-qa.cjs

brain-qa surfaces files whose lastReviewed exceeds the 180/365-day thresholds (warn/fail). Work the failures first, then the warnings, oldest first.

Batch size: 10-15 files per session. Larger batches cause quality drift.

2. Per-File Review Checklist

For each file, work through this checklist:

Frontmatter Check

• lastReviewed field exists and is a valid YYYY-MM-DD date
• description accurately reflects current content (not stale summary)
• For instructions: applyTo patterns still match the intended activation scope
• No dropped legacy fields present (type, application, tier, currency, inheritance, lifecycle) — see scripts/brain-qa.cjs for the canonical drop list

Content Freshness Check

• API references: Do referenced APIs still exist? Check for deprecations, renamed methods, new parameters
• Version numbers: Are pinned versions still current? (Node.js, VS Code API, Azure SDKs, etc.)
• URL links: Do referenced URLs still resolve? Are they the canonical location?
• Best practices: Has the community consensus changed? Check official docs, changelogs
• Tool availability: Are referenced CLI tools, extensions, or services still available?
• Code examples: Do code snippets still compile/run against current versions?

Semantic Consistency Check

• Terminology: File uses current terms, not deprecated synonyms (e.g., "skills" not "DK files", "skills/" not "domain-knowledge/")
• Claims match reality: Do documented features, commands, or capabilities actually exist in code?
• Cross-references: Links to other skills, agents, instructions, or files are correct and targets exist
• No contradictions: File doesn't conflict with related files covering the same topic
• Process logic: If the file documents a workflow, each step's prerequisites exist and outputs are real
• Architecture alignment: Any structural claims (counts, relationships, directory layouts) match current state

Structural Check

• No token waste: Remove Mermaid diagrams in non-teaching skills, redundant examples, excessive prose
• Progressive disclosure: Most important info first, advanced topics later
• Actionable content: Every section helps the LLM produce better output

3. Update and Stamp

Stamp only after ALL checklist sections pass — frontmatter, freshness, semantic consistency, and structural. A partial review does not earn a date stamp.

1. Make all content fixes (freshness + semantic + structural)
2. Update lastReviewed: YYYY-MM-DD in frontmatter to today's date
3. If the file's description changed meaning, update it

4. Verify

Re-run brain-qa to confirm the file now passes:

node scripts/brain-qa.cjs --stdout | grep "filename"
# PowerShell: node scripts/brain-qa.cjs --stdout | Select-String "filename"

Type-Specific Guidance

Skills

Instructions

Agents

Prompts

Muscles

Batch Audit Workflow

For auditing many files efficiently:

1. Sort by category: Group files by domain (Azure, VS Code, testing, etc.) so research carries across files
2. Research once per domain: Check the domain's latest changelog/release notes once, apply to all files in that domain
3. Parallel reads: Read 5-6 files at once, note which need changes, then edit in sequence
4. Stamp in bulk: After verifying all changes, stamp dates. Use brain-qa to confirm all pass

Common Findings

Stale File Decision Table (AC2)

When a file's currency stamp is expired or missing, decide before acting:

Rule: Never re-stamp without reviewing. A stamp is an attestation, not a date bump.

Related Skills

• doc-hygiene — Anti-drift rules for living documents
• skill-review — Five-gate audit (Gate 5 Currency & Coherence is the semantic companion to this skill's mechanical re-stamp)

Would Revise If

Revisit this skill by 2026-08-26 (90 days) or sooner if any of the following fires:

• Files stamped through this audit are flagged later by self-audit or coherence sweeps as carrying stale content the audit should have caught — ≥3 occurrences in a quarter
• Re-stamps without content change recur ≥3 times for the same file in a quarter (rubber-stamp pattern)
• Decisions on equivalent Stale File Decision Table entries diverge across audits ≥2 times in a quarter (decision table too soft)
• brain-qa.cjs changes the lastReviewed threshold defaults (currently 180-day warn / 365-day fail) — re-tune the 90-day audit cadence accordingly

Track these in docs/ledgers/curation-log.md tagged [CURRENCY-AUDIT-DRIFT].