Your Current Setup Audit

Inventory, trigger analysis, routing gaps, and scaling concerns for your 30 memory files.

Contents

Memory Inventory

Total: 30 files, ~51 KB of memory content in ~/.claude/projects/-home-visar/memory/.

#FileTypeTopicSize
1MEMORY.mdIndexAuto-memory index (23 entries)3.2 KB
2browser-automation.mdTool guideChrome MCP tools, Xvfb, context rules3.3 KB
3cv.mdProjectCV methodology, keyword targets3.1 KB
4feedback_background_servers.mdFeedbackUse systemd, not npm run dev0.7 KB
5feedback_cv_keywords.mdFeedbackRecruiter keyword checklist1.0 KB
6feedback_cv_versioning.mdFeedbackNever overwrite CV, create vN+10.6 KB
7feedback_evidence_based_debugging.mdFeedbackCite evidence, no speculation1.4 KB
8feedback_logs_before_fixes.mdFeedbackInstrument before fixing1.5 KB
9feedback_no_magic_strings.mdFeedbackUse constants, not string literals0.9 KB
10feedback_no_overflow_hidden.mdFeedbackFix root cause, don't hide overflow0.6 KB
11feedback_one_fix_at_a_time.mdFeedbackOne approved change at a time0.6 KB
12feedback_ownership_patterns.mdFeedbackRust-inspired one-writer ownership2.1 KB
13feedback_svelte5_imperative_bridge.mdFeedbackNo imperative APIs inside $effect1.5 KB
14feedback_sw_fetch_handler.mdFeedbackSW fetch must be no-op1.2 KB
15feedback_systemd_services.mdFeedbackUse systemctl, not manual kill0.7 KB
16feedback_timebox_then_pivot.mdFeedback1h timebox, then minimal repro1.1 KB
17feedback_verify_claims.mdFeedbackLabel claims, offer to verify2.1 KB
18google-docs.mdTool guideOAuth, workflow, naming convention0.7 KB
19hetzner-deploy.mdTool guideDeploy script, server, monorepo2.6 KB
20local-apps.mdReferencePorts, HTTPS, PWA patterns5.2 KB
21plane.mdTool guidePlane PM API, ticket management1.1 KB
22project_video_platform_ownership.mdProjectService ownership boundaries2.0 KB
23reference_pdf_convert.mdReferenceweasyprint for HTML-to-PDF0.4 KB
24reference_startup_albania_grant.mdReferenceGrant portal, doc links, scoring0.6 KB
25reference_streamonitor.mdReferenceSC protocol reference code0.8 KB
26svelte5-pitfalls.mdReference11 quick rules for Svelte 53.1 KB
27svelte5-pitfalls-detail.mdReferenceFull explanations with code examples11.0 KB
28user_identity.mdUserName, email, phone, location0.6 KB
29user_role.mdUserProfessional background, career1.4 KB
30xvfb.mdTool guideXvfb displays, peek command0.4 KB

Organization Assessment

What Works Well

Problems Found

Content duplication across files
Orphan file svelte5-pitfalls-detail.md (the largest file at 11 KB) is not listed in MEMORY.md. It is only reachable via a link inside svelte5-pitfalls.md. If Claude reads the quick-rules and decides it has enough context, the detail file never gets loaded.

Inconsistent frontmatter: Some files have YAML frontmatter, others don't. Files without: browser-automation.md, google-docs.md, hetzner-deploy.md, local-apps.md, plane.md, svelte5-pitfalls.md, svelte5-pitfalls-detail.md, xvfb.md.

Trigger Coverage Analysis

Files with Explicit Triggers in CLAUDE.md 7 files

Trigger in CLAUDE.mdMemory FileQuality
"Before deploying anything to Hetzner"hetzner-deploy.mdStrong
"Before interacting with Plane API or tickets"plane.mdStrong
"Before using any mcp__claude-in-chrome__* tools"browser-automation.mdStrong
"Before creating or updating a Google Doc"google-docs.mdStrong
"Before viewing or managing Xvfb displays"xvfb.mdStrong
"Before working on local-network apps"local-apps.mdStrong
"All Svelte 5 projects — before writing code"svelte5-pitfalls.mdStrong

Files with NO Trigger 23 files

These rely entirely on the LLM reading MEMORY.md one-line descriptions and deciding to load them:

CategoryFilesRisk
Universal behavioral rules (should apply to ALL tasks) feedback_no_magic_strings, feedback_one_fix_at_a_time, feedback_evidence_based_debugging, feedback_logs_before_fixes, feedback_verify_claims, feedback_timebox_then_pivot High
Context-specific feedback (need triggers) feedback_cv_*, feedback_background_servers, feedback_systemd_services, feedback_sw_fetch_handler, feedback_no_overflow_hidden, feedback_ownership_patterns, feedback_svelte5_imperative_bridge Medium
Reference/project data (on demand) cv.md, reference_*, project_video_platform_ownership, user_* Low
Orphan svelte5-pitfalls-detail.md High

The Routing Gap

What a Fresh Claude Instance Will Miss

All feedback_* behavioral rules A fresh Claude instance sees the MEMORY.md one-liners in context, but there is no instruction in CLAUDE.md saying "before writing any code, read the feedback memories." The general-purpose rules (one_fix_at_a_time, verify_claims, evidence_based_debugging) are the most likely to be missed because they apply to everything — there is no specific keyword trigger.

High risk of being missed:

Low risk: Tool guides (good triggers), user_identity and user_role (descriptive enough one-liners).

CLAUDE.md Effectiveness

Strengths

Weaknesses

  1. No "always read" section. Behavioral feedback rules that apply to ALL tasks have no entry point
  2. MEMORY.md is a passive index, not an active router. It's a directory listing, not a routing table
  3. No CV work trigger. Three CV-specific files with no routing
  4. No video platform ownership trigger. Not linked from the project entry
  5. Project entries list directories but not memory dependencies
  6. No trigger for reference_streamonitor.md

Scaling Concerns

ConcernCurrentAt 2x ScaleBreaking Point
MEMORY.md line count23 entries (~23 lines)~46 entries200-line limit hit at ~80-100 entries
Context window pressure~1.5 KB for one-liners~3 KBAt 100 entries: ~6 KB of noisy one-liners
"Load everything" temptationManageable15-30 KB burnedSignificant context waste
Feedback accumulation13 files~26 filesOne-liners blur together, overlaps grow
Archival/deprecationNo mechanismStale files lingerOutdated memories mislead
Growth projection At the current rate (~2-3 feedback memories per week of active development), the MEMORY.md line limit is hit in roughly 6-8 months. Before that, the signal-to-noise ratio of the index degrades as entries blur together.
← Claude Code Internals Memory Routing Problem →