diff --git a/SKILLS_AUDIT_2026-03-05.md b/SKILLS_AUDIT_2026-03-05.md
new file mode 100644
index 0000000..5024b14
--- /dev/null
+++ b/SKILLS_AUDIT_2026-03-05.md
@@ -0,0 +1,90 @@
+# Skills Audit Report - 2026-03-05
+
+## Overview
+Reviewed all 17 custom skills, 5 custom commands, and marketplace plugins against current Claude Code native capabilities.
+
+---
+
+## RETIRE (6 items)
+
+| Item | Type | Reason | Status |
+|------|------|--------|--------|
+| `/commit` | Command | Native `/commit` command does the same thing | DONE - deleted |
+| `/commit-push` | Command | Composition of native `/commit` + `git push` | DONE - deleted |
+| `notediscovery` | Skill | Redundant with cognitive-memory MCP (same triggers, inferior impl) | DONE - archived |
+| `orchestrator` | Skill | Superseded by native Agent Teams (`/agent-teams`) | DONE - archived |
+| `create-skill` | Skill | Superseded by native `/skill` command + official docs | DONE - archived |
+| `backlog` | Skill | Test native `/backlog` plugin; retire if feature parity | Pending testing |
+
+## KEEP + OPTIMIZE (11 items)
+
+| Item | Type | Key Optimization |
+|------|------|-----------------|
+| `cognitive-memory` | Skill | Add per-project graph isolation; auto-embed hook post-store |
+| `mcp-manager` | Skill | DONE - Fixed README (removed outdated MCPs, synced registry). Still TODO: add batch mode |
+| `project-plan` | Skill | Add Gitea integration to auto-create issues; add `--merge` mode |
+| `create-scheduled-task` | Skill | Upgrade runner to Agent SDK instead of raw `claude -p` |
+| `playwright-cli` | Skill | DONE - Added Chrome vs playwright-cli positioning table |
+| `proxmox` | Skill | DONE - Restructured MCP-first (310->110 lines), Python client as fallback |
+| `major-domo` | Skill | DONE - Extracted API/CLI refs (1005->173 lines), created API_REFERENCE.md + CLI_REFERENCE.md |
+| `paper-dynasty` | Skill | DONE - Fixed MemoryGraph -> cognitive-memory references |
+| `resume-tailoring` | Skill | DONE - Modularized from 1,320 to 82 lines (SKILL.md), 443 total across 5 files |
+| `/commit-push-pr` | Command | Convert from command to skill (on-demand loading, better Gitea/GitHub detection) |
+| `/sync-config` | Command | Automate via scheduled task (nightly) + keep as manual override |
+
+## KEEP AS-IS (5 items)
+
+| Item | Type | Reason |
+|------|------|--------|
+| `json-pretty` | Skill | Tiny, standalone, works perfectly, no native alternative |
+| `youtube-transcriber` | Skill | No native alternative (audio transcription via OpenAI) |
+| `z-image` | Skill | No native alternative (local GPU image generation) |
+| `optimise-claude` | Skill | Meta-guidance for CLAUDE.md writing (could condense itself though) |
+| `/save-memories` | Command | More sophisticated than auto-memory; verify no overlap with SessionEnd hook |
+
+---
+
+## Detailed Notes
+
+### Commands to Delete
+- `/commit` and `/commit-push` are functionally identical to native Claude Code `/commit` + git push.
+- Native `/commit` integrates with permission system and handles Co-Authored-By automatically.
+
+### Paper Dynasty Memory Fix — DONE
+- `paper-dynasty/SKILL.md` updated to use cognitive-memory MCP (previous session)
+- `card-creation/.claude/settings.local.json`: `Skill(memorygraph)` → `Skill(cognitive-memory)`, removed stale `Skill(pai-sync)`
+
+### NoteDiscovery Retirement Path
+1. Export all notes from notes.manticorum.com
+2. Ingest into cognitive-memory with tags
+3. Remove skill, archive for reference
+
+### Orchestrator Retirement Path
+1. Test `/agent-teams` with existing orchestrator use cases
+2. Validate feature parity (parallel agents, review cycles, dependency tracking)
+3. Archive skill once confirmed
+
+### Resume-Tailoring Modularization
+Current: 1,320 lines in single SKILL.md
+Proposed structure:
+```
+resume-tailoring/
+  SKILL.md          (~100 lines: routing + quick reference)
+  PHASES.md         (~800 lines: Phases 0-5 workflows)
+  MULTI_JOB.md      (~200 lines: multi-job mode)
+  MATCHING.md       (~150 lines: content matching)
+  EDGE_CASES.md     (~140 lines: error handling)
+  TESTING.md        (~80 lines: test procedures)
+  EXAMPLES.md       (~130 lines: usage examples)
+```
+
+### MCP-Manager Issues
+- Config file location docs are wrong (says `~/.claude/.mcp.json`, should be `~/.claude.json` global or `<project>/.mcp.json`)
+- MCP registry in code doesn't match README
+- No auto-detection hook (manual invocation only)
+- Each MCP change requires restart (no batching)
+
+### Proxmox Restructuring
+- MCP server exists but is secondary in docs
+- Should be MCP-first, Python client as fallback
+- Duplicate Python/CLI documentation for same operations