Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/stuartshields-review-memory
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/review-memory
Tools it pre-approves 5
From allowed-tools in SKILL.md: the agent may use these without asking you each time.
Read,Grep,Glob,Write,Edit
What's inside
- Skill name
review-memory- Description
- Review auto-memory for the current project. Shows accumulated learnings, identifies noise, and helps promote permanent patterns to CLAUDE.md, rules, or skills.
- License
- MIT
- Allowed tools
- Read,, Grep,, Glob,, Write,, Edit
- Instructions
- 133 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Skill: review-memory ## When to Use Run `/review-memory` when prompted by the memory review hook, or any time you want to audit what auto-memory has captured for this project. Use after completing a milestone, before starting new work, or when memory feels noisy. Do NOT run this mid-task. Finish what you're working on first. ## Mode Selection - **Full mode** (default): Reads all topic files, presents detailed categorisation table, acts on confirmation. - **Compact mode** (`/review-memory --compact`): Quick list of new memories with one-line summaries. Asks "promote, keep, or remove?" for each. No topic file reads. Use when context is low or time is short. --- ## Full Mode ### Step 1: Load Memory Read `MEMORY.md` from the project's memory directory. The path follows this pattern: `~/.claude/projects/<project-path-encoded>/memory/MEMORY.md` Where `<project-path-encoded>` is the project root path with `/` replaced by `-`. Also read any topic files referenced from MEMORY.md (e.g., `feedback_testing.md`, `project_auth.md`). List what was found: - Total entries in MEMORY.md - Topic files and their one-line descriptions - Last modified dates ### Step 2: Categorise Each Entry For each memory entry, evaluate all three categories **in this order**: 1. **Remove?** Is this stale, wrong, duplicated by CLAUDE.md/rules, or derivable from code/git? If yes → Remove. 2. **Promote?** Does this encode a convention, decision, or constraint that should survive memory cleanup? If losing it would cause mistakes in future sessions, it belongs in CLAUDE.md, a rule file, or a skill → Promote. 3. **Keep** only if the entry is still relevant but too situational or temporary for a permanent file (e.g., user preferences, in-flight research, external context not derivable from code). **Every "Keep" must include a one-line justification for why it doesn't belong in CLAUDE.md, a rule, or a skill.** If you can't articulate why, it's probably a Promote. Present the categorisation as a table: ``` | # | Entry | File | Category | Reason | |---|-------|------|----------|--------| | 1 | Always use tabs | MEMORY.md | Remove | Already in rules/style.md | | 2 | User is senior dev | user_role.md | Keep | User profile, not a project convention | | 3 | API uses JWT auth | project_api.md | Promote | Add to project CLAUDE.md | ``` ### Step 3: Act on Categorisation Wait for the user to confirm or adjust the categories. Then: **For "Promote" entries:** - Identify the right destination (CLAUDE.md, a rule file, or a skill) - Add the learning to that file - Remove or update the memory entry to avoid duplication **For "Remove" entries:** - Delete from MEMORY.md or the topic file - If a topic file becomes empty, delete the file and remove its reference from MEMORY.md **For "Keep" entries:** - Leave as-is ### Step 3.5: Post-Promote Contradiction Check After promoting any entries, check for contradictions: 1. **Grep the destination file** for any rules or statements that conflict with what was just promoted. Flag contradictions to the user. 2. **Scan remaining memory entries** for anything that is now superseded by the promoted rule. If a memory says "do X" and the new rule says "do Y instead", mark that memory for removal. 3. **Check other rule files** for overlap. If the promoted content duplicates or contradicts an existing rule in a different file, flag it: "This overlaps with [file:line] — merge or remove?" Only flag clear contradictions. Near-duplicates where both phrasings add value are fine to keep. ### Step 4: Update Review Timestamp Write the current unix timestamp to the review cache so the hook knows when the last review happened: ```bash date +%s > ~/.claude/projects/<project-path-encoded>/memory/.last-review ``` --- ## Compact Mode For use when context is low or the hook suggests `--compact`. ### Step 1: Read MEMORY.md Only Read `MEMORY.md` from the project's memory directory. Do NOT read topic files — use only the one-line descriptions from MEMORY.md. ### Step 2: Quick List Present each entry as a numbered list with the description from MEMORY.md: ``` 1. [feedback] Stop taking shortcuts when debugging — Rewrote debugging.md 2. [feedback] Write complete code always — Restructured discipline.md 3. [project] Auth migration deadline March 15 — Legal compliance driver ``` Ask: "Any to promote, remove, or update? Otherwise I'll mark as reviewed." ### Step 3: Act on Response - If the user says "all good" or similar → skip to Step 4 - If they flag specific entries → handle just those (promote/remove) - If a promotion triggers a contradiction → flag it briefly, don't deep-dive ### Step 4: Update Review Timestamp Same as full mode: ```bash date +%s > ~/.claude/projects/<project-path-encoded>/memory/.last-review
From stuartshields/claude-setup, MIT.
More from stuartshields/claude-setup
- 0.507score
swift-concurrency-review
stuartshields
Use when reviewing Swift 5.5+ code containing `await`, `actor`, `Task { }`, `@MainActor`, `@unchecked Sendable`, cancellable timer/deadline handles, or…
SkillCFreshness: C★ 2Claude Code +9 - 0.507score
module-depth-review
stuartshields
Use when reviewing whether module boundaries earn their keep — finding shallow modules that should be consolidated into deep ones.
SkillCFreshness: C★ 2Claude Code +9 - 0.507score
debugging
stuartshields
Structured debugging protocol.
SkillCFreshness: C★ 2Claude Code +9 - 0.504score
clean-worktrees
stuartshields
Remove stale agent worktree directories and their orphaned branches.
SkillCFreshness: C★ 2Claude Code +9 - 0.504score
block-journey
stuartshields
Discover all files for a block or component, trace the editorial and front-end user journeys, and write a journey document to docs/journeys/.
SkillCFreshness: C★ 2Claude Code +9 - 0.502score
tool-usage
stuartshields
Recovery protocol for tool failures and search budgets.
SkillCFreshness: C★ 2Claude Code +9 - 0.482score
audit-vs-fix-discipline
stuartshields
Use when the user asks to audit, review, investigate, check, scan, "look at", or "find issues in" code — anything that sounds like a diagnostic request…
SkillCFreshness: C★ 2Claude Code +9 - 0.440score
vibe-user
stuartshields
Open an app in Playwright and explore it as a real user with no prior knowledge.
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
debug-rules
stuartshields
Diagnose which rules loaded, which didn't, and why.
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
figma
stuartshields
Figma-to-code workflow using MCP tools.
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
playwright
stuartshields
Playwright MCP browser automation workflow.
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
qa-check
stuartshields
Auto-detects the project's tech stack, then audits for Accessibility, Performance, and Code Quality.
SkillDFreshness: D★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Mar 26, 2026 (7 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/stuartshields/claude-setup
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from stuartshields
- 0.507score
debugging
stuartshields
Structured debugging protocol.
SkillCFreshness: C★ 2Claude Code +9 - 0.507score
module-depth-review
stuartshields
Use when reviewing whether module boundaries earn their keep — finding shallow modules that should be consolidated into deep ones.
SkillCFreshness: C★ 2Claude Code +9 - 0.507score
swift-concurrency-review
stuartshields
Use when reviewing Swift 5.5+ code containing `await`, `actor`, `Task { }`, `@MainActor`, `@unchecked Sendable`, cancellable timer/deadline handles, or…
SkillCFreshness: C★ 2Claude Code +9 - 0.504score
clean-worktrees
stuartshields
Remove stale agent worktree directories and their orphaned branches.
SkillCFreshness: C★ 2Claude Code +9 - 0.504score
block-journey
stuartshields
Discover all files for a block or component, trace the editorial and front-end user journeys, and write a journey document to docs/journeys/.
SkillCFreshness: C★ 2Claude Code +9 - 0.502score
tool-usage
stuartshields
Recovery protocol for tool failures and search budgets.
SkillCFreshness: C★ 2Claude Code +9
Cite this page
review-memory skill. Agent Mart. https://agentmart.dev/skills/stuartshields-review-memory (last updated 2026-10-24)