Review

What Is Review?

The Review skill is a powerful audit tool designed for the Claude Code environment, specifically targeting the management and optimization of auto-memory. Auto-memory, a feature that captures and persists contextual insights from development activities, can quickly accumulate redundant, stale, or irrelevant data. The Review skill systematically analyzes this memory, providing a structured assessment that highlights promotion candidates, identifies obsolete or outdated entries, exposes opportunities for consolidation, and reports on overall health metrics. By invoking the /si:review command, development teams can maintain a lean, relevant, and actionable memory base to support ongoing self-improvement and code intelligence.

Why Use Review?

As codebases and teams grow, the auto-memory can become cluttered with repetitive observations, outdated references, and fragmented insights. This clutter not only dilutes the value of stored knowledge but can also lead to confusion, missed opportunities for automation, and increased cognitive load. The Review skill addresses these challenges by:

• Promoting clarity: Surfaces high-value insights ready for elevation or action.
• Maintaining relevance: Prunes memory entries that reference obsolete files or deprecated concepts.
• Optimizing utility: Highlights redundant or overlapping entries that can be consolidated, reducing noise and storage overhead.
• Ensuring compliance: Monitors memory size against configured limits (e.g., the 200-line startup limit), helping teams avoid performance issues or unexpected failures.

By integrating Review into regular engineering workflows, teams ensure that their auto-memory remains a reliable, actionable resource rather than an unchecked data dump.

How to Get Started

To use the Review skill, ensure it is installed as part of your Claude Code self-improving agent setup. The skill operates by analyzing the auto-memory directory associated with your current project.

Basic Usage

The primary command is:

/si:review

This initiates a full audit of the current project's auto-memory, providing detailed findings and recommendations.

Focused Reviews

For targeted inspections, the following options are available:

• Quick summary:

  bash

  /si:review --quick

  Outputs only counts and the top three candidates for promotion or action.

• Stale entries only:

  bash

  /si:review --stale

  Focuses exclusively on entries referencing outdated files or concepts.

• Promotion candidates:

  bash

  /si:review --candidates

  Shows entries identified as strong candidates for promotion or further action.

Memory Directory Resolution

The skill automatically detects your project's auto-memory directory based on your working directory. It checks the following path pattern:

MEMORY_DIR="$HOME/.claude/projects/$(pwd | sed 's|/|%2F|g; s|%2F|/|; s|^/||')/memory"

If the directory does not exist, Review will suggest verifying whether auto-memory is enabled for your project.

Key Features

1. Memory Directory

Discovery

Review locates the relevant memory directory for the current project. If the directory is missing, it proactively notifies the user and recommends checking the /memory configuration.

2. MEMORY.md Analysis

The core of the audit is the MEMORY.md file within the memory directory. Review reads this file, counts its lines, and compares it against a standard threshold (e.g., 200-line startup limit). This count helps teams preempt issues with oversized memory files.

Example:

wc -l $MEMORY_DIR/MEMORY.md
## Output: 185 MEMORY.md

3. Recurrence

Detection

The skill scans for recurring concepts, such as similar entries phrased differently or repeated references to persistent issues (e.g., "again", "still", "keeps happening"). It also detects overlapping themes across different topic files.

Example check for recurrence:

grep -i -E "again|still|keeps happening" $MEMORY_DIR/MEMORY.md

4. Staleness

Identification

Review verifies whether each memory entry references files that still exist in the repository. If an entry points to a deleted or renamed file, it is flagged as stale.

Example:

## For each file reference in MEMORY.md
if ! [ -f "$FILE_PATH" ]; then
  echo "Stale reference: $FILE_PATH"
fi

5. Promotion Candidate

Detection

Entries considered valuable or actionable (e.g., those summarizing major insights or repeatedly flagged by the team) are highlighted as candidates for promotion or further automation.

6. Consolidation

Opportunities

If multiple entries cover similar ground, Review suggests consolidation, listing overlapping entries for manual or automated merging.

7. Health

Metrics

The skill reports on key health indicators, such as total memory size, number of stale entries, count of promotion candidates, and consolidation suggestions.

Best Practices

• Run Regular Reviews: Integrate /si:review into your weekly or sprint-end workflows to avoid memory bloat.
• Act on Recommendations: Use the actionable output to promote, prune, or consolidate entries as soon as possible.
• Monitor Memory Size: Stay below configured memory limits to maintain system performance and reliability.
• Customize Patterns: Adjust recurrence and staleness detection patterns to better fit your team’s coding and documentation style.
• Combine with Other Skills: Use Review in conjunction with promotion or memory-editing tools to streamline knowledge curation.

Important Notes

• The Review skill depends on the structure of the Claude auto-memory directory and the presence of a MEMORY.md file.
• If the memory directory is missing or incorrectly mapped, the skill will not function and will prompt for configuration checks.
• Automated file existence checks may not capture all forms of staleness (e.g., logical deprecation), so manual review is occasionally recommended.
• The analysis is primarily textual and pattern-based; deeper semantic analysis may require integration with additional AI skills or manual oversight.
• Regular use of Review helps maintain a high-quality, actionable memory base—critical for scalable, self-improving engineering teams.