Agent-Skills.md

Agent Skills: Session Review

>-

UncategorizedID: philoserf/claude-code-setup/session-review

Repository

philoserf/claude-code-setup
philoserfLicense: MIT
9

Install this agent skill to your local

pnpm dlx add-skill https://github.com/philoserf/claude-code-setup/tree/HEAD/skills/session-review

Skill Files

Browse the full folder contents for session-review.

Download Skill

Loading file tree…

skills/session-review/SKILL.md

Skill Metadata

Name
session-review
Description
Analyzes the current session to extract patterns, preferences, and learnings. Use for session retrospectives, debriefs, post-mortems, or reflecting on insights worth remembering. Produces structured reviews capturing what worked, what to improve, and actionable takeaways.

Session Review

Comprehensive session analysis to build cumulative knowledge across interactions.

Objective

Extract reusable insights from the session that make future sessions more effective. Focus on patterns, not just facts.

When to Use

  • After significant debugging or problem-solving sessions
  • When you've learned something important about the codebase
  • After discovering user preferences through trial and error
  • When system relationships became clearer through investigation

When NOT to Use

  • Short, straightforward sessions with no corrections or surprises
  • When the user just wants a quick task done, not a retrospective
  • Use md-improve instead if the only goal is updating CLAUDE.md

Analysis Dimensions

| Dimension | Focus | | -------------------- | ------------------------------------------------------------ | | Problems & Solutions | Symptoms, root causes, dead ends, key insights | | Code Patterns | Design patterns, naming, data flow, error handling | | User Preferences | Explicit and implicit preferences across tools/style/process | | System Understanding | Components, dependencies, failure modes | | Knowledge Gaps | Misunderstandings, missing info, better approaches |

See analysis-dimensions.md for the full framework with questions, formats, and examples.

Process

  1. Review - Walk through the session conversation
  2. Extract - Identify insights in each dimension
  3. Synthesize - Connect related learnings
  4. Document - Create structured reflection
  5. Act - Generate concrete deliverables
  6. Save - Write the review to Obsidian (see below)

Deliverables

Based on the analysis, generate applicable items:

  • CLAUDE.md updates - Preferences and patterns to remember
  • Code comments - System understanding to preserve
  • Documentation - Workflows or processes to document
  • Future considerations - Things to address in later sessions
  • Obsidian note - Every session review is saved to the vault

Obsidian Storage

After presenting the review, save it to Obsidian using the obsidian CLI (invoke the obsidian-cli skill for reference):

obsidian create path="Session Reviews/YYYY-MM-DD <short description>.md" content="<review content>" silent
  • Title: # Session Review: <short description>
  • Date and repo/project as metadata at the top
  • Use the same markdown content shown to the user
  • Do not ask for confirmation — just save it

Verification

  • Review covers all 5 dimensions (or explicitly notes "N/A" for dimensions with no findings)
  • Each insight is grounded in a specific conversation moment, not hypothetical
  • CLAUDE.md updates are diffed before applying
  • Obsidian note was saved successfully (confirm with obsidian read)

Guidelines

  • Focus on reusable patterns, not session-specific facts
  • Capture the "why" behind decisions, not just the "what"
  • Preserve user voice when documenting preferences
  • Prioritize insights by impact on future effectiveness
  • Build cumulative knowledge, not just session notes

Reference Files