skills.rest
kieranpotts
kieranpottsCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

reflect

Extract durable lessons from agent sessions and persist them to memory or convention files.

Updated Feb 12, 2026
Search Tags:#workflow#claude code#reflection#agent memory#knowledge persistence#session lessons#convention file
One-click install
npx skills add https://github.com/kieranpotts/skills --skill reflect-kieranpotts
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: reflect
Source: https://github.com/kieranpotts/skills/tree/main/skills/reflect
Command: npx skills add https://github.com/kieranpotts/skills --skill reflect-kieranpotts

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Valuable lessons learned during an agent session — corrections, validated approaches, user preferences, and project decisions — are lost when the session ends, forcing the next session to relearn them. This Skill captures those durable lessons and persists them to agent memory or the project's committed convention file so future sessions start better-informed. ## Core Features & Use Cases - Signal Detection: Scans the conversation for four lesson types: corrections, quietly accepted non-obvious choices, revealed working preferences, and project decisions not in version control. - Ruthless Filtering: Drops anything derivable from code, standard best practices, one-off details, or lessons already recorded — persisting nothing is a valid outcome. - Typed Destinations: Classifies each lesson as user, feedback, project, or reference (routed to agent memory) or as a codebase convention (routed to the committed convention file like AGENTS.md or CLAUDE.md). - Interactive Approval: Proposes each candidate one at a time with a draft entry and blocks on user approval before writing anything. - Use Case: At the end of a working session where the user corrected your commit message style twice and revealed a stakeholder deadline, run the skill to save those lessons to project memory so the next session applies them automatically. ## Quick Start Ask the agent to reflect on this session and save the lessons worth remembering.

Frequently Asked Questions about reflect

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I save lessons learned from an AI agent session?▼

Invoke the reflect skill at the end of a session by asking the agent to reflect on the session. It scans the conversation for corrections, preferences, and decisions, proposes each candidate lesson individually, and writes only the ones you approve to memory or the project convention file.

What is the difference between agent memory and a project convention file?▼

Agent memory is private to the agent and stores working-style lessons like user preferences and project context. The convention file (such as AGENTS.md or CLAUDE.md) is committed to the repository and holds codebase rules that all contributors, human and agent, need to see.

Can this skill run unattended or in CI pipelines?▼

No. The skill is interactive by design: it proposes each candidate lesson one at a time and blocks on user approval before persisting anything. It is not suitable for unattended or CI use.

When should I run session reflection versus a handoff?▼

Run reflection at the end of a session to capture durable lessons about how to work with the user and project. Use a handoff skill instead when you need to record where a task got to so work can resume later; reflection deliberately does not record task state.

What happens if a session produced nothing worth saving?▼

The skill reports that nothing durable was learned and stops. It explicitly does not lower its filter to manufacture lessons, and an empty result is considered a correct outcome.

Does the skill protect sensitive data when saving memories?▼

Yes. It redacts aggressively, stripping API keys, tokens, real names, personal data, and internal-only URLs before persisting any entry, since memory persists well beyond the session that wrote it.

Related Skills

Explore More Software Engineering Skills→
adobeadobe
1.5k

session-retrospective

Capture session corrections and lessons into .ai/memory/lessons.md.

Official
Advanced
mi179mi179
1

context-save

Capture git state, decisions, and remaining work into a repeatable checkpoint.

Community
Advanced
MohabsmarMohabsmar

elite-longterm-memory

Implement layered memory with LanceDB, Git-notes, and Write-Ahead Logging.

Community
Advanced
kevinnftkevinnft
13

context-engineering

Load project-wide context rules, specs, and source files for agent sessions.

Community
Advanced