reflect

name	reflect
description	Review the current conversation for learnings that should improve existing skills, AGENTS.md files or memory. Use at the end of a session to capture non-obvious findings before they're lost.

Goal: review this conversation for anything worth persisting: corrections, validated approaches, new conventions or structural gaps in the agent setup.

Phase 1: Identify candidates

Scan the conversation for:

• Corrections to your approach ("no, don't do X")
• Confirmed non-obvious choices (explicit confirmation or absence of pushback on an unusual approach)
• Patterns or constraints established during the session
• Gaps: something you had to discover that agent setup files should have covered

Phase 2: Check existing files

Avoid redundancy by reading:

• Shared AGENTS.md and its imports
• ../snap/conventions/ (all files)
• Local AGENTS.md
• Skills in ../snap/agents/skills/ that overlap in scope with the proposed addition
• Existing memory entries

Phase 3: Present findings

For each finding, identify where it belongs:

• Memory — user preferences or project context specific to this machine; not shared across machines, so avoid it for suite-wide conventions
• Local AGENTS.md — package-specific architecture or conventions
• Shared AGENTS.md — conventions that apply across all packages in the suite and all machines
• A skill — a reusable multi-step workflow that would benefit future sessions

Present findings using the standard format defined in AGENTS.md.

Constraints

• If nothing meaningful was learned, say so. Don't invent improvements.
• Don't add things already covered. Read the target file before proposing an addition.
• Prefer updating an existing entry over adding a new one.