checkpoint

name	checkpoint
description	Save session findings, decisions, and discoveries back to Stele shared memory
user-invocable	true

Checkpoint Session to Stele

Review the current conversation and persist important findings to Stele so other team members and future sessions benefit.

Procedure

Step 1: Determine Project Scope

Look for the project's Stele scope in the current project's CLAUDE.md — find the "Stele Shared Memory Protocol" section and extract the scope from the **Scope:** line. If not found, ask the user.

Step 2: Review Conversation

Scan the current conversation for:

1. Decisions made — architectural choices, technology selections, convention agreements
2. Bugs fixed — especially non-obvious root causes or gotchas
3. Conventions established — coding standards, naming patterns, process rules
4. Architecture changes — new components, modified dependencies, removed services
5. Contract changes — API modifications, env var changes, shared interface updates

Step 3: Classify and Store

For each finding, determine the storage type and apply appropriate tags:

Flat Memory (facts, notes, decisions)

store_memory(
  scope: "{scope}",
  title: "{concise title}",
  content: "{detailed description with context}",
  memory_type: "{knowledge|decision|convention|troubleshooting|reference}",
  tags: ["{appropriate tags}"]
)

Tagging convention:

Tag	Meaning
#active	Currently implemented and enforced rules
#todo	Technical debt or pending migrations
#contract	Inter-service API definitions and shared interfaces
#breaking	Changes that require other agents/services to update
#wisdom	Non-obvious technical discoveries and gotchas
#conflict	Local rule that conflicts with a workspace-level rule

Check the project entity in Stele for project-specific tags (seeded during /stele:bootstrap).

Knowledge Graph (things with relationships)

For new components or services discovered:

create_entities(
  entities: [{name: "...", entity_type: "...", observations: ["..."]}],
  scope: "{scope}"
)

For new observations about existing entities:

add_observations(
  entity_name: "...",
  scope: "{scope}",
  observations: ["..."]
)

For discovered relationships:

create_relations(
  relations: [{from: "...", to: "...", relation_type: "..."}],
  scope: "{scope}"
)

Step 4: Confirm

Present a summary of what was saved:

• Number of flat memories stored
• Number of entities created or updated
• Number of relations created
• List each item briefly (title + tags for memories, name + type for entities)

Ask the user if anything was missed or if they want to add additional context to any item.