AgentSkillsCN

memento

管理会话持久化的记忆数据——在执行“/clear”与“/compact”命令之间，保存并恢复上下文。

SKILL.md
--- frontmatter
type: skill
name: memento
description: Manages mementos for session persistence - save and restore context across /clear and /compact
version: 0.1.0
tags:
  - core
  - memory
  - context

Memento

Mementos are persistent context snapshots that help Claude resume work after /clear, /compact, or in new conversations. Each memento captures the problem, approach, progress, and next steps for a work session.

Activation

This skill activates when:

  • •User invokes /aida memento [create|update|read|list|remove|complete]
  • •User needs to save or restore session context
  • •After /clear when active mementos exist (for suggestion)

Command Routing

When this skill activates, parse the command to determine:

  1. •Operation: create, update, read, list, remove, complete
  2. •Arguments: slug, description, source, options

Create Operations

For create operations:

  1. •Read references/memento-workflow.md for the full workflow
  2. •Run Phase 1 to get questions and inferred data
  3. •Ask user any required questions using AskUserQuestion
  4. •Run Phase 2 to create the memento
  5. •Report success and next steps

Sources:

  • •manual (default): User provides description
  • •from-pr: Extract context from current PR
  • •from-changes: Summarize current file changes

Script invocation:

bash
# Phase 1: Get questions
python {base_directory}/scripts/memento.py --get-questions \
  --context='{"operation": "create", "description": "user description", "source": "manual"}'

# Phase 2: Execute
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "create", "slug": "my-memento", "description": "...", ...}'

Read Operations

For read operations:

  1. •Load the memento file
  2. •Return full content to Claude's context
  3. •Claude can then continue with the restored context

Script invocation:

bash
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "read", "slug": "my-memento"}'

List Operations

For list operations:

  1. •Scan .claude/mementos/ directory
  2. •Parse frontmatter for metadata
  3. •Return formatted list with status, description, dates

Script invocation:

bash
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "list", "filter": "active"}'

Update Operations

For update operations:

  1. •Load current memento
  2. •Ask which section to update (progress, decisions, next_step)
  3. •Merge new content with existing
  4. •Update the updated timestamp

Script invocation:

bash
# Phase 1: Get section options
python {base_directory}/scripts/memento.py --get-questions \
  --context='{"operation": "update", "slug": "my-memento"}'

# Phase 2: Execute update
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "update", "slug": "my-memento", "section": "progress", "content": "..."}'

Complete Operations

For complete operations:

  1. •Set status to completed
  2. •Move file to .claude/mementos/.archive/
  3. •Report completion

Script invocation:

bash
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "complete", "slug": "my-memento"}'

Remove Operations

For remove operations:

  1. •Confirm deletion with user
  2. •Delete the memento file
  3. •Report removal

Script invocation:

bash
python {base_directory}/scripts/memento.py --execute \
  --context='{"operation": "remove", "slug": "my-memento"}'

Path Resolution

Base Directory: Provided when skill loads via <command-message> tags.

Script Execution:

text
{base_directory}/scripts/memento.py

Reference Files:

text
{base_directory}/references/memento-workflow.md

Templates:

text
{base_directory}/templates/work-session.md.jinja2
{base_directory}/templates/freeform.md.jinja2

Memento Storage:

text
{project_root}/.claude/mementos/          # Active mementos
{project_root}/.claude/mementos/.archive/ # Completed mementos

Two-Phase API

This skill uses a two-phase API pattern:

Phase 1: Get Questions

Analyzes context and returns:

  • •Inferred metadata (slug, source, tags)
  • •Questions that need user input
  • •Validation results

Phase 2: Execute

Performs the operation with:

  • •User responses (if any)
  • •Inferred data
  • •Configuration options

Example Workflows

Creating a Memento

text
User: /aida memento create "fix auth token expiry"

1. Parse: operation=create, description="fix auth token expiry"
2. Run Phase 1:
   - Infer: slug="fix-auth-token-expiry", source="manual"
   - Question: "What's the core problem you're solving?"
3. User provides problem description
4. Run Phase 2:
   - Create .claude/mementos/fix-auth-token-expiry.md
5. Report: "Created memento. Start working, then /aida memento update to track progress"

Loading After /clear

text
User: /aida memento read fix-auth-token-expiry

1. Parse: operation=read, slug="fix-auth-token-expiry"
2. Load .claude/mementos/fix-auth-token-expiry.md
3. Return full content
4. Claude summarizes: "I see you were working on X. Next step was Y. Ready to continue?"

Creating from PR

text
User: /aida memento create from-pr

1. Parse: operation=create, source="from-pr"
2. Detect current branch, fetch PR via `gh pr view`
3. Extract title, body, files changed
4. Pre-fill template with PR context
5. Create memento with PR-derived content

Resources

scripts/

  • •memento.py - Main management script with two-phase API

references/

  • •memento-workflow.md - Detailed workflow documentation

templates/

  • •work-session.md.jinja2 - Primary template for task-focused mementos
  • •freeform.md.jinja2 - Simple template for custom content