save

nijaru's avatarfrom nijaru

Update ai/ files and tk tasks.

⭐3stars🔀0forks📁View on GitHub🕐Updated Jan 10, 2026

When & Why to Use This Skill

This Claude skill automates the process of checkpointing session states by updating project-specific documentation and task lists. It ensures that all progress, architectural decisions, and pending tasks are persisted in specialized 'ai/' files and 'tk' task managers, enabling seamless context resumption and preventing information loss during long-term development projects.

Use Cases

  • Session Persistence: Automatically saving the current state of a development project before ending a workday or switching to a different project to ensure no context is lost.
  • Context Compression: Preparing a project for context window management (such as the /compact command) by summarizing recent changes and updating status files to fit within LLM limits.
  • Task Synchronization: Reviewing git diffs and logs to automatically update task statuses in 'tk', marking completed items and adding newly discovered sub-tasks.
  • Automated Documentation: Keeping technical documentation like DESIGN.md, DECISIONS.md, and STATUS.md synchronized with the actual code changes and project evolution.
  • Project Handover: Generating a comprehensive report of what was accomplished, what is pending, and the current health of the codebase for a smooth transition between different AI agents or human developers.
namesave
descriptionUpdate ai/ files and tk tasks.
allowed-toolsRead, Write, Edit, Glob, Grep, Bash

Context Save

Checkpoint session state to ai/ and tk. Goal: everything needed to resume is persisted.

When to Trigger

  • Before /compact
  • Session end ("done for today", "wrap up")
  • Context switch ("switching to X", "different project")
  • After significant work completes
  • Explicit: "/save", "save context"

When NOT to Trigger

  • Mid-task (save at completion)
  • Quick fixes that don't change project state
  • User says "skip" or "don't save"

1. Review Session

What was accomplished? Run in parallel:

git diff --stat
git log --oneline -5
tk ls

Read ai/STATUS.md to compare against current state.

2. Update tk Tasks (be thorough)

Review ALL tasks against what was done:

tk ls  # Current state

For EACH task:

  • Completed → tk done <id>
  • New work discovered → tk add "title"
  • Blocked → Note in STATUS.md

Don't leave stale tasks. If done, mark done. If new, add it.

3. Update ai/ Files

ai/STATUS.md (always):

  • Current state, metrics
  • What worked / didn't
  • Active work
  • Blockers
  • Prune old content

ai/DESIGN.md (if architecture changed):

  • New components, patterns
  • Updated interfaces

ai/DECISIONS.md (if decisions made):

  • Context → Decision → Rationale (append)

ai/ROADMAP.md (if phases changed):

  • Phase updates, scope changes

ai/SPRINTS.md (if sprint progress):

  • Mark completed tasks
  • Update current sprint status

4. Health Check

Issue Fix
Files >500 lines Prune, move to subdirs
Stale STATUS.md Remove old blockers/work
Outdated DESIGN.md Update to current

5. Commit

git add ai/ .tasks/
git commit -m "Update session context"

6. Report

## Session Saved

Tasks:
- Done: [list marked complete]
- Added: [list new]
- Pending: [N remaining]

ai/:
- STATUS.md: [summary]
- DESIGN.md: [updated/unchanged]
- SPRINTS.md: [updated/unchanged]

Committed: [yes/no]

Ready for /compact or new session.