kiyoraka/manage-project

Manage Project -- LRU Project Management Skill

Smart project tracking with automatic memory management.

Activation

When this skill activates, silently determine which command was triggered and execute the matching protocol section.

Context Guard

| Context | Status | |---------|--------| | User says "new project [name]" | ACTIVE -- create project | | User says "load project [name]" | ACTIVE -- load and resume | | User says "save project" | ACTIVE -- save current progress | | User says "list projects" | ACTIVE -- display all projects | | Mid-conversation (no project command) | DORMANT |

---

Protocol: New Project

Trigger: "new project [name]" or "create project [name]"

Step 1: Parse Command

• [ ] Extract project name from command
• [ ] Check if project name already exists in active or archived
• [ ] If exists: suggest loading instead, or ask for a different name

Step 2: Gather Project Details

• [ ] Ask user for brief description (1-2 sentences)
• [ ] Get current date using platform-appropriate time command
• [ ] Note any initial requirements, goals, or tech stack

Step 3: Create Project File

• [ ] Generate project file from Embedded Project Template (see below)
• [ ] Fill in: name, description, created date, status (Active)
• [ ] Save to projects/active/[name].md

Step 4: Apply LRU Positioning

• [ ] Insert new project at position #1
• [ ] Shift all existing active projects down by 1
• [ ] If position #11 exists after shift:
  • Move to projects/archived/
  • Update status to "Archived (LRU)"
  • Add archive date

Step 5: Update Project List

• [ ] Regenerate projects/project-list.md (see Project List Format)
• [ ] New project at top of active section
• [ ] Note any auto-archived project

Step 6: Update Session Memory

• [ ] Add to current-session.md:

  ## Active Project
  - Name: [project name]
  - Started: [date/time]
  - Context: [description]
  

Step 7: Confirm

• [ ] Display confirmation with project name, position #1, and description

---

Protocol: Load Project

Trigger: "load project [name]" or "resume project [name]"

Step 1: Search for Project

• [ ] Parse project name from command
• [ ] Search projects/active/ first
• [ ] Then search projects/archived/
• [ ] Use fuzzy matching if exact name not found (partial names, case-insensitive)

Step 2: Handle Search Results

• [ ] If multiple matches: show list, ask for selection
• [ ] If in archived: note that project will be reactivated
• [ ] If not found: suggest similar projects, offer to create new

Step 3: Load Project Data

• [ ] Read project file completely
• [ ] Extract: description, last accessed, status, recent sessions, duration

Step 4: Apply LRU Positioning

• [ ] If archived: move from archived/ to active/
• [ ] Move project to position #1
• [ ] Shift all other active projects down
• [ ] If position #11 exists: auto-archive

Step 5: Update Project File

• [ ] Update "Last Accessed" to current date/time
• [ ] Add session note to progress:

  ### [Current Date]
  - Project resumed from [previous position/archived]
  

Step 6: Update Project List

• [ ] Regenerate projects/project-list.md

Step 7: Load into Session Memory

• [ ] Update current-session.md:

  ## Active Project
  - Name: [project name]
  - Resumed: [date/time]
  - Last worked: [previous date]
  - Context: [description]
  - Recent progress: [last 3 session entries]
  

Step 8: Display Summary

• [ ] Show: name, position #1, last worked date, description, recent activity

---

Protocol: Save Project

Trigger: "save project" -- saves current project only, NOT AI memory

Step 1: Identify Active Project

• [ ] Check current-session.md for active project
• [ ] If no active project: inform user, suggest loading or creating one
• [ ] If project exists: proceed to save

Step 2: Gather Session Progress

• [ ] Collect work done in current session
• [ ] Capture code changes, decisions, milestones
• [ ] Note issues resolved or discovered
• [ ] Document resources or references added

Step 3: Parse Duration

• [ ] Read recent commit messages from session: git log --oneline --format="%s%n%b" -20
• [ ] Search for Time: field in commit bodies (from Auto-Commit System)
• [ ] Sum all Time: ~XX min values for this session
• [ ] Add to project's accumulated duration
• [ ] Format using Duration Thresholds (see below)

Step 4: Update Project File

• [ ] Load project file from projects/active/[name].md
• [ ] Update "Last Accessed" to current date/time
• [ ] Update "Duration" with accumulated total
• [ ] Update "Current Status" section
• [ ] Add session entry to Session History:

  ### [Current Date] - [Session Title]
  - **Changes**: [Summary of session work]
  - **Time Spent**: ~XX min
  

Step 5: Line Limit Check

• [ ] Count lines in project file
• [ ] If over 1000 lines: run Line Limit Enforcement (see below)
• [ ] Verify file is under 1000 lines after enforcement

Step 6: Update Project List

• [ ] Regenerate projects/project-list.md with updated timestamps

Step 7: Confirm Save

• [ ] Display: project name, saved timestamp, session summary, total duration

---

Protocol: List Projects

Trigger: "list projects" or "show projects"

Step 1: Read Project List

• [ ] Load projects/project-list.md
• [ ] Parse active projects (positions 1-10) and archived projects

Step 2: Display

• [ ] Show active projects with: position, name, last updated, status
• [ ] Show archived project count and names
• [ ] Highlight currently loaded project (if any)

---

LRU Engine

Rules

• Fixed Capacity: 10 active projects (positions 1-10)
• Position-Based: Projects ranked by most recent access/creation
• Auto-Archiving: Position #11 gets archived when new project added or existing one loaded
• Reactivation: Archived projects can be loaded back (replaces position #10, archives it)

Operations

Move to Position #1:

1. Identify project location (active position or archived)
2. Remove from current position
3. Insert at position #1
4. Shift all others down by 1
5. Archive position #11 if it exists
6. Regenerate project-list.md

Auto-Archive:

1. Move project file from active/ to archived/
2. Update project metadata: status = "Archived (LRU)", archive date added
3. Update project-list.md

Project List Format

# Project Portfolio - Dynamic List
*Auto-generated by LRU Project Management System*

## Active Projects (X/10)

| Pos | Project | Last Updated | Status |
|-----|---------|--------------|--------|
| 1 | **[Project Name](./active/name.md)** | [Date] | [Brief status] |
| 2 | ... | ... | ... |

## Archived Projects (X)

| Project | Status |
|---------|--------|
| **[Project Name](./archived/name.md)** | Archived (LRU) - [Date] |

---
## System Status
- **Total Projects**: X (Y active + Z archived)
- **Last Updated**: [Date] - [Action taken]

---

Duration Tracking

Parsing

• Read commit messages for Time: ~XX min or Time: XX min patterns
• Source: Auto-Commit System's SESSION CONTEXT section
• Sum all time values from commits made during the current session

Accumulation

• Each "save project" adds session time to the project's total Duration
• Duration stored in project file's Overview section

Display Thresholds

| Total Minutes | Display Format | Example | |---------------|----------------|---------| | 1-59 | X min | 45 min | | 60-1439 | X hours | 3 hours | | 1440-43199 | X days | 2 days | | 43200+ | X months | 1 month |

• Round to nearest 0.5 for clean display (e.g., "2.5 hours", "1.5 days")
• Conversion: 60 min = 1 hour, 1440 min = 1 day (24h), 43200 min = 1 month (30d)

---

Line Limit Enforcement

Maximum: 1000 lines per project file

When to Check

• Before completing any "save project" operation
• After adding new session content

When project file exceeds 1000 lines:

Step 1: Summarize Old Sessions

• [ ] Identify sessions older than last 5
• [ ] Extract key info: dates, major changes
• [ ] Create/update Historical Summary paragraph
• [ ] Remove detailed old session entries

Step 2: Trim Verbose Sections

• [ ] Keep only: Overview, Status, Sessions (5), History, Notes
• [ ] Remove any extra sections not in the template

Step 3: Verify

• [ ] Count lines after summarization
• [ ] If still > 1000: repeat summarization more aggressively
• [ ] Ensure file is under 1000 lines

Historical Summary Format

## Historical Summary
Project started [Start Date] as [initial scope]. Over [X] sessions spanning
[timeframe], developed [major features/changes]. Key milestones: [milestone 1],
[milestone 2], [milestone 3]. Previous sessions covered: [brief theme list].

---

Embedded Project Template

When creating a new project, use this template:

# [Project Name] - [Brief Description]
*[One-line project summary]*

## Project Overview
- **Type**: [Category: Web App/API/Library/Game/Tool/etc.]
- **Period**: [Start Date] - Active
- **Tech Stack**: [Backend] + [Frontend] + [Database]
- **Completion**: 0%
- **Duration**: 0 min

## Current Status
- **Last Session**: [Date] - Project created
- **Next Steps**: [Initial goals]
- **Known Issues**: None

## Session History (Last 5)

### [Date] - Project Created
- **Changes**: Initial project setup
- **Time Spent**: ~0 min

## Historical Summary
[No history yet -- this section is populated when session count exceeds 5]

## Technical Notes
- **Repository**: [Git URL or local path]
- **Key Dependencies**: [Critical packages or services]

---
**Last Updated**: [Date] | **Position**: #1/10 Active

---

Mandatory Rules

1. 10 active slots -- auto-archive at position #11. No exceptions
2. project-list.md regenerated after every LRU operation (new, load, save)
3. Line limit 1000 -- auto-summarize when exceeded, keep last 5 sessions
4. Duration tracking -- parse Time: from Auto-Commit messages, accumulate per project
5. Generic template -- no type-specific fields. One universal project format
6. "save project" is project-only -- does NOT save AI memory, personality, or preferences
7. Explicit save only -- no auto-save. User controls when to save
8. Fuzzy search on load -- partial names, case-insensitive matching
9. Preserve data on archive -- archived projects retain all history and can be reloaded
10. Command separation -- save = AI memory (handled elsewhere), save project = project progress

Edge Cases

| Situation | Behavior | |-----------|----------| | No active project on "save project" | Inform user, suggest load or create | | Project name already exists on "new" | Suggest loading existing or choosing different name | | Project not found on "load" | Show available projects, offer fuzzy matches | | Archive folder has 50+ projects | Suggest cleanup of old archives | | No Auto-Commit time data available | Use ~XX min estimate or ask user | | Multiple projects match search | Display list with positions and dates, ask for selection |

Synergy with Other Features

| Feature | Integration | |---------|-------------| | Auto-Commit System | Duration tracking parses Time: from commit messages | | Save Diary System | Diary captures daily narrative; project captures project-specific progress | | Reminders System | "Revisit this project next week" becomes a reminder | | Decision Log System | Project decisions can reference the decision log |

Level History

• Lv.1 -- Base: new/load/save/list commands, LRU engine (10 slots), auto-archiving, session history, project-list.md auto-generation. (Origin: Absorbed from separate protocol files + adapted from production AI companion project manager v3.1)
• Lv.2 -- Duration Tracking: Parse Auto-Commit time estimates, accumulate per project, progressive display format. Line Limit Enforcement: 1000-line cap with auto-summarization.