pi-autobacklog — autonomous backlog processing for pi

Install · Usage · How it works

Plan the work, iterate through each item, fix, verify, commit, repeat.

Process a backlog of codebase tasks autonomously: deprecated API removal, test fixes, lint rule cleanup, and more. The agent scans your codebase, builds a backlog, then works through each item one by one — fixing, verifying, and committing as it goes.

---

What's included

Extension	log_item tool + live widget + /autobacklog command
Skill	Guides the agent through planning and processing the backlog

Extension tools

Tool	Description
log_item	Records item result, updates autobacklog.json, auto-commits or auto-reverts

/autobacklog command

Subcommand	Description
/autobacklog <text>	Enter autobacklog mode. If autobacklog.json exists, resumes. Otherwise, sets up a new session.
/autobacklog off	Leave autobacklog mode.
/autobacklog clear	Delete autobacklog files and turn off.

Examples:

/autobacklog find all tests calling deprecated create_record() and update them to use new_create()
/autobacklog fix all files with eslint no-unused-vars violations
/autobacklog update all imports from old-package to new-package
/autobacklog off

Keyboard shortcuts

Shortcut	Description
Ctrl+B	Toggle dashboard expand/collapse

UI

• Status widget — always visible: 📋 5/23 done 1 failed │ 17 remaining
• Expanded dashboard — Ctrl+B shows full item list with statuses, progress bar, and verification config

---

Install

pi install github.com/andrewn617/autobacklog

---

Usage

1. Start autobacklog

/autobacklog find all tests calling deprecated create_record() and update them

The agent will:

1. Scan the codebase to discover all affected files
2. Create autobacklog.md (problem description, fix approach, verification)
3. Create autobacklog.json (item list with verification config)
4. Start processing items immediately

2. The loop

For each pending item, the agent:

1. Reads the relevant code
2. Implements the fix
3. Runs the verification command (scoped to the item)
4. Calls log_item — which auto-commits on success or auto-reverts on failure
5. Moves to the next item

3. Monitor progress

• Widget — always visible above the editor
• Ctrl+B — expand to see full item list
• Escape — interrupt anytime

---

Files

File	Purpose
autobacklog.md	Problem description, fix approach, verification, constraints
autobacklog.json	Config + item list with statuses

autobacklog.json format

{
  "name": "Fix deprecated create_record() calls",
  "verifyCommand": "bin/rails test",
  "maxAttempts": 3,
  "items": [
    {"item": "test/models/user_test.rb:45", "status": "pending"},
    {"item": "test/models/post_test.rb:12", "status": "complete", "attempts": 1, "commit": "abc1234"},
    {"item": "test/controllers/api_test.rb:78", "status": "failed", "attempts": 3}
  ]
}

---

How it works

The agent handles both planning (scanning the codebase, building the backlog) and execution (fixing each item). The extension provides infrastructure: the log_item tool for git management, the widget for progress tracking, and system prompt injection for loop rules.

┌──────────────────────┐     ┌──────────────────────────┐
│  Extension (global)  │     │  Skill (per-session)      │
│                      │     │                           │
│  log_item tool       │◄────│  how to plan the backlog  │
│  widget + dashboard  │     │  how to fix each item     │
│  /autobacklog cmd    │     │  loop rules + recovery    │
│  auto-resume         │     │                           │
└──────────────────────┘     └──────────────────────────┘

Two files keep the session alive across context resets:

autobacklog.json  — item list with statuses, verification config
autobacklog.md    — problem description, fix approach, notes on what's been tried

A fresh agent with no memory can read these two files and continue exactly where the previous session left off.

---

License

MIT