grind

name	grind
description	Autonomous bead grinding loop — picks open beads one by one, validates relevance, implements, tests, commits, and pushes. No human interaction needed. Use when user says /grind, "grind beads", "work through issues", "implement beads", or "clear the backlog". Runs until all ready beads are done or context runs low.

Grind

Autonomous loop: pick bead, validate, implement, test, commit, push. Repeat.

Rules

1. No subagents — do everything in the main session
2. No human interaction — make decisions autonomously
3. Scope discipline — if you discover tech debt, unrelated bugs, or scope creep while working, DO NOT fix it. Instead: bd create --title="..." --type=bug --priority=3 and move on
4. Follow codebase style — match existing patterns, imports, naming. Read files before modifying.
5. Verify before closing — run tests/linters, confirm code works. Never skip.
6. Commit after each bead — atomic commits, one bead per commit

Loop

1. bd ready                          # get available beads
2. Pick highest-priority bead
3. bd show <id>                      # read full details
4. ANALYZE: Is this bead still relevant?
   - Check if the code/feature it references still exists
   - Check if another bead already fixed it
   - If irrelevant → bd close <id> --reason="no longer relevant" → goto 1
5. bd update <id> --claim
6. IMPLEMENT:
   - Read relevant source files first
   - Follow existing code style and patterns
   - Add/update unit tests if the change is testable
   - Keep changes minimal and focused
7. VERIFY:
   - Run: pytest <relevant_test_files> (or full suite if unsure)
   - Run: ruff check <changed_files>
   - If tests/lint fail → fix → re-verify
   - Do NOT skip this step
8. COMMIT & PUSH:
   - git add <specific_files>
   - git commit -m "<type>: <longer description about the changes> (bd-<short_id>)"
   - bd close <id>
   - git push
9. goto 1

Stopping Conditions

Stop the loop when:

• bd ready returns no beads
• Context window is getting full (>80% used) — push remaining work and summarize
• A bead requires external input or decisions beyond scope — skip it, note in summary

Bead Selection

Priority order:

1. Bugs (highest priority first)
2. Tasks
3. Features
4. Chores

Skip beads that:

• Require UI/frontend changes if you're unsure about the design
• Need external API keys or services not available
• Are epics (too large for a single grind iteration)

Quality Gates

Before closing any bead:

• Changed Python files pass ruff check
• Relevant tests pass (run specific test files, not always full suite)
• No import errors in changed modules
• If you added a function/class, it has basic test coverage

Commit Message Format

<type>: <longer description about the changes and why> (bd-<short_id>)

Types: fix, feat, refactor, chore, test, docs

Example: fix: handle None in sharpe formatting (bd-k4ya)

Summary

After the loop ends, output a markdown summary:

## Grind Summary

### Completed

| Bead   | Title          | Type | Changes           |
| ------ | -------------- | ---- | ----------------- |
| bd-xxx | Fix None crash | bug  | vibe_quant/foo.py |

### Skipped

| Bead   | Title       | Reason             |
| ------ | ----------- | ------------------ |
| bd-yyy | Redesign UI | needs design input |

### New Issues Filed

| Bead   | Title          | Found While         |
| ------ | -------------- | ------------------- |
| bd-zzz | Tech debt in X | implementing bd-xxx |

### Stats

- Completed: N
- Skipped: N
- New issues: N
- Commits pushed: N