Back to list
thrashr888

development

by thrashr888

A Boss Repository Architecture for Multi-Repo AI Agent Orchestration

1🍴 0📅 Jan 20, 2026

SKILL.md


name: development description: AllBeads development process. Use when planning features, creating documentation, tracking work with beads, or understanding the documentation system. allowed-tools: Read, Edit, Write, Bash, Grep, Glob, Task, EnterPlanMode, ExitPlanMode, AskUserQuestion, TodoWrite

AllBeads Development Process Skill

Use this skill when planning features, creating documentation, tracking work with beads, or understanding how the documentation system works.

Quick Reference

Specs: specs/PRD-00.md - Core architecture specification Issues: .beads/ - Git-tracked issue database Plans: plans/ - Preserved plan history (git-tracked)

The 5-Style Documentation System

AllBeads uses 5 distinct doc types, each with a specific purpose:

StylePurposeLocationUpdate Trigger
1. Plan ModeDeep exploration before codingplans/Per feature
2. Beads IssuesTrack work across sessions.beads/Throughout feature
3. Evergreen SpecsSystem truth ("how it works")specs/Architecture changes
4. SkillsAgent guides ("how to implement").claude/skills/Pattern changes
5. User DocsHuman communicationREADME, CLAUDE.mdMajor features

Workflow: From Idea to Code

Step 1: Enter Plan Mode

For any non-trivial feature:

1. Use EnterPlanMode tool
2. Launch Explore agents to understand existing patterns
3. Read relevant specs (specs/PRD-00.md, specs/SPEC-*.md, etc.)
4. Ask clarifying questions via AskUserQuestion
5. Design implementation approach
6. Write plan to plan file
7. Exit plan mode with ExitPlanMode for approval
8. Copy approved plan to plans/ directory (see naming below)

Plan file naming: plans/YYYY-MM-DD-feature-name.md

Example: plans/2026-01-15-external-plugins.md

Plans are preserved even if abandoned - they document design decisions and exploration history.

Step 2: Create Beads Issues

After plan approval:

# Create feature epic
bd create --title="Feature X" --type=feature --priority=2

# Break into tasks
bd create --title="Implement X logic" --type=task --priority=2
bd create --title="Add X tests" --type=task --priority=2

# Set dependencies
bd dep add <task-id> <feature-id>  # Task depends on feature

# Claim work
bd update <id> --status=in_progress

Step 3: Implementation

During coding:

1. Read skills for "how to" guidance
2. Reference specs for architecture
3. Update beads (bd update <id> --status=in_progress)
4. Write code already formatted and linted (cargo fmt, cargo clippy)
5. Follow patterns from specs + skills
6. Run tests frequently (cargo test)

Step 4: Update Specs & Skills (if architecture changed)

- Specs: Declarative ("the Sheriff daemon syncs repos like this")
- Skills: Prescriptive ("implement a new sync adapter like this")

Step 5: Complete

# Verify quality gates
cargo fmt -- --check && cargo clippy -- -D warnings && cargo test

# Commit code
git add .
git commit -m "Implement Feature X"

# Sync beads
bd sync

# DO NOT close issues (user does after testing/pushing)
# DO NOT push to remote (user does)

Decision Framework

SituationAction
Starting new featureEnter plan mode
Feature spans sessionsCreate beads issue
New architectural patternUpdate spec in specs/
New implementation patternUpdate skill in .claude/skills/
Public-facing changeUpdate README.md
Agent guidance neededUpdate CLAUDE.md

Plan Preservation

Plans are git-tracked in plans/ to preserve institutional memory:

  • Approved plans - Document the "why" behind implementations
  • Abandoned plans - Show alternatives considered and why they were rejected
  • Partial plans - Capture exploration that may be revisited later

Plans complement beads issues: beads track what work was done, plans track how decisions were made.

When to save a plan:

  • After plan approval (before implementation)
  • When abandoning a plan (add note explaining why)
  • When significantly revising approach mid-implementation

Specs vs Skills

Specs say WHAT (evergreen truth):

# Sheriff Daemon Architecture

The Sheriff polls all registered Rigs for bead updates.
Shadow Beads MUST contain a pointer URI to the source bead.

Skills say HOW (implementation guide):

## Adding a New Integration

```rust
// Implement the Adapter trait
impl Adapter for JiraAdapter {
    async fn sync(&self) -> Result<Vec<ShadowBead>> {
        // Use ShadowBead::external() builder
        let shadow = ShadowBead::external(id, summary, uri)
            .with_status("open")
            .build();
        // ...
    }
}
```

Beads Essentials

Finding Work

bd ready               # Show issues ready to work (no blockers)
bd list --status=open  # All open issues
bd blocked             # Show blocked issues
bd show <id>           # View issue details

Managing Work

bd update <id> --status=in_progress  # Claim work
bd close <id>                        # Mark complete (ONLY after user tests)
bd sync                              # Sync with git remote

Dependencies

bd dep add <issue> <depends-on>  # Add dependency

Quality Gates

AllBeads requires ALL THREE to pass before commits:

cargo fmt -- --check   # Code formatting
cargo clippy -- -D warnings  # Linting
cargo test             # Tests

Run all three with:

cargo fmt -- --check && cargo clippy -- -D warnings && cargo test

Anti-Patterns

DON'T:

  • Write RFC-style docs that get stale ("We plan to add X")
  • Document temporary decisions in specs (use issues)
  • Put implementation details in specs (use skills)
  • Put architectural constraints in skills (use specs)
  • Close beads issues before user tests and pushes
  • Push to remote without user approval
  • Skip quality gates (fmt, clippy, test)

DO:

  • Use plan mode for all non-trivial features
  • Save plans to plans/ directory after approval (or abandonment)
  • Keep specs declarative and evergreen
  • Keep skills prescriptive and code-focused
  • Track multi-session work in beads
  • Update specs/skills when patterns change
  • Write formatted, linted code from the start
  • Run cargo test frequently during development

Issue Lifecycle Rules

CRITICAL: Follow this order:

  1. Code complete - Implementation is done, tests pass
  2. Commit code - Create git commit(s)
  3. Wait for user - DO NOT close issues yet
  4. User tests - Let user test the implementation
  5. User pushes - User pushes code to remote
  6. Then close issues - Only close after code is pushed and tested

Epics should only be closed after:

  • All child issues are closed
  • The feature has been tested by user
  • The code has been pushed to remote

Key Specs Reference

SpecPurpose
specs/PRD-00.mdCore architecture (Sheriff, Boss Board, Federated Graph)
specs/PRD-01-context-onboarding.mdContext and onboarding flows
specs/SPEC-handoff.mdAgent handoff patterns
specs/SPEC-governance.mdIssue governance model
specs/SPEC-aiki-integration.mdAiki integration spec

Score

Total Score

65/100

Based on repository quality metrics

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

+10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

1ヶ月以内に更新

+10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

+5

Reviews

💬

Reviews coming soon