Claude Code Plugins

Community-maintained marketplace

Feedback
29
0

Generate beads in bd using the repo's swarm-ready rules; explicit invocation only.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name gen-beads
description Generate beads in bd using the repo's swarm-ready rules; explicit invocation only.

Gen-Beads

Swarm-ready rules (surgeon mode)

  • One bead, one deliverable, one success state.
  • Smallest incision: no refactor unless required by outcome.
  • Parallel by default; add blocks only for true prerequisites.
  • Atomic scope: one agent, one sitting; split if it exceeds 1 day or >3 files.
  • Embed docs in each bead; no docs-only beads.
  • Acceptance is runnable: exact commands, file paths, expected results.
  • Epic cap: 7 max; split phases if more are needed.

Workflow

  1. Sanity check target
    • bd info --json
  2. Ensure memory template exists (seed if missing)
    • bd template list --json
    • If missing, create .beads/templates/memory.yaml:
      mkdir -p .beads/templates
      cat > .beads/templates/memory.yaml <<'YAML'
      name: memory
      description: |-
          ## Outcome (fill/update at close)
      
          [What changed? What is now true?]
      
          ## Why
      
          [Motivation / user story / constraints driving the change]
      
          ## Where (paths / systems)
      
          - [ ] Path(s):
          - [ ] Component(s):
      
          ## Verification (exact commands + results)
      
          - [ ] Command(s):
      
          ## Additional Context
      
          [Scratchpad while working. Distill into the bead before closing.]
      type: task
      priority: 2
      labels:
        - memory
      design: |-
          ## Decisions + rationale
      
          - [Decision]: [why]
      
          ## Alternatives considered
      
          - [Alt]: [why rejected]
      
          ## Constraints / invariants / gotchas
      
          - [Constraint]:
      
          ## Follow-ups
      
          - [ ] Created follow-up beads (if needed) and linked via discovered-from/blocks
      acceptance_criteria: |-
          - [ ] Outcome implemented
          - [ ] Verification commands recorded and pass
          - [ ] Bead fields are distilled (no placeholders)
          - [ ] Final "Closure Notes" saved in `notes` (per `AGENTS.md` / `~/.codex/AGENTS.md`)
      YAML
      
    • If .beads/templates/*.yaml is ignored, update .gitignore so the template can be committed.
    • Re-run bd template list --json to confirm.
  3. Draft the work graph
    • Cap epics at 7 for parallel workstreams; split phases if more are needed.
    • Define child beads that can be executed independently.
    • Mark only true prerequisites as blocks.
  4. Create beads
    • Prefer bd create --from-template memory ... --json.
    • If you cannot seed templates, fall back to template-free creation:
      • bd create -t <type> -p <priority> "<title>" --description "<...>" --design "<...>" --acceptance "<...>" --json
    • Use -t epic for epics and -t task (or feature/bug) for children.
    • Prefer hierarchical subtasks when coordination helps:
      • bd create --parent <epic-id> -t task "..." --description "<...>" --design "<...>" --acceptance "<...>" --json
    • Do not leave placeholders; distill fields immediately if needed.
  5. Wire dependencies
    • bd dep add <issue> <depends-on> -t blocks|related|parent-child|discovered-from --json
  6. Output
    • Print a compact index: epics → children, plus which beads are ready.
  7. Hard stop
    • Do not start implementation.