weegigs

execute

Execute an existing plan aggressively with minimal interruption. Use after /kick-off approval to implement the planned work.

weegigs 3 Updated 4mo ago

Resources

1
GitHub

Install

npx skillscat add weegigs/claude-kitbash/execute

Install via the SkillsCat registry.

SKILL.md

Workflow Execute

Aggressive execution mode. Find plan, work through tasks, minimize interruptions.

Mindset

See references/mindset.md for the "hard-nosed perfectionist" philosophy.

Key principles:

  • Every TODO is a lie — complete now or get approval
  • Workarounds compound — fix root causes
  • Scope is sacred — only user can reduce it
  • Done means done — partial completion is not completion

Mode: Ultrawork

  • Execute in dependency order, never skip ahead
  • Parallelize independent work where possible
  • Ask questions only when truly blocked
  • Run verification at checkpoints, not after every change
  • Code review required before commits

Task Tracking

Detect at start:

command -v bd &> /dev/null && echo "beads" || echo "todowrite"

See references/task-fallback.md for TodoWrite fallback patterns.

Knowledge Base

Detect ByteRover:

command -v brv &> /dev/null && [ -d ".brv" ] && echo "byterover" || echo "none"
  • byterover available: Curate discoveries after significant work
  • byterover unavailable: Skip curation steps

Process

Phase 1: Locate Plan

  1. Find active plan (check in order):

    • TodoWrite from /kick-off
    • In-progress tasks: bd list --status=in_progress (or TodoWrite)
    • Recent specs: .agent-os/specs/
    • Ready tasks: bd ready
  2. If no plan found — recommend /kick-off

  3. Load context:

    • Read TodoWrite for structure
    • Note parent epic — new tasks belong to it

Phase 2: Assess State

  1. Map execution state:

    Plan: [title]
    
    Completed:  ✓ task-1, task-2
    In Progress: → task-3
    Remaining:  ○ task-4, task-5
    Blocked:    ⊘ task-6 (waiting on task-4)
  2. Inject standards and skills (once at start):

    Standards (if agent-os configured):

    If .agent-os/standards/ exists:
      Invoke /standards-inject (auto-suggest mode)

    Language skills — detect from files to modify:

    • .rs → Load @rust (+ @tokio if async)
    • .ts, .tsx → Load @typescript
    • .svelte → Load @svelte + @typescript

    Always load @principles for quality standards.

Phase 3: Execute

  1. Work loop — repeat until complete:

    a. Select next task

    • Continue in-progress if exists
    • Otherwise pick highest-priority unblocked
    • Claim task (beads: bd update --claim, TodoWrite: mark in_progress)

    b. Execute task

    • Read acceptance criteria
    • Implement without asking for confirmation
    • Make decisions autonomously when reasonable
    • Only ask when genuinely blocked

    c. Verify task (lightweight)

    • Run lint, typecheck
    • Check acceptance criteria met
    • Fix issues immediately

    d. Code review (required before commit)

    Run in parallel:

    1. /codex-review — if Codex CLI available
    2. @reviewer agent — always (uses @principles + language skills)

    Both must pass. If Codex unavailable, @reviewer agent alone is sufficient.

    Address all issues before committing.

    e. Complete task

    • Commit: jj split -m "description" .
    • Close task (beads: bd close, TodoWrite: mark completed)
    • Update TodoWrite

    f. Check for checkpoint

    • If checkpoint task, run full /check
    • Otherwise continue to next
  2. Handle blockers:

    • Technical → attempt 2x, then ask
    • Missing info → ask concisely, continue other work
    • External dependency → note it, move to next task
  3. Handle discovered work:

    • See references/mindset.md for protocol
    • Never add TODO without user approval
    • Pause and explain what you found
    • If creating task: add to epic

Phase 4: Completion

  1. Final verification:

    • Run /check
    • Ensure all tasks closed including parent
    • Report completion summary
  2. Curate discoveries (if byterover available):

    If significant discoveries were made during execution:

    brv curate "Discovery context" -f relevant/file.rs

    Worth curating:

    • Bug root causes and non-obvious fixes
    • Architectural decisions and rationale
    • Gotchas that would save future time
    • Domain knowledge not obvious from code

    Skip if work was routine implementation.

  3. Completion report:

    Plan complete: [title]
    
    Executed: ✓ all tasks
    
    Commits:
      abc1234 Added user_preferences migration
      def5678 Implemented repository
    
    All verification passed.

Cop-Out Patterns

See references/cop-outs.md for patterns to avoid.

Never do:

  • Add TODO without asking
  • Use as any or lint suppressions
  • Say "out of scope" without user approval
  • Skip tests or swallow errors

Quick Reference

Operation With beads Without beads
Show task bd show <id> Read TodoWrite
Start task bd update --claim Mark in_progress
Complete bd close <id> Mark completed
Add task bd create --parent=<epic> Add to TodoWrite
Commit jj split -m "msg" . Same
Full verify /check Same