Skip to content

docs: align SKILL.md with agentskills.io best practices#3

Merged
sinzin91 merged 5 commits intomainfrom
feature/best-practices
Feb 5, 2026
Merged

docs: align SKILL.md with agentskills.io best practices#3
sinzin91 merged 5 commits intomainfrom
feature/best-practices

Conversation

@turian
Copy link
Copy Markdown
Member

@turian turian commented Jan 31, 2026

Summary

Align SKILL.md with agentskills.io spec and integrate the Ralph loop pattern into smithers-worker.

Changes

Frontmatter & Description

  • Fix description to third-person, trigger-only format ("Use when...")
  • Keep frontmatter to name + description only (per spec)

Progressive Disclosure

  • Extract smithers-worker agent definition to references/smithers-worker.md
  • SKILL.md now links to worker file instead of embedding inline

Ralph Loop Integration 🆕

  • Ralph Loop Pattern: Workers now iterate until all stop hooks (tests, lints, builds, type checks) pass
  • Stop Hooks: Defined quality gates (tests, lints, builds, types) that provide backpressure signals
  • Completion Promise: Workers only exit when all validations pass and requirements are met
  • Iterative Behavior: Detailed workflow showing the analyze → fix → retry loop
  • Architecture Comparison: Added table comparing depth-first Ralph vs breadth-first Smithers

New Sections

  • Add ## Progress Checklist for tracking workflow state
  • Add ## Configuration table with rationale for magic numbers
  • Add ## See Also section with Ralph pattern references

Compliance Fixes

  • Reduce DOT graph from 9 → 8 nodes (merge poll + check)
  • Update terminology from "beads" → "tasks" to match pluggable task sources
  • Remove redundant commentary

Architecture: Smithers vs Ralph

Smithers workers internally use Ralph loop discipline (iterate until stop hooks pass), while Smithers orchestrates multiple Ralph loops in parallel with review-based gating.

Aspect Ralph Smithers
Pattern Depth-first (one task, iterate until done) Breadth-first (parallel tasks, review gating)
Loop Scope Single task, infinite bash loop Multiple tasks, polling loop with gates
Worker Model Monolithic single process Parallel isolated workers

References

Test plan

  • Frontmatter has only name and description keys
  • ## Quick Reference and ## Failure Modes present
  • references/smithers-worker.md exists and includes Ralph loop pattern
  • SKILL.md includes architecture comparison and references
  • DOT graph ≤ 8 nodes
  • Description starts with "Use when" and is third-person
  • Workers iterate until stop hooks pass (Ralph loop discipline)
  • Completion promise documented

🤖 Generated with Claude Code

turian and others added 5 commits January 30, 2026 17:40
- Fix description to third-person, trigger-only format
- Extract smithers-worker agent to references/ (progressive disclosure)
- Add Progress Checklist for tracking workflow state
- Add Configuration table with rationale for magic numbers
- Remove redundant commentary

Reduces SKILL.md from 375 to 297 lines while improving discoverability
and compliance with repo best-practices doc.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Resolve conflicts while preserving best-practices compliance:
- Keep third-person, trigger-only description
- Keep progressive disclosure (external worker file)
- Keep Progress Checklist and Configuration sections
- Update terminology from "beads" to generic "tasks"

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Merge poll + check into single poll_check node to comply with
beads-skills best practices (≤8 nodes for DOT diagrams).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Integrate the Ralph loop pattern (iterative refinement driven by
backpressure) into smithers-worker specification:

- Add Ralph loop pattern description and principles
- Define stop hooks (tests, lints, builds, types) as quality gates
- Introduce completion promise concept
- Document iterative stop-hook behavior
- Update workflow to show Ralph loop phases
- Add architecture comparison (Smithers vs Ralph)
- Include references to Ralph pattern resources

Workers now iterate until all stop hooks pass before committing,
ensuring completion promise is fulfilled. Smithers orchestrates
multiple parallel Ralph loops with review-based gating.

Refs: https://ghuntley.com/ralph/
Refs: https://github.com/ghuntley/how-to-ralph-wiggum

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
@sinzin91 sinzin91 merged commit c721439 into main Feb 5, 2026
@sinzin91 sinzin91 deleted the feature/best-practices branch February 5, 2026 01:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants