Skip to content

feat(workflow): implement retry tracking and escalation flow#12

Merged
jeffwweee merged 2 commits intomainfrom
feature/commander-upgrade
Mar 4, 2026
Merged

feat(workflow): implement retry tracking and escalation flow#12
jeffwweee merged 2 commits intomainfrom
feature/commander-upgrade

Conversation

@jeffwweee
Copy link
Copy Markdown
Owner

Summary

Implements comprehensive retry tracking and escalation workflow to prevent infinite loops and ensure user awareness of persistent failures.

Changes

Workflow Skills

  • reviewer/SKILL.md: Add 3-retry limit with escalation, update decision matrix with retry column
  • background-tasks/SKILL.md: Add escalation triggers (3 consecutive failures, critical errors)
  • brainstorming/SKILL.md: Add explicit confirmation step before proceeding to implementation
  • subagent-driven-development/SKILL.md: Add post-verification step with retry handling

Infrastructure

  • scripts/task-state.sh: Add retry tracking API (increment_retry, get_retry_count, reset_retry)
  • references/workflow.md: Create comprehensive workflow documentation with failure handling

Workflow Flow

User → Commander → Brainstorm → Confirm → Design → Plan → Background
→ Subagent → Review → Verify → Complete
                 ↑               ↓           ↑            ↑
                 └─── 3 retries ←─┘           └── Escalate ←─┘

Test Plan

  • Verify retry tracking functions work correctly
  • Test escalation after 3 failures
  • Test "Try different approach" loops back to brainstorming
  • Verify explicit confirmation in brainstorming
  • Test plan verification integration

Token Efficiency

Main workflow skills: ~2,000-2,500 tokens per message (down from ~5,000 before optimization)

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

jeffwweee and others added 2 commits March 4, 2026 11:32
Refined 8 core skills from "superpower" context to minimal Pichu architecture:

- commander: 329 → 123 lines (63% reduction), added "never implement yourself" rule
- verification: 58 → 35 lines (40% reduction), removed redundant "Iron Law"
- brainstorming: 54 → 46 lines (15% reduction), added approval signals
- using-skill: 47 → 42 lines (11% reduction), trimmed verbose description
- background-tasks: added error handling (40 → 43 lines)
- writing-plans: added plan review checklist (27 → 30 lines)
- subagent-dev: added "why fresh" and "when not to use" (43 → 48 lines)
- reviewer: maintained lean structure (44 lines)

Created references/ folder structure for detailed documentation:
- Plan templates, handoff scripts, task templates
- Telegram format examples, verification patterns
- Review checklists, confidence scoring, notification templates

Reduced identity.md from 24 → 2 lines (92% reduction).

Total estimated savings: ~3,000 tokens per message (65% reduction)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add comprehensive retry tracking and escalation workflow to prevent
infinite loops and ensure user awareness of persistent failures.

Changes:
- reviewer: Add 3-retry limit with escalation, update decision matrix
- background-tasks: Add escalation triggers and failure tracking
- brainstorming: Add explicit confirmation step before proceeding
- subagent-dev: Add post-verification step with retry handling
- task-state.sh: Add increment_retry, get_retry_count, reset_retry API
- references/workflow.md: Create comprehensive workflow documentation
- project-status.md: Clear active work status

Workflow flow:
User → Commander → Brainstorm → Confirm → Design → Plan → Background
→ Subagent → Review → Verify → Complete
                 ↑               ↓           ↑            ↑
                 └─── 3 retries ←─┘           └── Escalate ←─┘

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@jeffwweee jeffwweee merged commit 35bd4f5 into main Mar 4, 2026
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.

1 participant