Menu
Execute implementation tasks with different approaches (single, dual-agent, plan-based).
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/Yuvasee/samocode --skill implementationThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Code quality analysis through cleanup analysis, multi-perspective reviews, and explicit finding decisions.
Interactive task definition with Q&A and documentation.
Extract durable lessons from the current conversation, classify them, and route each to the narrowest enforceable code, test, skill, or scoped instruction source without growing a generic CLAUDE.md memory dump.
Showing SKILL.md
| name | implementation |
| description | Execute implementation tasks with different approaches (single, dual-agent, plan-based). |
Executes implementation tasks using different approaches: direct execution, dual-agent comparison, or plan-based execution.
These steps apply to ALL actions after implementation is complete.
Write the fewest comments possible; the shorter, the better. A comment earns its place only by carrying non-obvious WHY the code cannot — a constraint, invariant, tradeoff, workaround, or surprising ordering — in one short line, essence only. Before committing, delete:
_overview.md.samocode file for MAIN_REPO, or use git root, or ask usernpx tsc --noEmit (it OOMs on large codebases)pr-readiness phase before done
_review_debt.md has undecided blocking/important rows_overview.md: Add Flow Log entry and Files entrycd [SESSION_DIR] && git add . && git commit -m "[action]: [description]"- [x]**Phase completed** ([TIMESTAMP_ITERATION])IMPORTANT: For implementation phases, use dop2 (dual-agent comparison).
DO NOT implement directly with Edit/Write until the comparison step is complete.
Rule of thumb: If you're considering dop, ask "Could there be a cleaner way?" If yes -> use dop2.
Execute a task directly with full implementation and documentation.
Session path: [SESSION_PATH from working memory] Task: $ARGUMENTS
Investigate and implement:
Document work - Create [SESSION_PATH]/[TIMESTAMP_FILE]-do-[task-slug].md:
# Task: [brief title]
Date: [TIMESTAMP_LOG]
## What was done
[Description of changes]
## Files Changed
- [file] - [what changed]
## Testing
[How verified]
## Notes
[Any important observations]
Complete common steps (see above)
Report back: Summary of what was done
Spawn two agents with different philosophies to solve the same task independently. Compare solutions and present options.
Session path: [SESSION_PATH from working memory] Task: $ARGUMENTS
Create 2 independent solution proposals. If the Task tool is available, spawn 2 sub-agents in parallel (model: sonnet). If the Task tool is not available, run the two proposal passes yourself sequentially and keep them independent.
Both agents solve the entire task independently with different philosophies.
Agent 1: Minimal Footprint
Session path: [SESSION_PATH] Task: $ARGUMENTS
Approach: Smallest possible change surface.
Deliverables:
Create [SESSION_PATH]/[TIMESTAMP_FILE]-solution-minimal.md:
# Solution: Minimal Footprint
Task: [description]
## Approach
[High-level strategy]
## Proposed Changes
### File: [path]
```[language]
[code]
Why: [rationale]
[What we accept]
DO NOT edit actual code files - proposals only
Report summary
Agent 2: Clean Foundation
Session path: [SESSION_PATH] Task: $ARGUMENTS
Approach: The "right way" with time to do it properly.
avoncore only when there are 2+ current Python service consumers. A frontend mirror does not count; a future consumer belongs in a ticket/TODO, not a premature shared abstraction.Deliverables:
Create [SESSION_PATH]/[TIMESTAMP_FILE]-solution-clean.md:
# Solution: Clean Foundation
Task: [description]
## Approach
[High-level strategy]
## Proposed Changes
### File: [path]
```[language]
[code]
Why: [rationale]
[New patterns, benefits]
DO NOT edit actual code files - proposals only
Report summary
After both agents complete:
Review both solutions
Create [SESSION_PATH]/[TIMESTAMP_FILE]-comparison.md:
# Comparison: [task]
## Minimal Footprint
**Pros:** [list]
**Cons:** [list]
## Clean Foundation
**Pros:** [list]
**Cons:** [list]
## Recommendation
[Which approach and why]
## Hybrid Possibility
[Can we combine best of both?]
Complete common steps for session files
Present to user:
Which approach?
1. Minimal footprint
2. Clean foundation
3. Hybrid
4. Neither (rethink)
Execute a specific phase from the plan without comparing approaches.
Only for TRIVIALLY SIMPLE 1-2 line changes. For anything else, use dop2.
Session path: [SESSION_PATH from working memory] Phase: $ARGUMENTS
IMPORTANT: Only work on this specific part. Don't do other parts of the plan.
Find the plan:
_overview.md -> Plans section for plan fileImplement:
Document work - Create [SESSION_PATH]/[TIMESTAMP_FILE]-dop-[phase-slug].md:
# Phase: [phase name]
Date: [TIMESTAMP_LOG]
Plan: [plan filename]
## Completed Items
- [x] [Item from plan]
- [x] [Item from plan]
## Changes Made
- [file] - [what changed]
## Testing
[How verified]
## Notes
[Any issues or observations]
Complete common steps (lint, commit code, update plan, update session, commit session)
Report back: Summary of completed items
DEFAULT IMPLEMENTATION STRATEGY - Use two independent proposal passes before editing code.
Execute a plan phase using dual-agent comparison, with automatic solution selection weighted toward clean approach.
CRITICAL: If the Task tool is available, use it to spawn sub-agents. If it is not available, create the two proposal documents yourself in separate passes. DO NOT implement directly before comparing the proposals.
Session path: [SESSION_PATH from working memory] Task: $ARGUMENTS
IMPORTANT: Only work on this specific task. Don't do other parts of the plan.
Before spawning agents, assess phase complexity:
Create 2 independent proposals with the context below. If the Task tool is available, spawn 2 sub-agents in parallel (model: sonnet). If the Task tool is not available, do the minimal proposal first, then the clean proposal, without letting the second proposal optimize around the first.
CRITICAL: Both agents solve the entire task independently. This is NOT task splitting - each agent produces a FULL solution with their own philosophy.
Agent 1: Minimal Footprint
Session path: [SESSION_PATH] Task: $ARGUMENTS
Your approach: Solve with the smallest possible change surface.
Guidelines:
Deliverables:
Create [SESSION_PATH]/[TIMESTAMP_FILE]-solution-minimal.md:
# Solution: Minimal Footprint
Date: [TIMESTAMP_LOG]
Task: [description]
## Approach
[High-level strategy]
## Proposed Changes
### File: [path]
```[language]
[code]
Why: [rationale]
[What we accept with this approach]
[Potential issues]
Update _overview.md: Flow Log and Files entries
DO NOT edit actual code files - suggestions only
Report back with summary
Agent 2: Clean Foundation
Session path: [SESSION_PATH] Task: $ARGUMENTS
Your approach: Solve the "right way" assuming time to do it properly.
Guidelines:
Deliverables:
Create [SESSION_PATH]/[TIMESTAMP_FILE]-solution-clean.md:
# Solution: Clean Foundation
Date: [TIMESTAMP_LOG]
Task: [description]
## Approach
[High-level strategy]
## Proposed Changes
### File: [path]
```[language]
[code]
Why: [rationale]
[New patterns, refactorings]
[Long-term advantages]
Update _overview.md: Flow Log and Files entries
DO NOT edit actual code files - suggestions only
Report back with summary
After both agents complete:
Create [SESSION_PATH]/[TIMESTAMP_FILE]-comparison-[task-slug].md:
# Comparison: [task]
## Key Differences
[What's fundamentally different]
## Solution 1: Minimal
**Pros:** [list]
**Cons:** [list]
**Risk:** [Low/Medium/High]
## Solution 2: Clean
**Pros:** [list]
**Cons:** [list]
**Risk:** [Low/Medium/High]
## Recommendation
[Which approach and why]
## Hybrid Possibility
[Can we combine best of both?]
Auto-Selection Logic:
Complete common steps for session files
Present comparison and ask:
Which approach?
1. Minimal footprint
2. Clean foundation
3. Hybrid
4. Neither (rethink)
- [x], add completion note)IMPORTANT! If unsure about something, ask first.
