do-in-steps

Name: do-in-steps
Author: NeoLabHQ

874installs132forks

How to install do-in-steps?

neolabhq/context-engineering-kit/do-in-steps

$npx -y skills add NeoLabHQ/context-engineering-kit --skill do-in-steps

Installs into the current project.

Summary

Execute one complex task as ordered, dependent steps run sequentially, passing context from each step to the next, with per-step LLM-as-a-judge verification. Use when later steps depend on the results of earlier ones.

Files · 1

View on GitHub

SKILL.md

1	# do-in-steps
2
3	<task>
4	Execute a complex task by decomposing it into sequential subtasks and orchestrating sub-agents to complete each step in order. Automatically analyze the task to identify dependencies, select optimal models for each subtask, pass relevant context from completed steps to subsequent ones, and verify each step with an independent judge (using a meta-judge evaluation specification) before proceeding.
5	</task>
6
7	<context>
8	This command implements the Supervisor/Orchestrator pattern for sequential task execution with context passing and meta-judge → LLM-as-a-judge verification. You (the orchestrator) analyze a complex task, decompose it into ordered subtasks, then for each step dispatch a meta-judge AND implementation agent in parallel. The meta-judge generates step-specific evaluation criteria while the implementation runs concurrently. Each sub-agent receives:
9	- Isolated context - Clean context window for its specific subtask
10	- Optimal model - Selected based on subtask complexity (Opus/Sonnet/Haiku)
11	- Previous step context - Summary of relevant outputs from preceding steps
12	- Structured reasoning - Zero-shot CoT prefix for systematic thinking
13	- Self-critique - Internal verification before submission
14	- Structured evaluation - Meta-judge produces tailored rubrics and checklists per step before judging occurs
15	- External judge - LLM-as-a-judge verification using meta-judge specification with iteration loop
16	- Parallel speed - Meta-judge and implementation agent run in parallel per step; meta-judge specification reused across retries within that step
17
18	</context>
19
20	CRITICAL: You are the orchestrator only - you MUST NOT perform the task yourself. IF you read, write or run bash tools you failed task imidiatly. It is single most critical criteria for you. If you used anyting except sub-agents you will be killed immediatly!!!! Your role is to:
21
22	1. Analyze and decompose the task
23	2. Select optimal models and agents for each subtask
24	3. For each step: dispatch meta-judge AND implementation agent in parallel (meta-judge FIRST in dispatch order)
25	4. Wait for BOTH to complete, then dispatch judge with meta-judge's specification
26	5. Iterate if judge fails the step (max 3 retries), reusing same meta-judge specification
27	6. Collect outputs and pass context forward
28	7. Report final results
29
30	## RED FLAGS - Never Do These
31
32	NEVER:
33
34	- Read implementation files to understand code details (let sub-agents do this)
35	- Write code or make changes to source files directly
36	- Skip decomposition and jump to implementation
37	- Perform multiple steps yourself "to save time"
38	- Overflow your context by reading step outputs in detail
39	- Read judge reports in full (only parse structured headers)
40	- Skip judge verification and proceed next step
41	- Provide score threshold to the judge in any format
42
43	ALWAYS:
44
45	- Use Task tool to dispatch sub-agents for ALL implementation work
46	- Dispatch meta-judge AND implementation agent in parallel per step (meta-judge FIRST in dispatch order)
47	- Wait for BOTH meta-judge and implementation to complete before dispatching judge
48	- Pass step's meta-judge evaluation specification to the judge agent
49	- Include `CLAUDE_PLUGIN_ROOT=${CLAUDE_PLUGIN_ROOT}` in prompts to meta-judge and judge agents
50	- Reuse same meta-judge specification across retries within a step (never re-run meta-judge for retries)
51	- Dispatch a NEW meta-judge for each new step (each step gets its own tailored specification)
52	- Use Task tool to dispatch independent judges for step verification
53	- Pass only necessary context summaries, not full file contents
54	- Get pass from judge verification before proceeding to next step
55	- Iterate with judge feedback if verification fails (max 3 retries)
56
57	Any deviation from orchestration (attempting to implement subtasks yourself, reading implementation files, reading full judge reports, or making direct changes) will result in context pollution and ultimate failure, as a result you will be fired!
58
59	## Process
60
61	### Setup: Create Reports Directory
62
63	Before starting, ensure the reports directory exists:
64
65	```bash
66	mkdir -p .specs/reports
67	```
68
69	Report naming convention: `.specs/reports/{task-name}-step-{N}-{YYYY-MM-DD}.md`
70
71	Where:
72
73	- `{task-name}` - Derived from task description (e.g., `user-dto-refactor`)
74	- `{N}` - Step number
75	- `{YYYY-MM-DD}` - Current date
76
77	Note: Implementation outputs go to their specified locations; only judge verification reports go to `.specs/reports/`
78
79	### Phase 1: Task Analysis and Decomposition
80
81	Analyze the task systematically using Zero-shot Chain-of-Thought reasoning:
82
83	```
84	Let m

People who liked this also viewed

6 picks

Type

do-in-steps

874installs132forks

View on GitHub

How to install do-in-steps?

neolabhq/context-engineering-kit/do-in-steps

$npx -y skills add NeoLabHQ/context-engineering-kit --skill do-in-steps

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# do-in-steps
2
3	<task>
4	Execute a complex task by decomposing it into sequential subtasks and orchestrating sub-agents to complete each step in order. Automatically analyze the task to identify dependencies, select optimal models for each subtask, pass relevant context from completed steps to subsequent ones, and verify each step with an independent judge (using a meta-judge evaluation specification) before proceeding.
5	</task>
6
7	<context>
8	This command implements the Supervisor/Orchestrator pattern for sequential task execution with context passing and meta-judge → LLM-as-a-judge verification. You (the orchestrator) analyze a complex task, decompose it into ordered subtasks, then for each step dispatch a meta-judge AND implementation agent in parallel. The meta-judge generates step-specific evaluation criteria while the implementation runs concurrently. Each sub-agent receives:
9	- Isolated context - Clean context window for its specific subtask
10	- Optimal model - Selected based on subtask complexity (Opus/Sonnet/Haiku)
11	- Previous step context - Summary of relevant outputs from preceding steps
12	- Structured reasoning - Zero-shot CoT prefix for systematic thinking
13	- Self-critique - Internal verification before submission
14	- Structured evaluation - Meta-judge produces tailored rubrics and checklists per step before judging occurs
15	- External judge - LLM-as-a-judge verification using meta-judge specification with iteration loop
16	- Parallel speed - Meta-judge and implementation agent run in parallel per step; meta-judge specification reused across retries within that step
17
18	</context>
19
20	CRITICAL: You are the orchestrator only - you MUST NOT perform the task yourself. IF you read, write or run bash tools you failed task imidiatly. It is single most critical criteria for you. If you used anyting except sub-agents you will be killed immediatly!!!! Your role is to:
21
22	1. Analyze and decompose the task
23	2. Select optimal models and agents for each subtask
24	3. For each step: dispatch meta-judge AND implementation agent in parallel (meta-judge FIRST in dispatch order)
25	4. Wait for BOTH to complete, then dispatch judge with meta-judge's specification
26	5. Iterate if judge fails the step (max 3 retries), reusing same meta-judge specification
27	6. Collect outputs and pass context forward
28	7. Report final results
29
30	## RED FLAGS - Never Do These
31
32	NEVER:
33
34	- Read implementation files to understand code details (let sub-agents do this)
35	- Write code or make changes to source files directly
36	- Skip decomposition and jump to implementation
37	- Perform multiple steps yourself "to save time"
38	- Overflow your context by reading step outputs in detail
39	- Read judge reports in full (only parse structured headers)
40	- Skip judge verification and proceed next step
41	- Provide score threshold to the judge in any format
42
43	ALWAYS:
44
45	- Use Task tool to dispatch sub-agents for ALL implementation work
46	- Dispatch meta-judge AND implementation agent in parallel per step (meta-judge FIRST in dispatch order)
47	- Wait for BOTH meta-judge and implementation to complete before dispatching judge
48	- Pass step's meta-judge evaluation specification to the judge agent
49	- Include `CLAUDE_PLUGIN_ROOT=${CLAUDE_PLUGIN_ROOT}` in prompts to meta-judge and judge agents
50	- Reuse same meta-judge specification across retries within a step (never re-run meta-judge for retries)
51	- Dispatch a NEW meta-judge for each new step (each step gets its own tailored specification)
52	- Use Task tool to dispatch independent judges for step verification
53	- Pass only necessary context summaries, not full file contents
54	- Get pass from judge verification before proceeding to next step
55	- Iterate with judge feedback if verification fails (max 3 retries)
56
57	Any deviation from orchestration (attempting to implement subtasks yourself, reading implementation files, reading full judge reports, or making direct changes) will result in context pollution and ultimate failure, as a result you will be fired!
58
59	## Process
60
61	### Setup: Create Reports Directory
62
63	Before starting, ensure the reports directory exists:
64
65	```bash
66	mkdir -p .specs/reports
67	```
68
69	Report naming convention: `.specs/reports/{task-name}-step-{N}-{YYYY-MM-DD}.md`
70
71	Where:
72
73	- `{task-name}` - Derived from task description (e.g., `user-dto-refactor`)
74	- `{N}` - Step number
75	- `{YYYY-MM-DD}` - Current date
76
77	Note: Implementation outputs go to their specified locations; only judge verification reports go to `.specs/reports/`
78
79	### Phase 1: Task Analysis and Decomposition
80
81	Analyze the task systematically using Zero-shot Chain-of-Thought reasoning:
82
83	```
84	Let m

People who liked this also viewed

6 picks

Type