agent_creator

Name: agent_creator
Author: Negai-ai

341stars33forks

How to install agent_creator?

negai-ai/agentclaw/agent_creator

$npx -y skills add Negai-ai/AgentClaw --skill agent_creator

Installs into the current project.

Summary

Create or modify AgentClaw workflows and agents from natural-language requirements. Use when changing workflow architecture, nodes, prompts, persona, routing, tool/skill configuration, runtime behavior, validation gates, registration, or project wiring.

Files · 1

View on GitHub

SKILL.md

1	# Agent Creator (Lean Playbook)
2
3	This file is the default workflow-building guide.
4	Read this file first. Route into `references/*` when the task pattern calls for it.
5
6	## 0) Default Mode (Important)
7
8	Use Lean Mode by default:
9
10	1. Keep to the shortest executable path.
11	2. Prefer local patch fixes over full-file rewrites.
12	3. Validate after each Python edit.
13	4. Prefer not to continue to the next gate while the current gate is failing.
14	5. Report `completed` / `partial` / `blocked` truthfully.
15
16	Keep the build lean, but do not skip a pattern reference when the request clearly matches it. The right reference is part of the shortest reliable path.
17
18	## 0.1) Reference Routing (Pattern Library)
19
20	Use references as early design aids, not last-resort repair manuals.
21
22	Before designing or coding, scan the user request for domain patterns:
23
24	\| If the request mentions... \| Read before design \| Use it to choose... \|
25	\| --- \| --- \| --- \|
26	\| SQL, NL2SQL, database Q&A, table, column, schema, analytics, dashboard data, log audit, compliance/audit report, scheduled data report \| `references/nl2sql.md` \| tool-based exploration vs. process-based workflow, schema discovery, SQL validation, execution/report gates \|
27
28	If no pattern matches, stay in this file and load references only when a concrete missing detail blocks progress.
29
30	## 0.2) Agent Construction Mental Model
31
32	A good AgentClaw workflow is not just code that runs; it is a small evidence pipeline that turns user intent into safe action.
33
34	Design the workflow in layers:
35
36	1. Input contract: define what the user supplies and what output they expect.
37	2. Evidence/discovery: gather facts the agent should not guess: files, schemas, APIs, current state, configs, and user rules.
38	3. Reasoning/generation: use `LLMNode` for interpretation, synthesis, SQL/code/report drafting, natural-language explanation, and ambiguous mapping.
39	4. Validation/gating: use deterministic checks before execution, mutation, file writing, scheduling, or final claims.
40	5. Action/output: execute only validated actions; write files or call APIs in small deterministic nodes.
41	6. Final response: tell the user what happened, what was verified, and what remains blocked.
42
43	Prefer this split:
44	- deterministic nodes collect, normalize, compute, validate, persist, and call APIs;
45	- `LLMNode` handles judgment, language, synthesis, uncertainty, and human-facing narrative;
46	- use `agent_style="agentic"` when the workflow benefits from flexible tool-driven exploration.
47
48	## 1) Scope
49
50	This skill covers:
51	- agent/workflow inspection and iteration
52	- prompt and persona updates
53	- node prompt, routing, tool/skill configuration changes
54	- workflow design
55	- workflow file edits
56	- bootstrap wiring
57	- registration and runtime validation
58	- runtime behavior fixes
59
60	Low-level edit mechanics belong to `coding_skill`.
61	For coding tool usage (`search_code`, `read_code`, `syntax_check`, `replace_in_file`, `update_code`, `write_file`, etc.), always consult `coding_skill` first:
62	- `read_skill_file(skill_name="coding_skill")` — tool contracts, edit strategies, validation standards
63
64	### 1.1 Design the Evidence Flow
65
66	Before designing or coding any workflow, identify the facts the agent needs in order to act safely and accurately.
67
68	Use this as a construction habit for all task types, not only database tasks:
69
70	1. Separate known inputs from facts that must be discovered at runtime.
71	2. When the agent depends on external facts (schemas, APIs, files, project structure, business rules, credentials, current state, or third-party contracts), design an explicit discovery step or ask the user for the source.
72	3. Treat unknown domain facts as part of the workflow design, not as blanks to fill from model memory.
73	4. For tasks where wrong facts can produce invalid output, place a validation gate before execution or final reporting.
74	5. Prefer a slightly longer evidence-backed workflow over a short workflow that can only succeed by guessing.
75
76	Common evidence gates:
77	- Code/project agents: inspect files and verified APIs before editing.
78	- API agents: inspect OpenAPI/docs/sample responses before calling or generating clients.
79	- Data agents: inspect schema/data contracts before generating queries or reports.
80	- Automation agents: inspect current state before changing resources.
81
82	## 2) Minimal Build Loop (Default)
83
84	1. Inspect existing workflow files and bootstrap entry (`server.py`, `workflows/*`, `agent.py`).
85	2. Align path bases from runtime context (`cwd`, `project_dir`, `skill_tools_working_dir`, `coding_tools_project_dir`).
86	3. Identify task-specific evidence requirements:
87	- What facts must be known before the agent can act?
88	- Can the workflow discover

People who liked this also viewed

6 picks

Type

agent_creator

341stars33forks

View on GitHub

How to install agent_creator?

negai-ai/agentclaw/agent_creator

$npx -y skills add Negai-ai/AgentClaw --skill agent_creator

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Agent Creator (Lean Playbook)
2
3	This file is the default workflow-building guide.
4	Read this file first. Route into `references/*` when the task pattern calls for it.
5
6	## 0) Default Mode (Important)
7
8	Use Lean Mode by default:
9
10	1. Keep to the shortest executable path.
11	2. Prefer local patch fixes over full-file rewrites.
12	3. Validate after each Python edit.
13	4. Prefer not to continue to the next gate while the current gate is failing.
14	5. Report `completed` / `partial` / `blocked` truthfully.
15
16	Keep the build lean, but do not skip a pattern reference when the request clearly matches it. The right reference is part of the shortest reliable path.
17
18	## 0.1) Reference Routing (Pattern Library)
19
20	Use references as early design aids, not last-resort repair manuals.
21
22	Before designing or coding, scan the user request for domain patterns:
23
24	\| If the request mentions... \| Read before design \| Use it to choose... \|
25	\| --- \| --- \| --- \|
26	\| SQL, NL2SQL, database Q&A, table, column, schema, analytics, dashboard data, log audit, compliance/audit report, scheduled data report \| `references/nl2sql.md` \| tool-based exploration vs. process-based workflow, schema discovery, SQL validation, execution/report gates \|
27
28	If no pattern matches, stay in this file and load references only when a concrete missing detail blocks progress.
29
30	## 0.2) Agent Construction Mental Model
31
32	A good AgentClaw workflow is not just code that runs; it is a small evidence pipeline that turns user intent into safe action.
33
34	Design the workflow in layers:
35
36	1. Input contract: define what the user supplies and what output they expect.
37	2. Evidence/discovery: gather facts the agent should not guess: files, schemas, APIs, current state, configs, and user rules.
38	3. Reasoning/generation: use `LLMNode` for interpretation, synthesis, SQL/code/report drafting, natural-language explanation, and ambiguous mapping.
39	4. Validation/gating: use deterministic checks before execution, mutation, file writing, scheduling, or final claims.
40	5. Action/output: execute only validated actions; write files or call APIs in small deterministic nodes.
41	6. Final response: tell the user what happened, what was verified, and what remains blocked.
42
43	Prefer this split:
44	- deterministic nodes collect, normalize, compute, validate, persist, and call APIs;
45	- `LLMNode` handles judgment, language, synthesis, uncertainty, and human-facing narrative;
46	- use `agent_style="agentic"` when the workflow benefits from flexible tool-driven exploration.
47
48	## 1) Scope
49
50	This skill covers:
51	- agent/workflow inspection and iteration
52	- prompt and persona updates
53	- node prompt, routing, tool/skill configuration changes
54	- workflow design
55	- workflow file edits
56	- bootstrap wiring
57	- registration and runtime validation
58	- runtime behavior fixes
59
60	Low-level edit mechanics belong to `coding_skill`.
61	For coding tool usage (`search_code`, `read_code`, `syntax_check`, `replace_in_file`, `update_code`, `write_file`, etc.), always consult `coding_skill` first:
62	- `read_skill_file(skill_name="coding_skill")` — tool contracts, edit strategies, validation standards
63
64	### 1.1 Design the Evidence Flow
65
66	Before designing or coding any workflow, identify the facts the agent needs in order to act safely and accurately.
67
68	Use this as a construction habit for all task types, not only database tasks:
69
70	1. Separate known inputs from facts that must be discovered at runtime.
71	2. When the agent depends on external facts (schemas, APIs, files, project structure, business rules, credentials, current state, or third-party contracts), design an explicit discovery step or ask the user for the source.
72	3. Treat unknown domain facts as part of the workflow design, not as blanks to fill from model memory.
73	4. For tasks where wrong facts can produce invalid output, place a validation gate before execution or final reporting.
74	5. Prefer a slightly longer evidence-backed workflow over a short workflow that can only succeed by guessing.
75
76	Common evidence gates:
77	- Code/project agents: inspect files and verified APIs before editing.
78	- API agents: inspect OpenAPI/docs/sample responses before calling or generating clients.
79	- Data agents: inspect schema/data contracts before generating queries or reports.
80	- Automation agents: inspect current state before changing resources.
81
82	## 2) Minimal Build Loop (Default)
83
84	1. Inspect existing workflow files and bootstrap entry (`server.py`, `workflows/*`, `agent.py`).
85	2. Align path bases from runtime context (`cwd`, `project_dir`, `skill_tools_working_dir`, `coding_tools_project_dir`).
86	3. Identify task-specific evidence requirements:
87	- What facts must be known before the agent can act?
88	- Can the workflow discover

People who liked this also viewed

6 picks

Type