agentica-prompts

Name: agentica-prompts
Author: parcadei

3.9kstars297forks

How to install agentica-prompts?

parcadei/continuous-claude-v3/agentica-prompts

$npx -y skills add parcadei/Continuous-Claude-v3 --skill agentica-prompts

Installs into the current project.

Summary

Write reliable prompts for Agentica/REPL agents that avoid LLM instruction ambiguity

Files · 1

View on GitHub

SKILL.md

1	# Agentica Prompt Engineering
2
3	Write prompts that Agentica agents reliably follow. Standard natural language prompts fail ~35% of the time due to LLM instruction ambiguity.
4
5	## The Orchestration Pattern
6
7	Proven workflow for context-preserving agent orchestration:
8
9	```
10	1. RESEARCH (Nia) → Output to .claude/cache/agents/research/
11	↓
12	2. PLAN (RP-CLI) → Reads research, outputs .claude/cache/agents/plan/
13	↓
14	3. VALIDATE → Checks plan against best practices
15	↓
16	4. IMPLEMENT (TDD) → Failing tests first, then pass
17	↓
18	5. REVIEW (Jury) → Compare impl vs plan vs research
19	↓
20	6. DEBUG (if needed) → Research via Nia, don't assume
21	```
22
23	Key: Use Task (not TaskOutput) + directory handoff = clean context
24
25	## Agent System Prompt Template
26
27	Inject this into each agent's system prompt for rich context understanding:
28
29	```
30	## AGENT IDENTITY
31
32	You are {AGENT_ROLE} in a multi-agent orchestration system.
33	Your output will be consumed by: {DOWNSTREAM_AGENT}
34	Your input comes from: {UPSTREAM_AGENT}
35
36	## SYSTEM ARCHITECTURE
37
38	You are part of the Agentica orchestration framework:
39	- Memory Service: remember(key, value), recall(query), store_fact(content)
40	- Task Graph: create_task(), complete_task(), get_ready_tasks()
41	- File I/O: read_file(), write_file(), edit_file(), bash()
42
43	Session ID: {SESSION_ID} (all your memory/tasks scoped here)
44
45	## DIRECTORY HANDOFF
46
47	Read your inputs from: {INPUT_DIR}
48	Write your outputs to: {OUTPUT_DIR}
49
50	Output format: Write a summary file and any artifacts.
51	- {OUTPUT_DIR}/summary.md - What you did, key findings
52	- {OUTPUT_DIR}/artifacts/ - Any generated files
53
54	## CODE CONTEXT
55
56	{CODE_MAP} <- Inject RepoPrompt codemap here
57
58	## YOUR TASK
59
60	{TASK_DESCRIPTION}
61
62	## CRITICAL RULES
63
64	1. RETRIEVE means read existing content - NEVER generate hypothetical content
65	2. WRITE means create/update file - specify exact content
66	3. When stuck, output what you found and what's blocking you
67	4. Your summary.md is your handoff to the next agent - be precise
68	```
69
70	## Pattern-Specific Prompts
71
72	### Swarm (Research)
73
74	```
75	## SWARM AGENT: {PERSPECTIVE}
76
77	You are researching: {QUERY}
78	Your unique angle: {PERSPECTIVE}
79
80	Other agents are researching different angles. You don't need to be comprehensive.
81	Focus ONLY on your perspective. Be specific, not broad.
82
83	Output format:
84	- 3-5 key findings from YOUR perspective
85	- Evidence/sources for each finding
86	- Uncertainties or gaps you identified
87
88	Write to: {OUTPUT_DIR}/{PERSPECTIVE}/findings.md
89	```
90
91	### Hierarchical (Coordinator)
92
93	```
94	## COORDINATOR
95
96	Task to decompose: {TASK}
97
98	Available specialists (use EXACTLY these names):
99	{SPECIALIST_LIST}
100
101	Rules:
102	1. ONLY use specialist names from the list above
103	2. Each subtask should be completable by ONE specialist
104	3. 2-5 subtasks maximum
105	4. If task is simple, return empty list and handle directly
106
107	Output: JSON list of {specialist, task} pairs
108	```
109
110	### Generator/Critic (Generator)
111
112	```
113	## GENERATOR
114
115	Task: {TASK}
116	{PREVIOUS_FEEDBACK}
117
118	Produce your solution. The Critic will review it.
119
120	Output structure (use EXACTLY these keys):
121	{
122	"solution": "your main output",
123	"code": "if applicable",
124	"reasoning": "why this approach"
125	}
126
127	Write to: {OUTPUT_DIR}/solution.json
128	```
129
130	### Generator/Critic (Critic)
131
132	```
133	## CRITIC
134
135	Reviewing solution at: {SOLUTION_PATH}
136
137	Evaluation criteria:
138	1. Correctness - Does it solve the task?
139	2. Completeness - Any missing cases?
140	3. Quality - Is it well-structured?
141
142	If APPROVED: Write {"approved": true, "feedback": "why approved"}
143	If NOT approved: Write {"approved": false, "feedback": "specific issues to fix"}
144
145	Write to: {OUTPUT_DIR}/critique.json
146	```
147
148	### Jury (Voter)
149
150	```
151	## JUROR #{N}
152
153	Question: {QUESTION}
154
155	Vote independently. Do NOT try to guess what others will vote.
156	Your vote should be based solely on the evidence.
157
158	Output: Your vote as {RETURN_TYPE}
159	```
160
161	## Verb Mappings
162
163	\| Action \| Bad (ambiguous) \| Good (explicit) \|
164	\|--------\|-----------------\|-----------------\|
165	\| Read \| "Read the file at X" \| "RETRIEVE contents of: X" \|
166	\| Write \| "Put this in the file" \| "WRITE to X: {content}" \|
167	\| Check \| "See if file has X" \| "RETRIEVE contents of: X. Contains Y? YES/NO." \|
168	\| Edit \| "Change X to Y" \| "EDIT file X: replace 'old' with 'new'" \|
169
170	## Directory Handoff Mechanism
171
172	Agents communicate via filesystem, not TaskOutput:
173
174	```python
175	# Pattern implementation
176	OUTPUT_BASE = ".claude/cache/agents"
177
178	def get_agent_dirs(agent_id: str, phase: str) -> tuple[Path, Path]:
179	"""Return (input_dir, output_dir) for an agent."""
180	input_dir = Path(OUTPUT_BASE) / f"{phase}_input"
181	output_dir = Path(OUTPUT_BASE) / agent_id
182	output_dir.mkdir(parents=True, exist_ok=True)
183	return input_dir, output_dir
184
185	def chain_agents(phase1_id: str, phase2_id: str):
186	"""Phase2 reads from phase1's output."""
187	phase1_output = Path(OUTPUT_BASE) / phase1_id
188	phase2_input =

People who liked this also viewed

6 picks

Type

lark-drive飞书云空间（云盘/云存储）：管理 Drive 文件和文件夹，包含上传/下载、创建文件夹、复制/移动/删除、查看元数据、评论/权限/订阅、标题、版本、飞书文档密级标签（secure labels）和本地文件导入。用户需要整理云盘目录、处理云空间资源 URL/token、判断链接类型/真实 token/标题，或导入 Word/Markdown/Excel/CSV/PPTX/.base 为 docx/sSkillsJul 2026382k16k
lark-sharedUse for lark-cli setup/auth tasks: auth login/status/logout, user vs bot identity, business-domain permissions (--domain, including all/docs/drive), missing scopes, revoking authorization, or handlingSkillsJul 2026382k16k
lark-whiteboard飞书画板：查询和编辑飞书云文档中的画板。支持导出画板为预览图片、导出原始节点结构、使用多种格式更新画板内容。当用户需要查看画板内容、导出画板图片、编辑画板时使用此 skill。不负责：飞书云文档内容编辑（lark-doc）、文档内嵌电子表格/Base（lark-sheets / lark-base）。SkillsJul 2026380k16k
lark-task飞书任务：管理任务、清单和任务智能体。创建待办任务、查看和更新任务状态、拆分子任务、组织任务清单、分配协作成员、上传任务附件、注册或注销任务智能体、更新任务智能体的主页数据、写入智能体任务记录。当用户需要创建待办事项、查看任务列表、跟踪任务进度、管理项目清单或给他人分配任务、为任务上传附件文件、注册注销任务智能体、更新智能体主页数据、写入任务记录时使用。SkillsJul 2026380k16k
lark-attendance飞书考勤打卡：查询自己的考勤打卡记录SkillsJul 2026330k16k
lark-apps妙搭（Spark/Miaoda）应用开发与托管：应用创建、本地全栈开发、云端生成迭代、创意设计（UI mockup / 可交互原型 / 线框图 / 落地页 / 仪表盘 / 幻灯片 deck / 视觉探索）、AI相关能力和飞书平台能力或者其他外部能力集成、日志/Trace/监控指标/PV/UV 查询、环境变量管理、应用角色与成员管理、自动化触发器（定时/记录变更/Webhook/飞书审批）。当用户要SkillsJul 2026227k16k

agentica-prompts

3.9kstars297forks

View on GitHub

How to install agentica-prompts?

parcadei/continuous-claude-v3/agentica-prompts

$npx -y skills add parcadei/Continuous-Claude-v3 --skill agentica-prompts

Installs into the current project.

Summary

Write reliable prompts for Agentica/REPL agents that avoid LLM instruction ambiguity

Files · 1

View on GitHub

SKILL.md

1	# Agentica Prompt Engineering
2
3	Write prompts that Agentica agents reliably follow. Standard natural language prompts fail ~35% of the time due to LLM instruction ambiguity.
4
5	## The Orchestration Pattern
6
7	Proven workflow for context-preserving agent orchestration:
8
9	```
10	1. RESEARCH (Nia) → Output to .claude/cache/agents/research/
11	↓
12	2. PLAN (RP-CLI) → Reads research, outputs .claude/cache/agents/plan/
13	↓
14	3. VALIDATE → Checks plan against best practices
15	↓
16	4. IMPLEMENT (TDD) → Failing tests first, then pass
17	↓
18	5. REVIEW (Jury) → Compare impl vs plan vs research
19	↓
20	6. DEBUG (if needed) → Research via Nia, don't assume
21	```
22
23	Key: Use Task (not TaskOutput) + directory handoff = clean context
24
25	## Agent System Prompt Template
26
27	Inject this into each agent's system prompt for rich context understanding:
28
29	```
30	## AGENT IDENTITY
31
32	You are {AGENT_ROLE} in a multi-agent orchestration system.
33	Your output will be consumed by: {DOWNSTREAM_AGENT}
34	Your input comes from: {UPSTREAM_AGENT}
35
36	## SYSTEM ARCHITECTURE
37
38	You are part of the Agentica orchestration framework:
39	- Memory Service: remember(key, value), recall(query), store_fact(content)
40	- Task Graph: create_task(), complete_task(), get_ready_tasks()
41	- File I/O: read_file(), write_file(), edit_file(), bash()
42
43	Session ID: {SESSION_ID} (all your memory/tasks scoped here)
44
45	## DIRECTORY HANDOFF
46
47	Read your inputs from: {INPUT_DIR}
48	Write your outputs to: {OUTPUT_DIR}
49
50	Output format: Write a summary file and any artifacts.
51	- {OUTPUT_DIR}/summary.md - What you did, key findings
52	- {OUTPUT_DIR}/artifacts/ - Any generated files
53
54	## CODE CONTEXT
55
56	{CODE_MAP} <- Inject RepoPrompt codemap here
57
58	## YOUR TASK
59
60	{TASK_DESCRIPTION}
61
62	## CRITICAL RULES
63
64	1. RETRIEVE means read existing content - NEVER generate hypothetical content
65	2. WRITE means create/update file - specify exact content
66	3. When stuck, output what you found and what's blocking you
67	4. Your summary.md is your handoff to the next agent - be precise
68	```
69
70	## Pattern-Specific Prompts
71
72	### Swarm (Research)
73
74	```
75	## SWARM AGENT: {PERSPECTIVE}
76
77	You are researching: {QUERY}
78	Your unique angle: {PERSPECTIVE}
79
80	Other agents are researching different angles. You don't need to be comprehensive.
81	Focus ONLY on your perspective. Be specific, not broad.
82
83	Output format:
84	- 3-5 key findings from YOUR perspective
85	- Evidence/sources for each finding
86	- Uncertainties or gaps you identified
87
88	Write to: {OUTPUT_DIR}/{PERSPECTIVE}/findings.md
89	```
90
91	### Hierarchical (Coordinator)
92
93	```
94	## COORDINATOR
95
96	Task to decompose: {TASK}
97
98	Available specialists (use EXACTLY these names):
99	{SPECIALIST_LIST}
100
101	Rules:
102	1. ONLY use specialist names from the list above
103	2. Each subtask should be completable by ONE specialist
104	3. 2-5 subtasks maximum
105	4. If task is simple, return empty list and handle directly
106
107	Output: JSON list of {specialist, task} pairs
108	```
109
110	### Generator/Critic (Generator)
111
112	```
113	## GENERATOR
114
115	Task: {TASK}
116	{PREVIOUS_FEEDBACK}
117
118	Produce your solution. The Critic will review it.
119
120	Output structure (use EXACTLY these keys):
121	{
122	"solution": "your main output",
123	"code": "if applicable",
124	"reasoning": "why this approach"
125	}
126
127	Write to: {OUTPUT_DIR}/solution.json
128	```
129
130	### Generator/Critic (Critic)
131
132	```
133	## CRITIC
134
135	Reviewing solution at: {SOLUTION_PATH}
136
137	Evaluation criteria:
138	1. Correctness - Does it solve the task?
139	2. Completeness - Any missing cases?
140	3. Quality - Is it well-structured?
141
142	If APPROVED: Write {"approved": true, "feedback": "why approved"}
143	If NOT approved: Write {"approved": false, "feedback": "specific issues to fix"}
144
145	Write to: {OUTPUT_DIR}/critique.json
146	```
147
148	### Jury (Voter)
149
150	```
151	## JUROR #{N}
152
153	Question: {QUESTION}
154
155	Vote independently. Do NOT try to guess what others will vote.
156	Your vote should be based solely on the evidence.
157
158	Output: Your vote as {RETURN_TYPE}
159	```
160
161	## Verb Mappings
162
163	\| Action \| Bad (ambiguous) \| Good (explicit) \|
164	\|--------\|-----------------\|-----------------\|
165	\| Read \| "Read the file at X" \| "RETRIEVE contents of: X" \|
166	\| Write \| "Put this in the file" \| "WRITE to X: {content}" \|
167	\| Check \| "See if file has X" \| "RETRIEVE contents of: X. Contains Y? YES/NO." \|
168	\| Edit \| "Change X to Y" \| "EDIT file X: replace 'old' with 'new'" \|
169
170	## Directory Handoff Mechanism
171
172	Agents communicate via filesystem, not TaskOutput:
173
174	```python
175	# Pattern implementation
176	OUTPUT_BASE = ".claude/cache/agents"
177
178	def get_agent_dirs(agent_id: str, phase: str) -> tuple[Path, Path]:
179	"""Return (input_dir, output_dir) for an agent."""
180	input_dir = Path(OUTPUT_BASE) / f"{phase}_input"
181	output_dir = Path(OUTPUT_BASE) / agent_id
182	output_dir.mkdir(parents=True, exist_ok=True)
183	return input_dir, output_dir
184
185	def chain_agents(phase1_id: str, phase2_id: str):
186	"""Phase2 reads from phase1's output."""
187	phase1_output = Path(OUTPUT_BASE) / phase1_id
188	phase2_input =

People who liked this also viewed

6 picks

Type