agentica-claude-proxy

Name: agentica-claude-proxy
Author: vibeeval

519stars44forks

How to install agentica-claude-proxy?

vibeeval/vibecosystem/agentica-claude-proxy

$npx -y skills add vibeeval/vibecosystem --skill agentica-claude-proxy

Installs into the current project.

Summary

--- name: agentica-claude-proxy description: Guide for integrating Agentica SDK with Claude Code CLI proxy allowed-tools: [Read, Bash] user-invocable: false ---

Files · 1

View on GitHub

SKILL.md

1	# Agentica-Claude Code Proxy Integration
2
3	Use this skill when developing or debugging the Agentica-Claude proxy integration.
4
5	## When to Use
6
7	- Setting up Agentica agents to use Claude Code tools
8	- Debugging agent hallucination issues
9	- Fixing permission errors in file operations
10	- Understanding the REPL response format
11
12	## Architecture Overview
13
14	```
15	Agentica Agent → S_M_BASE_URL → Claude Proxy → claude -p → Claude CLI (with tools)
16	(localhost:2345) (localhost:8080)
17	```
18
19	## Critical Requirements
20
21	### 1. --allowedTools Flag (REQUIRED)
22
23	Claude CLI in `-p` mode restricts file operations. You MUST add:
24
25	```python
26	subprocess.run([
27	"claude", "-p", prompt,
28	"--append-system-prompt", system_prompt,
29	"--allowedTools", "Read", "Write", "Edit", "Bash", # REQUIRED
30	])
31	```
32
33	Without this, agents will report "permission denied" for Write/Edit operations.
34
35	### 2. SSE Streaming Format (REQUIRED)
36
37	Agentica expects SSE streaming, not plain JSON:
38
39	```python
40	# Response format
41	yield f"data: {json.dumps(chunk)}\n\n"
42	yield "data: [DONE]\n\n"
43	```
44
45	### 3. REPL Response Format (REQUIRED)
46
47	Agents MUST return results as Python code blocks with a return statement:
48
49	```python
50	return "your result here"
51	```
52
53	Agentica's REPL parser extracts code between \`\`\`python and \`\`\`.
54
55	## Anti-Hallucination Prompt Engineering
56
57	Agents will hallucinate success without actually using tools unless you explicitly warn them:
58
59	```
60	## ANTI-HALLUCINATION WARNING
61
62	STOP AND READ THIS CAREFULLY:
63
64	You have access to these tools: Read, Write, Edit, Bash
65
66	When the task asks you to create/modify/run something:
67	1. FIRST: Actually invoke the tool (Read, Write, Edit, or Bash)
68	2. SECOND: Wait for the tool result
69	3. THIRD: Then return your answer based on what actually happened
70
71	DO NOT skip the tool invocation and just claim success!
72
73	If you didn't invoke a tool, you CANNOT claim the action succeeded.
74	```
75
76	## Path Sandboxing
77
78	Both Claude Code and Agentica have sandboxes:
79
80	- `/tmp/` paths are blocked by Claude Code
81	- Files outside project directory blocked by Agentica
82
83	Solution: Use project-relative paths like `workspace/` instead of `/tmp/`
84
85	## Debugging
86
87	### Check Agent Logs
88
89	```bash
90	cat logs/agent-<N>.log
91	```
92
93	Note: Logs only show final conversational response, not tool invocations.
94
95	### Test Proxy Directly
96
97	```bash
98	curl -s http://localhost:8080/v1/chat/completions \
99	-H "Content-Type: application/json" \
100	-d '{"model": "claude", "messages": [{"role": "user", "content": "Create file at workspace/test.txt"}], "stream": false}'
101	```
102
103	### Verify File Operations
104
105	```bash
106	# After agent claims to create file
107	ls -la workspace/test.txt
108	cat workspace/test.txt
109	```
110
111	## Server Commands
112
113	### Start Servers
114
115	```bash
116	# Terminal 1: Proxy
117	uv run python scripts/agentica/claude_proxy.py --port 8080
118
119	# Terminal 2: Agentica Server
120	cd workspace/agentica-research/agentica-server
121	INFERENCE_ENDPOINT_URL=http://localhost:8080/v1/chat/completions uv run agentica-server --port 2345
122	```
123
124	### Use Swarm
125
126	```bash
127	S_M_BASE_URL=http://localhost:2345 uv run python your_script.py
128	```
129
130	### Health Checks
131
132	```bash
133	curl http://localhost:8080/health # Proxy
134	curl http://localhost:2345/health # Agentica
135	```
136
137	## Reference Files
138
139	- Proxy implementation: `scripts/agentica/claude_proxy.py`
140	- REPL_BASELINE prompt: `scripts/agentica/claude_proxy.py:49-155`
141	- Comprehensive test: `workspace/test_swarm_all_tools.py`
142	- DependencySwarm: `scripts/agentica/dependency_swarm.py`
143
144	## Common Errors
145
146	\| Error \| Cause \| Fix \|
147	\|-------\|-------\|-----\|
148	\| "Permission denied" \| Missing --allowedTools \| Add `--allowedTools Read Write Edit Bash` \|
149	\| Agent claims success but file not created \| Hallucination \| Add anti-hallucination prompt section \|
150	\| "Cannot access /tmp/..." \| Sandbox restriction \| Use project-relative paths \|
151	\| "APIConnectionError" \| Wrong response format \| Use SSE streaming (data: {...}\n\n) \|
152	\| "NameError: view_file" \| Agent using REPL functions \| Add REPL_BASELINE with native tool examples \|

People who liked this also viewed

6 picks

Type

lark-openapi-explorer飞书/Lark 原生 OpenAPI 探索：从官方文档库中挖掘未经 CLI 封装的原生 OpenAPI 接口。当用户的需求无法被现有 lark-* skill 或 lark-cli 已注册命令满足，需要查找并调用原生飞书 OpenAPI 时使用。SkillsJul 2026378k16k
lark-skill-maker创建 lark-cli 的自定义 Skill。当用户需要把飞书 API 操作封装成可复用的 Skill（包装原子 API 或编排多步流程）时使用。SkillsJul 2026377k16k
azure-cost-optimizationOfficial agent plugin providing skills and MCP server configurations for Azure scenarios.SkillsJul 2026209k1.3k
vercel-react-native-skillsVercel's official collection of agent skillsSkillsJul 2026172k29k
mcp-builderGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate externaSkillsJul 202693k163k
gws-sharedgws CLI: Shared patterns for authentication, global flags, and output formatting.SkillsJul 202653k30k

agentica-claude-proxy

519stars44forks

View on GitHub

How to install agentica-claude-proxy?

vibeeval/vibecosystem/agentica-claude-proxy

$npx -y skills add vibeeval/vibecosystem --skill agentica-claude-proxy

Installs into the current project.

Summary

--- name: agentica-claude-proxy description: Guide for integrating Agentica SDK with Claude Code CLI proxy allowed-tools: [Read, Bash] user-invocable: false ---

Files · 1

View on GitHub

SKILL.md

1	# Agentica-Claude Code Proxy Integration
2
3	Use this skill when developing or debugging the Agentica-Claude proxy integration.
4
5	## When to Use
6
7	- Setting up Agentica agents to use Claude Code tools
8	- Debugging agent hallucination issues
9	- Fixing permission errors in file operations
10	- Understanding the REPL response format
11
12	## Architecture Overview
13
14	```
15	Agentica Agent → S_M_BASE_URL → Claude Proxy → claude -p → Claude CLI (with tools)
16	(localhost:2345) (localhost:8080)
17	```
18
19	## Critical Requirements
20
21	### 1. --allowedTools Flag (REQUIRED)
22
23	Claude CLI in `-p` mode restricts file operations. You MUST add:
24
25	```python
26	subprocess.run([
27	"claude", "-p", prompt,
28	"--append-system-prompt", system_prompt,
29	"--allowedTools", "Read", "Write", "Edit", "Bash", # REQUIRED
30	])
31	```
32
33	Without this, agents will report "permission denied" for Write/Edit operations.
34
35	### 2. SSE Streaming Format (REQUIRED)
36
37	Agentica expects SSE streaming, not plain JSON:
38
39	```python
40	# Response format
41	yield f"data: {json.dumps(chunk)}\n\n"
42	yield "data: [DONE]\n\n"
43	```
44
45	### 3. REPL Response Format (REQUIRED)
46
47	Agents MUST return results as Python code blocks with a return statement:
48
49	```python
50	return "your result here"
51	```
52
53	Agentica's REPL parser extracts code between \`\`\`python and \`\`\`.
54
55	## Anti-Hallucination Prompt Engineering
56
57	Agents will hallucinate success without actually using tools unless you explicitly warn them:
58
59	```
60	## ANTI-HALLUCINATION WARNING
61
62	STOP AND READ THIS CAREFULLY:
63
64	You have access to these tools: Read, Write, Edit, Bash
65
66	When the task asks you to create/modify/run something:
67	1. FIRST: Actually invoke the tool (Read, Write, Edit, or Bash)
68	2. SECOND: Wait for the tool result
69	3. THIRD: Then return your answer based on what actually happened
70
71	DO NOT skip the tool invocation and just claim success!
72
73	If you didn't invoke a tool, you CANNOT claim the action succeeded.
74	```
75
76	## Path Sandboxing
77
78	Both Claude Code and Agentica have sandboxes:
79
80	- `/tmp/` paths are blocked by Claude Code
81	- Files outside project directory blocked by Agentica
82
83	Solution: Use project-relative paths like `workspace/` instead of `/tmp/`
84
85	## Debugging
86
87	### Check Agent Logs
88
89	```bash
90	cat logs/agent-<N>.log
91	```
92
93	Note: Logs only show final conversational response, not tool invocations.
94
95	### Test Proxy Directly
96
97	```bash
98	curl -s http://localhost:8080/v1/chat/completions \
99	-H "Content-Type: application/json" \
100	-d '{"model": "claude", "messages": [{"role": "user", "content": "Create file at workspace/test.txt"}], "stream": false}'
101	```
102
103	### Verify File Operations
104
105	```bash
106	# After agent claims to create file
107	ls -la workspace/test.txt
108	cat workspace/test.txt
109	```
110
111	## Server Commands
112
113	### Start Servers
114
115	```bash
116	# Terminal 1: Proxy
117	uv run python scripts/agentica/claude_proxy.py --port 8080
118
119	# Terminal 2: Agentica Server
120	cd workspace/agentica-research/agentica-server
121	INFERENCE_ENDPOINT_URL=http://localhost:8080/v1/chat/completions uv run agentica-server --port 2345
122	```
123
124	### Use Swarm
125
126	```bash
127	S_M_BASE_URL=http://localhost:2345 uv run python your_script.py
128	```
129
130	### Health Checks
131
132	```bash
133	curl http://localhost:8080/health # Proxy
134	curl http://localhost:2345/health # Agentica
135	```
136
137	## Reference Files
138
139	- Proxy implementation: `scripts/agentica/claude_proxy.py`
140	- REPL_BASELINE prompt: `scripts/agentica/claude_proxy.py:49-155`
141	- Comprehensive test: `workspace/test_swarm_all_tools.py`
142	- DependencySwarm: `scripts/agentica/dependency_swarm.py`
143
144	## Common Errors
145
146	\| Error \| Cause \| Fix \|
147	\|-------\|-------\|-----\|
148	\| "Permission denied" \| Missing --allowedTools \| Add `--allowedTools Read Write Edit Bash` \|
149	\| Agent claims success but file not created \| Hallucination \| Add anti-hallucination prompt section \|
150	\| "Cannot access /tmp/..." \| Sandbox restriction \| Use project-relative paths \|
151	\| "APIConnectionError" \| Wrong response format \| Use SSE streaming (data: {...}\n\n) \|
152	\| "NameError: view_file" \| Agent using REPL functions \| Add REPL_BASELINE with native tool examples \|

People who liked this also viewed

6 picks

Type