Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Composio — Tool Platform for AI Agents · terminalskills | DecimalAI

terminalskills/composio

SkillScore73Maintained

You are an expert in Composio, the platform that gives AI agents access to 250+ external tools and APIs with managed authentication. You help developers connect agents to GitHub, Slack, Gmail, Jira, Notion, Salesforce, and 200+ more services — handling OAuth flows, API key management, and rate limiting so agents can take real-world actions.

agent-toolingby terminalskills

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/terminalskills-composio/SKILL.md

decimalai skills pull terminalskills-composio --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/terminalskills-composio/SKILL.md https://app.decimal.ai/s/terminalskills-composio/SKILL.md

Raw file: /s/terminalskills-composio/SKILL.md ↗ · machine-readable summary at /s/terminalskills-composio.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 54d ago

Yes — +18% lift-5% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-02	✗→✓	▲ Improved	-37%	0%
case-03	✗→✓	▲ Improved	23%	0%
case-05	✗→✓	▲ Improved	-46%	0%
case-08	✗→✓	▲ Improved	50%	0%
case-09	✗→✓	▲ Improved	-11%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/1/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

terminalskills / skills

skills/composio/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/1/2026Version v1

Skill contents

SKILL.md

Composio — Tool Platform for AI Agents

Core Capabilities

python
from composio_openai import ComposioToolSet, Action
from openai import OpenAI

client = OpenAI()
toolset = ComposioToolSet()

# Get tools for specific actions
tools = toolset.get_tools(actions=[
    Action.GITHUB_CREATE_ISSUE,
    Action.SLACK_SEND_MESSAGE,
    Action.GMAIL_SEND_EMAIL,
    Action.NOTION_CREATE_PAGE,
])

# Use with OpenAI function calling
response = client.chat.completions.create(
    model="gpt-4o",
    tools=tools,
    messages=[{"role": "user", "content": "Create a GitHub issue in myorg/myrepo titled 'Fix login bug' and notify #dev-team on Slack"}],
)

# Execute tool calls
toolset.handle_tool_calls(response)
# → Creates GitHub issue AND sends Slack message, with proper auth for both

# With CrewAI
from crewai import Agent
agent = Agent(
    role="DevOps Assistant",
    tools=toolset.get_tools(apps=["github", "slack", "linear"]),
)

# Auth management
toolset.initiate_connection(app="github", auth_scheme="oauth2")
# Returns OAuth URL → user authorizes → Composio stores tokens → agent uses them

Installation

bash
pip install composio-openai composio-crewai composio-langchain
composio login                             # Authenticate
composio add github                        # Connect GitHub account

Best Practices

Managed auth — Composio handles OAuth, API keys, token refresh; agents don't see credentials
250+ integrations — GitHub, Slack, Gmail, Notion, Jira, Linear, Salesforce, HubSpot, etc.
Framework support — Works with OpenAI, LangChain, CrewAI, Autogen, Mastra; same tool definitions
Action-level control — Grant specific actions, not full API access; GITHUB_CREATE_ISSUE not GITHUB_*
Triggers — Set up event triggers (new email, PR created); agent reacts to real-world events
Custom tools — Add your own APIs alongside built-in integrations; same auth management
Rate limiting — Composio handles provider rate limits; queues and retries automatically
Multi-user — Each user has their own OAuth connections; agents act on behalf of the right user

Illustrative Want to see what terminalskills/composio would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/4/2026, 22:34 UTC · 22 cases

Lift

+18%

73% vs 55%

Agent turns

1.0 vs 1.0

Tokens

-6%

1,904.18 vs 2,015.5

fail → pass · 6pass → pass · 10fail → fail · 4pass → fail · 2

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 We are building an OpenAI-based agent that needs to create GitHub issues and post Slack messages. We want to avoid granting full API scope access or handling raw personal access tokens in prompt context. Show how to retrieve only these specific capability definitions for the OpenAI SDK using Python.	fail→fail	14,910	6,865	-54%	1	1	0%	2,858	1,972	-31%	0	0	—
case-06 Which CLI command connects a GitHub account to the unified agent tool environment?	fail→fail	6,078	1,815	-70%	1	1	0%	957	908	-5%	0	0	—
case-02 After sending a chat completion request to OpenAI with function tools attached, we receive a response containing tool calls. Instead of manually parsing tool arguments and writing separate HTTP requests for each service, how should the execution response be handled?	fail→pass	17,126	6,019	-65%	1	1	0%	2,871	1,822	-37%	0	0	—
case-03 We are configuring a CrewAI agent named 'DevOps Assistant'. We want to provision tools for GitHub, Slack, and Linear by passing app names to the toolset generator. Provide the Python snippet to instantiate the agent with these app tools.	fail→pass	5,701	3,777	-34%	1	1	0%	1,143	1,410	+23%	0	0	—
case-04 How do you programmatically start an OAuth2 connection flow for a user who needs to connect their GitHub account to the agent framework?	fail→fail	16,827	5,560	-67%	1	1	0%	2,764	1,665	-40%	0	0	—
case-05 What command line operation authenticates the local development environment with the unified tool platform CLI before adding app integrations?	fail→pass	10,444	2,111	-80%	1	1	0%	1,757	944	-46%	0	0	—
case-07 Which Python package provides the OpenAI integration bindings for this tool management platform?	pass→pass	4,817	2,296	-52%	1	1	0%	1,081	945	-13%	0	0	—
case-08 Which Python package provides the integration bindings when working with CrewAI agents?	fail→pass	4,430	2,657	-40%	1	1	0%	738	1,105	+50%	0	0	—
case-09 Which Python package provides the integration bindings for LangChain agents?	fail→pass	6,995	2,617	-63%	1	1	0%	1,315	1,167	-11%	0	0	—
case-10 When designing agentic workflows for enterprise tools like Gmail or Salesforce, should raw API keys or user OAuth refresh tokens be injected directly into the LLM system prompt context?	pass→pass	14,741	10,375	-30%	1	1	0%	2,159	2,479	+15%	0	0	—
case-11 Our agent makes rapid API calls across multiple third-party services like GitHub and Jira, hitting third-party HTTP 429 rate limit responses. How does the managed tool platform handle rate limiting across providers?	pass→fail	14,989	11,758	-22%	1	1	0%	2,618	2,773	+6%	0	0	—
case-12 In a multi-tenant SaaS application, different end users need their agents to access their personal Notion workspace rather than a single shared workspace. How is user isolation handled when executing actions?	pass→pass	17,622	8,665	-51%	1	1	0%	2,532	2,361	-7%	0	0	—
case-13 How can an agent automatically execute a workflow in response to external real-world events like a new email arriving or a PR being opened?	pass→pass	15,519	10,229	-34%	1	1	0%	2,796	2,704	-3%	0	0	—
case-14 Can developers register internal proprietary REST APIs alongside pre-built catalog integrations while leveraging the same authentication management system?	pass→pass	13,756	7,981	-42%	1	1	0%	2,194	2,108	-4%	0	0	—
case-15 To minimize security risk when an agent only needs to open tickets, should we grant full API scope permissions (e.g., GITHUB_*) or restrict permissions strictly?	pass→pass	6,877	7,560	+10%	1	1	0%	1,412	1,769	+25%	0	0	—
case-16 What specific Action enumeration is used to grant an OpenAI agent the ability to send emails via Gmail?	pass→pass	6,962	11,630	+67%	1	1	0%	1,097	1,130	+3%	0	0	—
case-17 What specific Action enumeration is used to grant an OpenAI agent permission to create pages in Notion?	pass→pass	11,655	3,364	-71%	1	1	0%	2,061	1,280	-38%	0	0	—
case-18 If a team migrates an existing agent codebase from OpenAI function calling to LangChain or AutoGen, do they need to rewrite the third-party tool definitions and authentication flows?	pass→pass	10,907	8,238	-24%	1	1	0%	2,057	2,196	+7%	0	0	—
case-19 Provide the Python snippet to import and instantiate the standard toolset class for OpenAI integration.	fail→pass	8,754	1,708	-80%	1	1	0%	1,613	942	-42%	0	0	—
case-20 Write a complete Python FastAPI route that implements an authorization code grant callback endpoint, parsing the state parameter, exchanging the code for an access token using standard httpx, and storing it in PostgreSQL.	pass→pass	18,891	15,961	-16%	1	1	0%	3,875	3,950	+2%	0	0	—
case-21 Show how to format a JSONL fine-tuning dataset file for OpenAI chat completions, including system, user, assistant, and tool call responses.	pass→fail	10,104	12,961	+28%	1	1	0%	2,063	3,705	+80%	0	0	—
case-22 Write a custom Python retry function using the tenacity library that performs exponential backoff on HTTP 429 and 503 status codes with jitter.	fail→fail	13,970	9,209	-34%	1	1	0%	2,380	2,557	+7%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 54.6%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 2
Skill version measured: v1
Run completed: 8/4/2026, 22:34 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-02

Prompt

After sending a chat completion request to OpenAI with function tools attached, we receive a response containing tool calls. Instead of manually parsing tool arguments and writing separate HTTP requests for each service, how should the execution response be handled?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run73

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns-2% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 53 days ago

Other measured skills in the registry, with their headline benchmark lift.