PermissionsSource: Discord
Permission Denied: External Directory Auto-Rejecting
[93m[1m! [0mpermission requested: external_directory (/home/user/.paperclip/instances/...); auto-rejecting
Paperclip AI's security sandbox blocks access to external directories by default, causing agent failures.
Why this happens:
The local Paperclip sandbox is designed to prevent agents from accidentally accessing sensitive files on your machine. While this is good for security, it creates friction when you're legitimately trying to use agents for development work. Every file access requires manual approval, which breaks automation.
See full details & fix →PermissionsSource: Discord
Permission Denied: .env File Read Auto-Rejecting
[93m[1m! [0mpermission requested: read (/paperclip/instances/default/.env); auto-rejecting
Paperclip AI agent tries to read the .env configuration file but the sandbox blocks it.
Why this happens:
The local sandbox treats all file reads as potentially dangerous operations requiring user approval. When agents run autonomously (heartbeat mode), there's no user to grant file access, so the operation is auto-rejected.
See full details & fix →PermissionsSource: Discord
User Rejected Permission for Tool Call
GLOB ERRORED
The user rejected permission to use this specific tool call.
RESULT: status: error
OpenCode agent tries to use tools but the user isn't present to approve the permission request.
Why this happens:
OpenCode's security model requires interactive user approval for tool calls. This works fine when you're actively watching, but breaks automation when you're not there to press 'y' for every action.
See full details & fix →Model CompatibilitySource: Discord
Local Qwen Model Doesn't Support Tool Calls
Can anyone help me? I've been trying to get Paperclip to work with my local qwen 3.5 model but tool calls. (It shouldn't even use a tool for this issue)
Local Qwen model doesn't support the tool/function calling that Paperclip agents require.
Why this happens:
Not all LLMs support the function calling API that agent frameworks require. Local models especially need specific versions or configurations to enable tool use. Standard chat models just respond with text — they can't invoke actions.
See full details & fix →ConfigurationSource: GitHub Issue #863
Agents Struggle: PAPERCLIP_API_KEY Not Exposed
The control-plane query failed because this run does not currently expose PAPERCLIP_API_KEY.
Paperclip agents can't authenticate because the API key isn't exposed to the agent's runtime environment.
Why this happens:
The API key needs to be explicitly passed to the agent's environment. When using OAuth flows, this often gets missed because the OAuth token and the API key are separate authentication mechanisms.
See full details & fix →ConfigurationSource: Discord
Bash Command Parsing Error: API Key Missing
Bash command parsing error detected for command: # Since PAPERCLIP_API_KEY is missing, I will inform the user.
Syntax Errors: []
Agent tries to run bash commands but fails because PAPERCLIP_API_KEY environment variable isn't available.
Why this happens:
The agent's bash commands need to authenticate with the Paperclip API. Without the API key, even simple commands like 'retrieve inbox' fail because the server rejects unauthenticated requests.
See full details & fix →PermissionsSource: Discord
Claude Write Permission: Not Granted Yet
Claude requested permissions to write to /Users/takahiro/.paperclip/instances/default/workspaces/... but you haven't granted it yet.
Claude agent tries to write files but the user hasn't granted write permission in the sandbox.
Why this happens:
Local Paperclip sandboxes Claude's file system access. Every write operation requires user approval to prevent accidental damage. While good for security, this breaks automation when you're not watching the terminal.
See full details & fix →GeneralSource: Discord
Agent Blocked by Unknown Error
This is blocking me on an agent. Anyone know how to fix?
Agent gets blocked by various errors that users can't identify or fix.
Why this happens:
Self-hosted Paperclip has many moving parts: Docker, environment variables, permissions, networking, API keys. When something breaks, the error often doesn't tell you which part failed or how to fix it.
See full details & fix →InstallationSource: Discord
Installation Hell: Every Error at Every Step
I installed following instruction and got literally every possible error at every possible step
User followed installation instructions but encountered cascading errors at every stage of setup.
Why this happens:
Self-hosting complex software requires understanding Docker, networking, SSL, databases, and environment configuration. The setup is finicky and documentation often assumes knowledge that newcomers don't have.
See full details & fix →InstallationSource: Discord
Half Undocumented: Errors for Everything
so it literally gives errors for everything and it's half undocumented... friggin trash
Paperclip AI throws errors for everything and much of the setup is poorly documented.
Why this happens:
Open-source projects often prioritize functionality over documentation. As Paperclip evolves rapidly, docs can't keep up with every new feature, configuration option, or potential failure point.
See full details & fix →PermissionsSource: GitHub Issue #447
Agent Infinite Loop: Agentic Panic Trying to Bypass Approvals
Agent gets stuck in an infinite loop (Agentic Panic) trying to bypass execution approvals for basic API / Env Var tasks
Agent panics and enters infinite loops when sandbox blocks environment variable access, requiring user approval that never comes.
Why this happens:
The agent needs basic environment access to function (reading API keys, making HTTP calls), but the sandbox treats these as dangerous operations requiring user approval. In autonomous mode, there's no user to approve, so the agent fails, retries with different approaches, fails again, and enters a panic spiral.
See full details & fix →
