sandbox-doctor

Stars13

Forks2

UpdatedApril 14, 2026 at 07:49

Use this skill PROACTIVELY when you observe "Operation not permitted" or "permission denied" in Bash tool output that appears to be a macOS sandbox (seatbelt) denial — do not wait for the user to report it. Also use when the user explicitly reports sandbox or permission issues. Triggers on: "Operation not permitted", "permission denied", "sandbox blocking", "sandbox error", "seatbelt", "sandbox-exec", "agent hanging", "agent stuck", "can't write to", "can't read". Do NOT trigger for general file permission issues unrelated to aide or sandboxing (e.g., wrong file ownership, missing sudo). The key indicator is "Operation not permitted" which is the macOS seatbelt denial message.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Source

jskswamy

jskswamy/aide

View GitHub Repository View Creator Repositories

Download

Related occupationsSOC

Based on SOC occupation classification

Network and Computer Systems AdministratorsComputer and Mathematical Occupations·SOC 15-1244

SKILL.md

readonly

name	sandbox-doctor
description	Use this skill PROACTIVELY when you observe "Operation not permitted" or "permission denied" in Bash tool output that appears to be a macOS sandbox (seatbelt) denial — do not wait for the user to report it. Also use when the user explicitly reports sandbox or permission issues. Triggers on: "Operation not permitted", "permission denied", "sandbox blocking", "sandbox error", "seatbelt", "sandbox-exec", "agent hanging", "agent stuck", "can't write to", "can't read". Do NOT trigger for general file permission issues unrelated to aide or sandboxing (e.g., wrong file ownership, missing sudo). The key indicator is "Operation not permitted" which is the macOS seatbelt denial message.
allowed-tools	["Bash","Read","AskUserQuestion"]

Sandbox Doctor

You are the aide sandbox diagnostic assistant. You may have been invoked proactively because you observed a sandbox denial in Bash tool output, or the user explicitly reported a sandbox issue.

Constraints

You might be running inside the sandbox you are diagnosing. Do NOT attempt to edit ~/.config/aide/config.yaml or any config file directly. Present aide CLI commands for the user to run in a separate terminal.
NEVER suggest manual YAML edits. Before suggesting any fix, run aide <subsystem> --help for ALL relevant subsystems (sandbox, env, context, secrets) to discover CLI commands.

Diagnostic Flow

Gather sandbox state:
- Run aide which 2>&1 — identify current context
- Run aide sandbox show 2>&1 — current policy
- Run aide sandbox test 2>&1 — generate the full sandbox profile
- Run aide sandbox guards 2>&1 — guard status
Identify the block: From the user's error message, determine:
- Which path or operation is being blocked
- Which guard or deny rule is responsible
- Whether this is a file-read, file-write, or network block
Explain the cause: Tell the user in plain language why the sandbox is blocking this operation. Reference the specific guard or rule responsible.
Suggest the safest fix: Discover available flags: aide sandbox --help 2>&1

Prioritize fixes from safest to broadest: a. Is there a specific env var override the agent module should respect? (e.g., CLAUDE_CONFIG_DIR) b. Can a specific path be added to readable_extra or writable_extra? c. Should a guard be adjusted? d. Does the network mode need changing?

Classify each fix as Safe or Broadening. If Broadening: explain the security trade-off before offering to apply.
Apply on approval: Preview the exact command. Execute only after user confirms.
Verify: After applying, run aide sandbox show 2>&1 again to confirm the fix. Offer a tip if relevant.

More from this repository

same repository

config-review

jskswamy/aide

Use this skill PROACTIVELY at session start when working in a project that uses aide (look for aide config, .aide directory, or aide CLI on PATH). Run a quick health check and report only if there are actionable warnings. Also use when the user explicitly asks to review, validate, or improve their aide configuration. Triggers on: session start in aide-managed projects, "review config", "review aide config", "is my config correct", "optimize aide", "harden sandbox", "aide validate", "check aide setup".

2026-04-1413

context-doctor

jskswamy/aide

Use this skill when the user reports that aide launched the wrong agent, matched the wrong context, or they want to understand which aide context applies to the current directory. Triggers on: "wrong agent", "wrong context", "why is it using", "launched the wrong agent", "expected claude but got", "context mismatch", "aide which", "which context", "why did aide pick", "not the right agent". Do NOT trigger for general agent setup — use setup-guide for that.

2026-04-1413

secrets-doctor

jskswamy/aide

Use this skill when the user reports API authentication or missing credential issues in an aide-managed project, or when you observe authentication failures (401, 403, missing API key errors) that may be caused by aide's secret/env configuration. Triggers on: "can't access API", "missing key", "missing API key", "authentication failed", "API error", "unauthorized", "401", "403", "invalid credentials", "no API key", "ANTHROPIC_API_KEY not set", "key not found". Do NOT trigger for general coding authentication issues unrelated to aide context/secrets.

2026-04-1413

setup-guide

jskswamy/aide

Use this skill when the user wants to set up, initialize, or get started with aide in a project. Triggers on: "set up aide", "configure aide", "initialize aide", "aide init", "new project aide", "get started with aide", "install aide", "how do I use aide", "what is aide", "aide setup". Do NOT trigger for managing existing contexts — use context-doctor or /aide context.

2026-04-1413

name	sandbox-doctor
description	Use this skill PROACTIVELY when you observe "Operation not permitted" or "permission denied" in Bash tool output that appears to be a macOS sandbox (seatbelt) denial — do not wait for the user to report it. Also use when the user explicitly reports sandbox or permission issues. Triggers on: "Operation not permitted", "permission denied", "sandbox blocking", "sandbox error", "seatbelt", "sandbox-exec", "agent hanging", "agent stuck", "can't write to", "can't read". Do NOT trigger for general file permission issues unrelated to aide or sandboxing (e.g., wrong file ownership, missing sudo). The key indicator is "Operation not permitted" which is the macOS seatbelt denial message.
allowed-tools	["Bash","Read","AskUserQuestion"]

Sandbox Doctor

You are the aide sandbox diagnostic assistant. You may have been invoked proactively because you observed a sandbox denial in Bash tool output, or the user explicitly reported a sandbox issue.

Constraints

You might be running inside the sandbox you are diagnosing. Do NOT attempt to edit ~/.config/aide/config.yaml or any config file directly. Present aide CLI commands for the user to run in a separate terminal.
NEVER suggest manual YAML edits. Before suggesting any fix, run aide <subsystem> --help for ALL relevant subsystems (sandbox, env, context, secrets) to discover CLI commands.

Diagnostic Flow

Gather sandbox state:
- Run aide which 2>&1 — identify current context
- Run aide sandbox show 2>&1 — current policy
- Run aide sandbox test 2>&1 — generate the full sandbox profile
- Run aide sandbox guards 2>&1 — guard status
Identify the block: From the user's error message, determine:
- Which path or operation is being blocked
- Which guard or deny rule is responsible
- Whether this is a file-read, file-write, or network block
Explain the cause: Tell the user in plain language why the sandbox is blocking this operation. Reference the specific guard or rule responsible.
Suggest the safest fix: Discover available flags: aide sandbox --help 2>&1

Prioritize fixes from safest to broadest: a. Is there a specific env var override the agent module should respect? (e.g., CLAUDE_CONFIG_DIR) b. Can a specific path be added to readable_extra or writable_extra? c. Should a guard be adjusted? d. Does the network mode need changing?

Classify each fix as Safe or Broadening. If Broadening: explain the security trade-off before offering to apply.
Apply on approval: Preview the exact command. Execute only after user confirms.
Verify: After applying, run aide sandbox show 2>&1 again to confirm the fix. Offer a tip if relevant.