From autopilot
This skill should be used when a tool call is denied in autonomous mode. Guides through checking existing permissions for alternative paths, and if none exist, proposing well-scoped permission rules — without it, denied tool calls tend to either block on user input or cause subtasks to be silently skipped.
How this skill is triggered — by the user, by Claude, or both
Slash command
/autopilot:resolve-denied-toolcallThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Proposing a permission is costly — it blocks progress until the user notices and approves it, and they may be away, asleep, or running other agents. Invest real effort into checking whether the allow list already covers what's needed through a different path.
Proposing a permission is costly — it blocks progress until the user notices and approves it, and they may be away, asleep, or running other agents. Invest real effort into checking whether the allow list already covers what's needed through a different path.
Read .claude/settings.json and .claude/settings.local.json (in the working directory, not global). Look for permitted commands that accomplish the same thing — especially project scripts and allowed package manager invocations.
If the deno sandbox is available, consider whether the task can be done by writing TypeScript in the sandbox instead.
Examples of non-obvious alternatives:
Bash(bash scripts/check-gpu.sh *) is allowed — no need for ssh lab-server nvidia-smi. The script already SSHes to the configured server.Bash(bun run db:query *) is allowed — no need for psql -c "SELECT...". The script runs queries with proper credentials.Bash(uv run scripts/download_model.py *) is allowed — no need for curl -L -o models/gpt2.bin https://.... The script handles authenticated downloads from the project's model registry.If the denied command only blocks one part of the task, move on to other parts. Batch the permission proposal for when no further progress is possible.
Only after confirming no permitted alternative exists, read references/adding-permissions.md for how to propose a well-scoped rule.
npx claudepluginhub crazytieguy/alignment-hive --plugin autopilotGuides Claude Code permission configuration, rules (allow/ask/deny), modes, tool-specific patterns (Bash/read/edit/WebFetch), /permissions command, and troubleshooting via docs-management delegation.
Configures tool permissions for Claude Code agents via --allowedTools flags, path patterns for file access, and Bash command prefixes. Reference for secure launches.
Configure allow/deny/ask permission rules in .claude/settings.json for Claude Code tools like Bash(git:*), Write, Edit. Builds layered policies with glob patterns for git commands.