askuserquestion-depth-limit

Use this skill when building a post-manufacture auditor that scans agent execution logs and transcripts for pipeline instruction violations expressed in the…

Use this skill when a new pattern or convention needs to apply to all agents — write it once in CLAUDE.md rather than editing every agent file individually.

Use pipe-delimited strings in bash associative arrays to map a key to multiple values, then split with IFS='|' read -ra when consuming.

When a Claude Code PreToolUse hook intercepts an Agent tool call, the prompt and subagent_type live under .tool_input, not at the top level of the hook stdin…

How to wire cross-agent shared state through Claude Code PreToolUse/PostToolUse hooks: the orchestrator creates brain.json, hooks inject it into sub-agent…

After a worker agent returns in an orchestrator, verify the feature branch is ahead of main to detect cases where the worker committed to the wrong branch or…

How to safely capture the real project root path before entering a worktree, so that files written at end-of-manufacture land in the permanent project…

Env vars set via `export` inside a Claude Code Bash tool call are invisible to PreToolUse/PostToolUse hook processes — use a pointer file at a fixed path as…

In Claude Code PreToolUse hooks, stdout is the override channel for the tool input JSON — all diagnostic logging must go to stderr, never stdout.

When an agent detects that a long-lived resource (PR, branch, deployment) already exists, perform the side effects for that path (commit, push, record the…

Opens a pull request on GitHub using a fix already applied to the working tree.

Run a mandatory debugging workflow for non-obvious bugs using the Lewibs bug template and Solving a Bug checklist, then save a deduplicated audit log in…