blast-radius-rm-hold-unattended-subagent-stall
A Bash call containing rm -rf or rm -f (test loop, staging script, cleanup) is held by a safety hook (e.g. Blast Radius) for 10 minutes and never runs, with 'no answer within 10 minutes' or a denial for an unresolved shell variable target. Use when writing any agent-authored script, test harness or cleanup step, or when dispatching a report-only subagent that may run shell commands.
- License 许可
- MIT
- 最近更新
- 2026-11-07
- Stars
- ★ 1
- Forks
- 1
01 前言 Front Matter
- name
- blast-radius-rm-hold-unattended-subagent-stall
- description
- A Bash call containing rm -rf or rm -f (test loop, staging script, cleanup) is held by a safety hook (e.g. Blast Radius) for 10 minutes and never runs, with 'no answer within 10 minutes' or a denial for an unresolved shell variable target. Use when writing any agent-authored script, test harness or cleanup step, or when dispatching a report-only subagent that may run shell commands.
03 SKILL.md 正文
Agent scripts with rm: the approval hold that stalls unattended agents
Symptom
Blast Radius held this command and did not run it: no answer within 10 minutes. It would have: delete nothing: no file matches $O.Safety check blocked the rm -rf in my test loop.The agent rewrites the script to use a fresh temp dir.- Unattended subagents tried to delete reviewer probe files they did not own. A report-only sweep reviewer sat blocked for 10 minutes; the next dispatch needed a warning that a previous sweeper hung on a blocked call.
Root cause
An approval hook (e.g. Blast Radius) asks for confirmation on every delete and cannot resolve
computed paths ($O, loop variables). A subagent has no human to answer, so
the hold runs to its timeout. Reviewer briefs did not forbid destructive
commands, so agents reached for rm as routine tidy-up.
Fix
- Never put
rmin agent-written scripts or loops. Make deletion unnecessary:OUT="$(mktemp -d "${TMPDIR:-/tmp}/scenarios.XXXXXX")" # fresh dir per run # (or use the session scratch dir) run_scenario "$OUT/case-1" # nothing to clean up afterwards - Write harnesses to a file with the Write tool and run the file, rather
than long inline
zsh -c '...'one-liners: files are reviewable and re-runnable. This is NOT a way around the hook. If a delete is genuinely required, stop and hand the user the exact command; never moverminto a file to get past the hook. - Never delete files you did not create. Leave cleanup to the user or a scratch-dir TTL.
- If a command is held, rewrite it without the delete. Do not retry it unchanged and do not wait out the timeout.
- Dispatch prompts for reviewers and verifiers (paste this):
COMMAND LIMITS: report-only. Do not run rm, mv, network or install commands. Use a fresh mktemp -d "${TMPDIR:-/tmp}/<name>.XXXXXX" for scratch output. Tool-call budget: 25. If any command is held or blocked by a hook, STOP and report it as blocked; do not retry or work around it.
Evidence
Mined from session digests; 8 sessions across 3 projects. claude-skills
094ecb81 (the Blast Radius hold text above), a private web project 41c27adc (safety
check blocked an rm -rf in a test loop; rewritten to a temp dir per
scenario), a private-bot worktree 60710e29 (a report-only sweeper
hung on a blocked call).
Related skills
core-workflow:subagent-report-truncation-delivery-contract— also paste its report block into the same reviewer dispatches.core-workflow:subagent-buildverify-tool-grant-check— the grant check to run before dispatching shell-running agents.core-workflow:worktree-isolation-guard-refuses-complex-bash— the other hook that refuses agent shell commands.
04 同仓库更多 Skills
second-wind
Orchestrate long unattended Claude Code runs across multiple repos with the wind CLI. Use when the user wants to run Claude Code overnight, resume sessions after the 5-hour usage limit, run Claude in many repos at once via tmux, or says "set up second wind", "wind up", "overnight run", "resume after limit", "usage limit orchestrator".
structured-output-adaptive-thinking-token-bloat
A Claude API call using output_config.format (json_schema structured outputs) with thinking:{type:"adaptive"} burns most of its output budget on thinking (observed ~21k of ~24k tokens) without improving JSON quality, making the run up to ~5x more expensive than the same call with thinking omitted — use when calling the Messages API with a json_schema output format for extraction, ranking, classification, or summarization tasks, or when a schema-constrained call's usage.output_tokens is far larger than the returned JSON.
review
Apply a code-review lens to a diff, branch, or file — find correctness bugs, security issues, and quality problems, ranked by severity, with a concrete fix for each. Use when the user says "review this", "review my diff", "check this PR", "look this over before I commit", or "is this safe". Skip when the user wants the code written, not reviewed.
refresh
Rebuild the prompt-craft command registry and usage profile, and print a summary of installed commands and your personalized recommendations. Use when you install or update plugins and want the advisor to pick them up immediately, or say "refresh prompt-craft", "rebuild the command registry".
refactor
Apply a refactoring lens — restructure code without changing behavior, guarded by tests that pass before and after. Use when the user says "refactor", "clean this up", "extract this", "simplify this", "reduce duplication", or "make this readable" on existing working code. Skip when the change is meant to alter behavior (that's a feature or a fix).
plan
Decompose a non-trivial task into an explicit, tracked plan with goals and per-step acceptance criteria before writing code. Use when the user says "plan this", "break this down", "what's the plan", or asks for a multi-step feature/refactor that touches more than one file. Pairs with plan mode. Skip for single-edit changes that need no decomposition.