Agent Safety Pack
Agent Safety Pack: Fail-Closed Guards, Deletion Safety Gate, and Evidence-Gated Delegation. $22, saves $5 versus $27 singles. do not let my agent do something stupid. Catch dead guards, prove a path is unreferenced before deleting, and gate delegation on evidence, not self-reports.
Works with any agent that signs in to Agensi, such as Claude, ChatGPT, Claude Code, Cursor and Codex CLI.
One payment, lifetime access. 2 skills unlock instantly in your library.
What's included
2 skillsA read-only, fail-closed check your agent runs before deleting any worktree, clone, cache or build folder. It screens scheduled jobs, crontab, config files, running processes and shared git object stores for references to the path, and returns SAFE only when every screen finishes with zero hits.
A framework for letting sub-agents, worker fleets or pipelines act for you safely. Gate each task family on a track record, verify every result with deterministic evidence instead of the delegate's own "it passed", and climb a Shadow → Propose → Autonomous ladder one rung at a time.
See it in action
Deletion Safety Gate — Fail-Closed Check Before rm -rf
Example session with this skill installed
My disk is almost full. Before deleting these old worktrees, run deletion-safety-gate on each path and only delete the ones it reports SAFE:
~/code/worktrees/checkout-redesign
~/code/worktrees/nightly-report
~/code/worktrees/old-spike
- Read your context and instructions
- Compiled the deletion safety gate
deletion-safety-gate — 3 candidates, checked one at a time (read-only)
-
~/code/worktrees/checkout-redesign
spellings: $HOME/code/worktrees/checkout-redesign | ~/code/worktrees/checkout-redesign | /worktrees/checkout-redesign
loaded jobs: 0 | job files: 0 | crontab: 0 | config/scripts: 0 | processes: 0 | shared git store: not a hub
→ SAFE (git clean, 0 unpushed, recovery ref saved: refs/archive/checkout-redesign) -
~/code/worktrees/nightly-report
loaded jobs: 1 HIT — com.dev.nightly-report (WorkingDirectory=/worktrees/nightly-report, matched bare spelling)
job files: 1 HIT — ~/Library/LaunchAgents/com.dev.nightly-report.plist.bak
config/scripts: 1 HIT — ~/bin/run-report.sh:4
→ REFUSE (referenced by a live scheduled job; git status was clean, which is exactly why this check exists) -
~/code/worktrees/old-spike
config/scripts: could not read ~/.config/tooling/state.db (permission denied)
→ REFUSE (screen incomplete — fail closed)
Deleted: 1 (checkout-redesign). Kept: 2.
Next steps: retarget or unload com.dev.nightly-report before deleting #2; fix read access and re-run for #3.
Not screened: open file descriptors, remote CI definitions, paths built at runtime.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads these skills and uses them on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every skill in this bundle goes through our trust checks.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.