Tenacity — Never Give Up

WarnAudited by ClawScan on May 18, 2026.

Overview

This skill openly makes the agent keep working without confirmations, but it gives broad file/command authority, persistence, and Telegram reporting that need careful review.

Install or invoke this only for tightly scoped tasks. Before use, edit the standing order to limit allowed tools, require approval for deletes/commands/external actions, remove the hard-coded Telegram recipient, set an expiration for any cron/background run, and clean checkpoints/memory logs when done.

Findings (6)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The agent may take many actions without interim review, even when the user would normally expect checkpoints or approvals.

Why it was flagged

This intentionally changes the agent's normal stopping and confirmation behavior for potentially broad tasks.

Skill content
Default behavior: Ask for confirmation only when blocked, not at every step. If the path is clear, execute.
Recommendation

Use only with a narrow written scope, explicit stop conditions, and separate approval for destructive, external, or high-impact actions.

What this means

A mistaken instruction or bad intermediate result could lead to unwanted file edits, deletions, or command execution before the user sees it.

Why it was flagged

The no-ask authority includes destructive file operations and broad command/tool execution.

Skill content
What I Can Do Without Asking ... Read, write, edit, delete files in workspace ... Run scripts and commands ... Use any tool in the workspace
Recommendation

Require confirmation for deletes, bulk changes, risky commands, and any tool outside a task-specific allowlist.

What this means

A standing order could be interpreted as ongoing consent for later actions, not just the immediate task.

Why it was flagged

The skill frames standing orders as permanent delegated authority, but the artifacts do not define expiry, revocation, or mandatory re-confirmation.

Skill content
| Standing Orders | Grants permanent authority to execute within scope |
Recommendation

Make every standing order task-specific, time-limited, and automatically revoked or reviewed at completion.

What this means

Task summaries or status updates could be sent to a recipient that is not the installing user.

Why it was flagged

The cron example sends announcements to a hard-coded Telegram recipient, creating an unclear external data boundary.

Skill content
--announce \
  --channel telegram \
  --to "834732674"
Recommendation

Remove the hard-coded recipient, require the user to configure the destination, and confirm before sending sensitive task summaries.

What this means

The agent may continue operating in the background after the initial interaction unless the user actively stops it.

Why it was flagged

The skill encourages persistent autonomous execution that can survive restarts and run on a schedule.

Skill content
| Task Flow | Durable multi-step state across gateway restarts | ... | Cron | Scheduling and enforcement |
Recommendation

Require explicit opt-in for cron/background runs, set an end time, and document how to list, pause, and remove scheduled Tenacity jobs.

What this means

Old or tampered checkpoint data could influence a resumed task, and local checkpoint files may reveal task details.

Why it was flagged

Checkpoint state is stored locally and later reused as resume context, which is purpose-aligned but should not contain secrets or be blindly trusted.

Skill content
CHECKPOINT_DIR="${CHECKPOINT_DIR:-/tmp/tenacity-checkpoints}" ... cat "$LATEST"
Recommendation

Use a private checkpoint directory, avoid storing secrets in checkpoint state, and clean or validate checkpoints before resuming.