Approval Report Skill

Track agent approval requests, generate weekly reports of repeated approval patterns, and add explicitly confirmed allow-prefix rules.

Why

Approving every repeated command gets old fast. Allowing everything by default is far too dangerous.

The just-in-time "always allow" option is often too narrow: it may approve one exact command while leaving similar future commands to ask again. This skill surfaces common approval patterns, ranks them by risk, and makes it easier to add safer, more generic prefixes to your config.

What It Does

• Logs PermissionRequest events without approving or denying them.
• Reports repeated approval patterns grouped by conservative candidate prefixes.
• Excludes commands already covered by existing allow-prefix rules.
• Cleans processed raw logs only after a successful report.
• Adds prefix rules only when explicitly requested.

Install

From this skill directory:

python3 scripts/install.py --agent codex

The installer resolves Codex home from $CODEX_HOME, then ~/.codex.

Once the skill is installed, you can also ask Codex to run it for you, for example:

Use the approval-report skill to show this week's approval report.

Run A Report

python3 scripts/report.py --agent codex

Add Confirmed Prefixes

Only run this after deciding the exact prefixes are safe:

python3 scripts/add_prefix_rules.py '["docker", "logs"]'

Claude

The shared report and prefix scripts support --agent claude, but automatic Claude hook installation is intentionally not claimed until the target machine's Claude hook schema is inspected.