| name | codex-fewer-permission-prompts |
| description | Codex permission prompts helper for /fewer-permission-prompts default dry-run, safe prefix_rule proposals, execpolicy verification, apply, and rollback. |
Use this skill to reduce repeated approval prompts without weakening Codex's sandbox or approval model. Keep approval_policy and sandbox_mode intact; work only through official .rules files and prefix_rule(...).
Default Action
When the user invokes this skill with Codex Fewer Permission Prompts, $codex-fewer-permission-prompts, /fewer-permission-prompts, or no subcommand, do not explain what the skill is. Start the dry-run workflow:
python -m codex_fewer_permission_prompts doctor --json
python -m codex_fewer_permission_prompts propose --dry-run
Summarize the environment status, the new candidate count, and the most useful proposed low-risk rules. Existing matching prefix_rule(...) entries in the current rules file are omitted from the dry-run candidate list; if no new candidates remain, say that the existing rules already cover the observed safe commands. State clearly that dry-run did not modify files. Do not run verify against the current default.rules for a dry-run proposal; positive examples are expected not to match until rules are applied. If the user asks to verify a dry-run proposal before applying, write the proposal JSON to a temporary file and run:
python -m codex_fewer_permission_prompts verify --proposal-json <proposal.json>
Trigger Patterns
- Natural language:
减少 Codex 权限确认, 生成低风险 rules 建议, 清理/验证 Codex allow rules
- Lifecycle: install, update, uninstall, doctor, status, scan, propose, apply, verify, rollback
- Slash-style text:
/fewer-permission-prompts, /fewer-permission-prompts doctor, /fewer-permission-prompts propose, /fewer-permission-prompts apply, /fewer-permission-prompts rollback
Codex currently documents built-in slash commands only. Treat /fewer-permission-prompts ... as user text that this skill maps to the matching workflow; do not claim it registers a native Codex slash command.
How To Execute
Prefer the installed module:
python -m codex_fewer_permission_prompts
python -m codex_fewer_permission_prompts /fewer-permission-prompts
python -m codex_fewer_permission_prompts doctor
python -m codex_fewer_permission_prompts status
python -m codex_fewer_permission_prompts scan
python -m codex_fewer_permission_prompts propose --dry-run
python -m codex_fewer_permission_prompts verify --rules-file $HOME\.codex\rules\default.rules
python -m codex_fewer_permission_prompts apply --rules-file $HOME\.codex\rules\default.rules
python -m codex_fewer_permission_prompts apply --rules-file $HOME\.codex\rules\default.rules --write
python -m codex_fewer_permission_prompts rollback --rules-file $HOME\.codex\rules\default.rules --remove-block
When running from the source repo before install, use the wrapper:
python scripts\codex_fewer_permission_prompts.py doctor
When Codex loads this as a plugin skill and the Python package is not installed, use the bundled skill wrapper:
python <skill-folder>\scripts\codex_fewer_permission_prompts.py doctor
The CLI also accepts python -m codex_fewer_permission_prompts /fewer-permission-prompts doctor as a slash-style compatibility form.
Lifecycle
- Install: fetch and follow
.codex/INSTALL.md from this repo.
- Update: fetch and follow
.codex/UPDATE.md.
- Uninstall: fetch and follow
.codex/UNINSTALL.md.
- After installing, updating, uninstalling, or changing rules, tell the user to restart Codex App or open a new CLI session so skill discovery or rules are reloaded.
Safety Model
- Default to dry-run. Do not edit rules unless the user explicitly approves the apply step.
- Recommend only frequent, explainable, low-risk read-only or diagnostic command prefixes.
- Never recommend delete, move, write, install, registry, ACL, service,
git reset, git checkout, git clean, git push, git commit, or complex shell script rules.
- Reject commands with redirection, variable expansion, wildcards, shell control flow, or sensitive path hints.
- Output only command summaries and counts. Do not print raw transcript text, secrets, tokens, or full conversation content.
- Before writing a
.rules file, create a timestamped backup and update only the codex-fewer-permission-prompts sentinel block.
Command Map
doctor / status: locate CODEX_HOME, rules files, sessions, history, and logs; summarize JSONL shapes without printing content.- default / bare
/fewer-permission-prompts: run doctor --json, then propose --dry-run; never writes rules.
scan / analyze: count observed shell commands from Codex session JSONL files.propose: classify observed commands and print new candidate prefix_rule(...) entries with match and not_match examples, omitting commands already covered by the current rules file.verify: run codex execpolicy check --pretty --rules <rules-file> -- <command...> for each generated or sentinel rule example. With --proposal-json and no --rules-file, verify against a temporary rules file generated from the proposal.apply: show a unified diff by default. Add --write to ask for confirmation, back up the rules file, then create the sentinel block or append only new uncovered candidates to it. Use --yes only for controlled tests or when the user already approved the exact write.rollback: restore the latest backup, restore a named backup, or remove only the sentinel block.
If Codex official docs or local codex execpolicy check behavior disagree with this skill, treat the current official docs and local CLI behavior as the source of truth.
