safe-ops
Skill1 starsChanged 41 days agoMITView on GitHub
Safely perform authorized remote, destructive or security-relevant operations, including Git push/merge, releases, infrastructure and data mutations, credential changes and bulk cleanup. Not required for routine read-only inspection or ordinary scoped repository edits.
- Commits and pushes
In this repository6 other files it gives its agents.
- guides/AGENTS.mdAGENTS.md
- adversarial-reviewSkill
- bddSkill
- mean-qaSkill
- morpheusSkill
- toolshedSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Ask when the action/target is ambiguous, blast radius exceeds the request, or new evidence materially changes the risk.
- Distinctive: line 27
Never infer ST/PROD write authorization from local-development permission.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Safe Ops
Make the safe path easy. Check authority, preview material effects, execute the named operation, and read back the result. No ritual approval loops.
One authorization policy
- A clear request naming an operation and a resolvable target authorizes that scope. "Commit and push this branch" and "open the PR" do not need another generic "proceed?". Show a compact preview and continue.
- A review, diagnosis or QA request does not authorize publishing or fixing. "Ship it" does not silently authorize merge, tag, deploy or database changes.
- Ask when the action/target is ambiguous, blast radius exceeds the request, or new evidence materially changes the risk. Do not turn an inferred goal into authority for a remote mutation.
- Destructive/history-rewriting actions require approval of the exact preview: targets, old/new state, losses, recovery and its limits. An earlier explicit approval of that unchanged preview remains valid.
- Higher-priority and environment-specific gates still apply, including production restrictions and the user's exact-diff approval for global Codex instructions/configuration/skills. Unknown environment = production. Never infer ST/PROD write authorization from local-development permission.
Risk, proportional ceremony
| Level | Typical action | Handling |
| L1 | Status, diff, bounded read-only query | Proceed within read-access rules |
| L2 | Authorized local commit, topic-branch fast-forward push | Short preview, execute, read back |
| L3 | Shared branch push/merge, PR publication, infrastructure/API writes | Verify named target, effects and existing authorization; ask only if missing |
| L4 | Force push, hard reset, destructive deletion, schema/data loss | Exact reviewed preview, explicit approval and recovery plan |
A topic push can trigger CI or preview deployment; discover meaningful side effects before treating it as L2. "Reversible" does not mean cost-free.
References
The policy above is canonical. Read only the relevant domain; "confirm" in a recipe means establish authorization under this policy, not automatically ask again. Examples are not authority.
- Git, CI and PRs: dev-workflow.md.
- Infrastructure: infrastructure.md.
- Databases/APIs: data-services.md.
- Files/processes/secrets: local-system.md.
- Multi-step or uncertain operations: safety-protocol.md.
Report one compact outcome per coherent operation: changed scope, observed result, important residual effects and recovery when material. No audit block for every file or command. Stop on unexpected state and inspect before retrying.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.