agentleFS
VeeiyVeeiy/vantage-os/skills/permissions/SKILL.md

permissions

Skill0 starsChanged 5 months agoMITView on GitHub

Permissions and kill-switch. Use when the operator says "permissions", "authority", "tier", "kill switch", "stop", "shut it down", "pause", "raise tier", "lower tier", or any time the system is about to perform an action with real-world consequences (sending email, posting listing, committing money). Defines and enforces the authority tier the coordinator and all sub-agents operate under.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

20 rules. 0 are common, 17 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 17

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No spending >$50 in a single action without the operator's explicit per-action approval

    line 50
  2. Distinctive:

    No trading. the operator trading restrictions apply and the system has no broker access.

    line 51
  3. Distinctive:

    No posting to public social media on the operator's personal accounts.

    line 53
  4. Distinctive:

    No a partner-related actions (her platform, their work) without her explicit OK.

    line 54
  5. Distinctive:

    When the operator says "raise to tier 2" or similar:

    line 60
  6. Distinctive:

    Verify the tier change is intentional.

    line 62
  7. Distinctive:

    If raising to Tier 3, list the auto-rules currently configured and confirm.

    line 63
  8. Very specific:

    Update permissions.md. Log the change with timestamp and reason.

    line 64
  9. Distinctive:

    When the operator says "stop", "kill", "shut it down", "emergency":

    line 69
  10. Distinctive:

    Write the kill event to the audit log.

    line 73
  11. Distinctive:

    Ask the operator what triggered the kill so the lesson is captured.

    line 74
  12. Very specific:

    Before any sub-agent or skill takes an external action, it must consult permissions.md:

    line 78
  13. Distinctive:

    Check if action is allowed at this tier.

    line 81
  14. Distinctive:

    If not allowed, refuse and propose what tier would be needed.

    line 82
  15. Distinctive:

    If allowed, perform action and log to audit.

    line 83
  16. Very specific:

    When nightly-wrap runs and the operator will be away for 8+ hours:

    line 87
  17. Distinctive:

    If tier is 2 or 3, default-recommend dropping to 1.

    line 89
  18. Distinctive:

    If the operator explicitly wants to maintain tier overnight, log the override with reason.

    line 90
  19. Distinctive:

    Tier never auto-raises: Only the operator raises.

    line 118
  20. Distinctive:

    Audit log is append-only: Never edit history.

    line 120

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Permissions

The blast-radius controller for vantage-os.

State file

~/Vault/Vantage-OS/permissions.md

Schema:

## Current tier: [1 / 2 / 3]
Last changed: YYYY-MM-DD HH:MM by [the operator]
Reason: [...]

## Tier 2 templates pre-approved (only consulted at Tier 2+)
- [name]: [description, e.g., "rental-fleet standard pre-arrival message"]

## Tier 3 auto-rules (only consulted at Tier 3)
- [rule]: [trigger -> action, e.g., "If rental-fleet trip ends -> send post-trip thank-you using template Z"]

## Audit log (last 30 days)
- YYYY-MM-DD HH:MM | tier | action | result

Tier definitions

Tier 1 - Research and draft only (DEFAULT)

  • Vantage-OS may research, analyze, draft messages/listings/code/emails.
  • Vantage-OS may NOT send, post, commit money, sign anything, or take any action visible to a third party.
  • Drafts queue for the operator's explicit per-item approval.
  • This is the default for all new sessions and overnight periods.

Tier 2 - Send pre-approved templates

  • Everything in Tier 1, plus:
  • Vantage-OS may send messages and post listings using templates the operator has pre-approved and added to permissions.md under "Tier 2 templates pre-approved."
  • Each sent action is logged in the audit log.
  • Vantage-OS still may not commit money or sign anything.
  • the operator must add templates explicitly. No template = no send.

Tier 3 - Act on documented rules

  • Everything in Tier 2, plus:
  • Vantage-OS may execute documented rules under "Tier 3 auto-rules" without per-action approval.
  • Hard limits regardless of tier:
  • No spending >$50 in a single action without the operator's explicit per-action approval
  • No trading. the operator trading restrictions apply and the system has no broker access.
  • No signing contracts.
  • No posting to public social media on the operator's personal accounts.
  • No a partner-related actions (her platform, their work) without her explicit OK.

Process

Setting tier

When the operator says "raise to tier 2" or similar:

  1. Verify the tier change is intentional. Ask: "Tier 2 lets me send pre-approved template messages without per-message approval. Right now you have [list templates] pre-approved. Confirm raise?"
  2. If raising to Tier 3, list the auto-rules currently configured and confirm.
  3. Update permissions.md. Log the change with timestamp and reason.
  4. Notify in chat: "Now Tier N. Blast radius: [list current Tier N capabilities]."

Kill switch

When the operator says "stop", "kill", "shut it down", "emergency":

  1. Immediately set tier to 1.
  2. Halt any in-progress sub-agent dispatches.
  3. Write the kill event to the audit log.
  4. Ask the operator what triggered the kill so the lesson is captured.

Pre-action check

Before any sub-agent or skill takes an external action, it must consult permissions.md:

  1. Read current tier.
  2. Check if action is allowed at this tier.
  3. If not allowed, refuse and propose what tier would be needed.
  4. If allowed, perform action and log to audit.

Tier auto-drop overnight

When nightly-wrap runs and the operator will be away for 8+ hours:

  1. If tier is 2 or 3, default-recommend dropping to 1.
  2. If the operator explicitly wants to maintain tier overnight, log the override with reason.

Pre-approved Tier 2 template format

### [Template name]
- Channel: [e.g., rental-fleet message, Gmail draft, etc.]
- Trigger: [when this template applies]
- Body:
 [exact text, with {{variables}} for substitution]
- Variables: {{var1}} (description), {{var2}} (description)

Audit examples

Good entry:

2026-05-17 09:14 | Tier 2 | Sent rental-fleet pre-arrival message to renter Jane D. for Model 3 Performance trip starting 2026-05-17 14:00 | Sent successfully

Bad entry (the kind that should trigger the operator's attention):

2026-05-16 02:30 | Tier 3 | Attempted post-trip thank-you to renter Carlos M. | FAILED: rental-fleet MCP unavailable. Queued for retry.

Rules

  1. Default is Tier 1. Always. Sessions start at Tier 1 unless the operator explicitly raises in-session.
  2. Tier never auto-raises. Only the operator raises.
  3. Hard limits override tier. Even at Tier 3, the hard limits hold.
  4. Audit log is append-only. Never edit history.
  5. If permissions.md is missing or corrupted, default to Tier 1 and warn the operator.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.