budget-guardrails
Skill4 starsChanged 4 months agoMITView on GitHub
Cost discipline protocol for every Paperclip agent — check remaining budget before expensive operations, tier actions as cheap/moderate/expensive, switch to warning behavior at 80% of the monthly cap, hard-stop and report at 100%, and never work around a budget limit. Use before any costly operation, when asked "can I afford this", "check my budget", "how much have I spent", or whenever a budget warning or cap is hit. Don't use for deciding whether work is worth doing (ceo-operating-doctrine) or writing a ticket's budget field (delegation-protocol).
In this repository74 other files it gives its agents.
- agent-eval-harnessSkill
- context-budget-managerSkill
- incident-rollbackSkill
- memory-hygieneSkill
- model-cost-optimizerSkill
- prompt-debuggerSkill
- skill-author-coachSkill
- third-party-skill-vetterSkill
- api-endpoint-testerSkill
- bug-triage-protocolSkill
- changelog-release-notesSkill
- docs-readme-writerSkill
- env-secrets-hygieneSkill
- github-pr-reviewSkill
- git-ship-flowSkill
- lovable-brief-writerSkill
- supabase-migration-writerSkill
- vercel-deploy-checkSkill
- cold-outreach-sequencerSkill
- competitor-pricing-scanSkill
- contract-redline-scanSkill
- customer-support-inboxSkill
- invoice-runnerSkill
- kpi-dashboard-weeklySkill
- lead-research-dossierSkill
- payment-reconciliationSkill
- pipeline-trackerSkill
- proposal-builderSkill
- stripe-payment-linkSkill
- testimonial-harvesterSkill
- weekly-revenue-reportSkill
- brand-voice-guardianSkill
- case-study-builderSkill
- content-repurposerSkill
- deep-research-briefSkill
- email-nurture-sequenceSkill
- landing-page-copywriterSkill
- linkedin-content-packSkill
- newsletter-assemblerSkill
- seo-blog-writerSkill
- social-listening-reportSkill
- x-thread-builderSkill
- youtube-script-writerSkill
- ceo-operating-doctrineSkill
- delegation-protocolSkill
- editor-personaSkill
- heartbeat-standupSkill
- hiring-briefSkill
- ops-agent-personaSkill
- researcher-personaSkill
- sales-agent-personaSkill
- support-agent-personaSkill
- weekly-retroSkill
- writer-personaSkill
- calendar-conciergeSkill
- decision-journalSkill
- email-triageSkill
- file-organizerSkill
- habit-trackerSkill
- inbox-zero-sprintSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 19
Before starting any moderate or expensive operation (see tiers below).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Budget Guardrails
Every Paperclip agent has a hard monthly budget. This skill is the spending reflex that keeps the org solvent: know the cost tier of what you're about to do, check the balance before anything non-cheap, get loud at 80%, and stop dead at 100%. The standard: no spend surprises the human, ever — and a cap is a wall, not a suggestion.
When to Use
- Before starting any moderate or expensive operation (see tiers below).
- At the start of every wake cycle and before accepting any new ticket.
- The moment a budget warning fires or spend crosses 80% of the cap.
- Not for: deciding whether work is worth doing at all (that's
ceo-operating-doctrineprioritization) or writing the ticket's budget field (that'sdelegation-protocol— this skill checks the money is actually there).
Role Boundary
- You own: knowing your own spent/remaining numbers, tiering every action before you take it, reporting spend per ticket, stopping at limits.
- You escalate: any single action estimated over 10% of your monthly cap (CEO pre-approval on the ticket), crossing 80% (warning report to CEO), hitting 100% (hard stop + report, then silence on paid work), and any real-money purchase of any size (human approval, always).
- You never: work around a cap — no splitting one expensive action into many cheap ones, no borrowing against next month, no routing spend through another agent, no "finishing this one thing" past the limit.
Quick Reference
| Tier | Typical actions | Rule before acting |
| Cheap | Ticket reads/comments, planning, small file edits, short generations | Go. Tracked, not gated. |
| Moderate | Web research runs, long-document generation, batch processing | Check remaining budget first; proceed if the estimate fits under the 80% line. |
| Expensive | Large model calls in bulk, big scrapes/datasets, any paid API or real-money purchase | Check budget + CEO pre-approval on the ticket if >10% of cap. Real money → human approval, no exceptions. |
| Threshold | Behavior |
| <80% of cap | Normal work; log spend per ticket |
| 80–99% | Warning mode: essentials only, CEO notified, expensive tier frozen |
| 100% | Hard stop: finish the sentence, post the report, do nothing else that costs |
Procedure
- Wake-cycle check — first action of every wake: fetch spent vs. cap. Success: you can state "spent X of Y (Z%)" before touching any ticket.
- Tier the action — before any operation, name its tier out loud (in your working notes or the ticket). Unsure which tier? Treat it as the higher one.
- Gate moderate/expensive actions — estimate the cost; confirm
spent + estimatestays under the 80% line. Expensive actions >10% of cap additionally need CEO approval as a ticket comment before execution. Real-money purchases need explicit human approval, always, with a dry-run/quote posted first ("this will charge $X to Y — confirm?"). - Log spend to the ticket — after each moderate/expensive action, note actual cost on the ticket so goal ancestry shows what each goal really costs.
- At 80% — warning mode — post the warning report (template below) to the CEO, freeze the expensive tier, and only continue moderate work that an in-flight ticket's deadline requires. New tickets get accepted only with CEO acknowledgment of your budget state.
- At 100% — hard stop — stop all cost-incurring work mid-ticket if necessary. Post the hard-stop report to your current ticket and to the CEO. Then do only free actions (reading, commenting, planning). Do not resume until the human or the platform raises the cap — a CEO cannot override a hard cap.
- Never route around — if you catch yourself designing a workaround (splitting actions, using another agent's budget, deferring the logging), that is the failure. Stop and escalate instead.
Output Template
## Budget {Warning 80% | HARD STOP 100%} — {agent name}, {date}
Spent: {amount} / {cap} ({%}) | Month resets: {date}
### Where it went (top tickets)
- {ticket-id} ({project → goal}): {amount}
- {ticket-id} ({project → goal}): {amount}
### In flight
- {ticket-id}: {state — e.g. "draft posted, revision round pending, est. {cost} to finish"}
### Recommendation
{e.g. "Approve {amount} top-up to finish T-14 by Friday, or reassign it — all else can wait for reset."}Pitfalls
- Estimate optimism — calling a scrape "moderate", watching it run expensive, discovering it at 97%. Recovery: when actuals exceed estimate by >50%, stop the operation, log the real number, and re-gate before continuing.
- The cap workaround — ten cheap calls that add up to the expensive one you weren't approved for. Recovery: tiering is by total intended operation, not per call; recombine the estimate, and if it needed approval, get it retroactively flagged to the CEO — visibly, on the ticket.
- Silent 80% crossing — warning threshold passes mid-task and nobody hears until the hard stop. Recovery: the wake-cycle check (step 1) exists for this; if you find yourself past 80% unreported, post the warning report immediately, timestamped honestly.
- Stopping without reporting — agent hits 100% and goes quiet; the CEO thinks it's stalled. Recovery: the hard stop is stop-AND-report; if you stopped silently, the report is free — post it now to the current ticket and CEO.
Verification
- Spent/cap/% stated at the top of this wake cycle
- Every moderate+ action this cycle was tiered and gated before execution
- Actual costs logged to their tickets (goal ancestry shows spend)
- If ≥80%: warning report posted, expensive tier frozen
- If 100%: all paid work stopped and hard-stop report posted
- Zero workarounds: no split actions, no borrowed budgets, no unlogged spend
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.

