Plugin marketplace
Agent skills that make an agent's work checkable: numbers with receipts, fake data that can't pass for real, handovers verified by a fresh agent, run settings shown before a costly run, and work planned around usage limits.
Add this marketplace
Then install any plugin below by name — each page has its exact command.
I-PASS handovers for coding agents: /shift-change hands your session to the next one and quizzes a fresh agent on it, /shift-start reads it back before acting, /shift-recover rebuilds a session that ended without one, /shift-brief packages a task for a collaborator.
4 skills
Before an agent states results, metrics, counts or costs, it writes a receipt for each number (file + exact quote, logged command output, calculation, or the user's words) and a checker verifies it. Receipts expire when their inputs change. A Stop hook sends back any reply with an unbacked number. Unmeasured values are N/A, never 0.
1 skill · hooks
Before training, a sweep, a cluster submission or any costly command, runcard resolves every setting it will actually use, says where each came from, flags quietly wrong combinations, and asks once.
1 skill · hooks
When an agent generates fake, mock, seed or synthetic data: only reserved values (RFC 2606 emails, NANP 555-01xx and Ofcom drama phones, processor test cards, never-issued SSNs, RFC 5737 IPs), a seeded generator script, and a data card. decoy.py lints the result.
1 skill
Reads the status line rate limits (5-hour and 7-day windows), measures burn rate, and orders the backlog so heavy tasks land in fresh headroom and long non-model jobs run during the lockout. v0 prototype.
1 skill
Shows how many plugins this repo ships, and links here.