Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/gitwormq/datasieve-mcp/skillnpx skills add gitwormq/datasieve-mcp --skill skillgit clone --depth 1 https://github.com/gitwormq/datasieve-mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00123 | $0.00886 |
| Opus 5 | $0.00062 | $0.00443 |
| Sonnet 5 | $0.00025 | $0.00177 |
| Haiku 4.5 | $0.00012 | $0.00089 |
Grade A, and why
datasieve scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 60 lines — stays where its author put it; the contents beside it link to each section on GitHub.
DataSieve — standing infrastructure for agents
You are ephemeral; DataSieve stays awake between your runs. One x402 payment
(USDC on Base, no account, no API key) buys a standing watch. If the
datasieve MCP server is connected, use its tools; otherwise any x402 client
against https://api.datasieve.xyz works (docs: GET /llms.txt).
When to reach for it (proactively)
-
Before long or risky autonomous work: arm a dead-man switch.
create_watchwith{type:"heartbeat", intervalSeconds: <2-3x your expected loop time>, label:"<task-name>"}, thenping_heartbeateach loop. If you crash, the alert fires and waits in the mailbox for whoever runs next. Pair with awebhookUrlthat re-launches you for self-healing loops. -
"Do X at/every TIME" with nobody awake to do it: schedule watch.
{type:"schedule", cron:"0 9 * * 1-5", tz:"America/New_York", payload:{...context for the woken agent...}}— the payload is echoed in every fire, so it is your note-to-future-self. One-shots:at. Simple loops:everySeconds. -
"Tell me when this page changes": url-change watch.
{type:"url-change", url, selector:".price"}— always scope with a CSS selector or regex; whole-page hashes false-positive on noisy pages. -
On-chain conditions on Base:
wallet-activity,token-transfers(whale alerts),price-cross,new-pairs. -
Fan-in / mutual exclusion across agents:
coord_workspaceonce, thencoord_barrier(wake me when all N sub-agents finish; hand each its arriveUrl) andcoord_lock(leases auto-expire — a crashed agent cannot deadlock the fleet). -
Evidence another agent/service can check: every event carries
receipt— an Ed25519 attestation by DataSieve. Verify withverify_receiptor client-side against/.well-known/datasieve-receipts.json.
Practical rules
- Call
payment_statusfirst (free) — reports wallet, balance, prices. - No wallet configured? All read tools still work; to buy, the user must fund a burner key (see the npm README). Never ask for a main wallet key.
- Tiers: standard $0.10/7d (1 condition), composite $0.25/7d (≤5 any-match, mix types freely), standard30 $0.35/30d, composite30 $0.85/30d. Prefer *30 tiers when nobody will be around to renew weekly.
- Save the returned watch id + secret into task state/memory — the id is the
persistent memory an ephemeral agent lacks. The MCP server also remembers
them in
~/.datasieve/watches.json. - Planned shutdown?
pause_watch(free) so the dead-man switch doesn't cry wolf;resume_watchrestarts clocks backlog-free. - No webhook infrastructure? Omit
webhookUrland drain the mailbox later withdrain_events— events wait server-side. - Failed x402 calls are never billed; settlement failures roll back.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 60 lines · 123 tokens per session scan A bac01ca84e4f
datasieve is a skill published in the GitHub repository gitwormq/datasieve-mcp (0 stars, last pushed 1mo ago), licensed MIT. It adds 123 tokens to every session and 886 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agent-self-scheduling
Schedule AI agent runs with cron, loops, or external clocks while avoiding unsafe tight autonomous timers.
plugin-develop
Implement Zhin.js plugin features with Plugin Runtime: defineCommand, defineAgentTool, middleware, pages. Use when asked to add a command, register a tool, wire cron, extend middleware, or build a console page. Triggers: 插件开发, add command, 加命令, 写插件功能, defineCommand.
create-hook
Create Cursor hooks. Use when you want to create a hook, write hooks.json, add hook scripts, or automate behavior around agent events.
cursor-coding
Spawn a Cursor coding agent for real code work. Use when the user asks for actual code changes -- refactors, bug fixes, new features, repo edits -- that need full file/terminal access and iterative tool calls. Hand off via the cursoragent MCP tool.
find-skills
Discover and install agent skills from skills.sh — the definitive open skills registry. Use when the user asks "how do I do X", "find a skill for X", "is there a skill for X", wants to extend agent capabilities, or when a task might benefit from a specialized skill you don't already have. Based on vercel-labs/skills…
dedalus-machines
How Agent Machines work -- providers (Dedalus, E2B, Sprites), agent runtimes (Hermes, OpenClaw, Claude Code, Codex), MCP catalog, bootstrap, and closed-loop verification. Use when asked about the runtime, scaling, sleep/wake, destroy, or inspecting this environment.