Agentspec — Agent, Tool & Guardrail Designer

Agentspec — Agent, Tool & Guardrail Designer — a premium, 100% offline digital tool by Aceplore. Runs entirely in your browser: no accounts, no subscriptions, your data never leaves your device.

A private, offline design bench for agents — the goal, the boundary, every tool classified by what it can change, the planned steps, the guardrails, the risk register and what one run costs at rates you enter — 31 tools in one, and the worst case at the step ceiling is costed as well as the planned path.

31 built-in tools in one app. An agent roster with a goal page whose keyword checks read your own wording, a boundary page listing what the agent must NOT do beside the tools that can change something, and a one-page agent record; a tool inventory where every tool is classified read, write, external or spend, an approvals view separating gated tools from the ones that fire unattended, and a permissions matrix of every agent against every distinct tool name with a cell for each pair; the planned step sequence carrying your own token estimates in and out per step, a costed breakdown for each one, a flow read top to bottom with cost accumulating down it, a step ceiling, and structural loop detection that reads the return-to-step field and names what bounds each cycle; guardrails grouped into input, output, tool, budget and human, a risk register scored severity by likelihood on scales you set, a matrix, coverage measured two ways because the two disagree, the unmitigated list in full, failure modes taken from the register AND from the structure of the design, fallbacks and handoffs; plus cost per run with the arithmetic shown and a worst case at the ceiling, a budget page checked against that worst case rather than the happy path, a printable specification, CSV and Markdown export, and reports across every agent.

Cost per run. A step you estimate at 3,000 input and 500 output tokens costs 0.0165 at the rates on your own card — 0.009 in plus 0.0075 out — and four planned steps come to 0.066. A 12-step ceiling is carried through and costed as well, because the ceiling is what has to be budgeted for. An agent with no steps reports no cost figure at all rather than printing 0.00 as though it had been costed.

Side effects. Every tool falls into one of four classes, and the order is an escalation: a read wastes a call, a write has to be undone, an external call has already been seen by someone outside, a spend has already left. Tools with nothing recorded are listed as unclassified rather than assumed harmless.

Coverage & the unmitigated list. Coverage is mitigated divided by total with the arithmetic shown, and the risks the percentage leaves out are listed by name rather than only implied. A risk ticked as mitigated with no guardrail linked to it is reported as a claim, not a control.

What's included:

• Agentspec.html
• Quick-Start-Guide.html
• How-To-Use-Guide.html
• README.txt
• LICENSE.txt

Includes a personalized welcome wizard, an interactive guided tour, a ⌘K command palette, light/dark themes with 8 accents, and one-click JSON backup & restore.

Agentspec is a planning, costing and record-keeping tool for people building with AI. Every figure it reports is computed by a stated formula from the rates, sizes, counts and thresholds YOU enter. It ships no model prices, no context windows and no tokeniser: those differ by provider and model and change without notice, so you enter them and the app shows which ones it used. Any token figure derived from the length of a text is an ESTIMATE, not a count — real tokenisers are model-specific and cannot run offline — so treat it as a planning figure and confirm actual usage against your provider's own reporting and billing before you commit to a budget. Nothing here is a guarantee of a cost, a saving, a quality level or a result, and none of it is business, financial, legal or engineering advice. You are solely responsible for your own decisions.