Language Release

ICC DSL v1.01

First public ICC DSL release with native percent references, intent-cell routing, execution constraints, output commands, branching, artifacts, and legacy reference migration.

Abstract

ICC DSL v1.01 defines a compact notation for executable intent cells. It separates model routing, execution limits, workflow commands, and data references into four leading symbol families.

>   routing, provider, model, orchestration
<   execution constraints
@   execution commands and flow
%   data references

The release replaces the older double-brace reference style with native ICC references:

%output.c2
%var.c2.pnl
%file.c3:strategy_code.py
%input

Cell Structure

An ICC cell has a header zone and a prompt body. The header is the leading block of service lines beginning with >, <, or @. The prompt body starts at the first ordinary text line or blank boundary.

> claude.max
< tokens <= 50000
@file strategy_code.py
@text <300

Write strategy code for the accepted hypothesis.

Input:
%output.c2

Routing

> openai
> claude.max
> openai:gpt-5.5
> auto
> fast
> cheap
> best
> (openai + claude)
> (openai + claude).best
> (xai.max + claude.max).synthesis

Constraints

< cost <= $3.33
< latency <= 3m
< tokens <= 50000
< iterations <= 5
< $3.33
< 50000tok

Constraints are hard caps, not business logic. Branching over business results belongs in @if.

Execution Commands

@file
@file .md
@file strategy_code.py
@text
@text <300
@forward c2
@forward! c2
@chain c1 > c2 > c3
@if pnl > 0 -> c3
@else -> c4

References

%input
%output.c2
%var.c2.pnl
%file.c3
%file.c3:strategy_code.py
%files.c3
%prompt.c1
%header.c1
%meta.c1.cost
%error.c2

Use %% to write a literal percent reference without interpreting it.

Legacy Migration

Deprecated forms are detected and should be migrated:

{{c2.output}}                    -> %output.c2
{{c2.vars.pnl}}                  -> %var.c2.pnl
{{c3.artifact.strategy_code.py}} -> %file.c3:strategy_code.py
@input                           -> %input

Minimal Workflow

c1 Generate candidate

> (openai + claude).best
< latency <= 3m
@forward! c2

Find a realistic HFT hypothesis for Binance USDM.

c2 Evaluate PnL

> openai.max
@if pnl > 0 -> c3
@else -> c4

Analyze the hypothesis from %input.
Return:

pnl = <number>
reason = <short explanation>

c3 Write strategy code

> claude.max
< tokens <= 50000
@file strategy_code.py
@text <300

Write strategy code for the accepted hypothesis.

Input:
%output.c2