🧭 Language Fundamentals
The grammar behind the 100 commands: syntax, rules, chains, and presets—and the family of truth and verification that separates fact from guesswork before any decision.
By the end of the track, you’ll write compound commands with parameters and chain protocols with /chain and audits any claim against the source of truth.
The language’s promise: turn vague intent into execution with a protocol, a record, and declared confidence.
Course roadmap
Detailed content
⌨️ The /inema language
From the first slash to the full runtime: how a command declares work, receives parameters, and combines into chains.
A short DSL for controlling how models and agents research, verify, decide, and execute. Each command carries a work protocol.
Without a language, you reuse scattered prompts. With one, the same request produces standardized, verifiable output.
DSL · declarative command · protocol · reproducibility.
Slash + short name, with optional parameters in key=value format: /rank criteria=impact,cost,risk.
Syntax is the only mechanism you need to memorize—everything else is command vocabulary.
prefix / · key=value parameter · space-separated composition.
The official rules: commands are declarative instructions; order matters in an explicit chain; in a conflict, the more specific rule takes precedence, then the more recent one.
It’s what prevents unpredictable behavior when two requests collide.
declarativeness · specific > recent precedence · runtime records everything.
Several commands on the same line form a bundle of instructions executed together.
The composition covers 80% of daily use without needing an explicit /chain.
instruction package · synergy between families · parsimony.
/chain(/research -> /truth -> /rank -> /decide) sets the order and passes the relevant output from one command to the next.
Chains produce repeatable workflows—the foundation for the presets in module 6.2.
explicit order · output handoff · pipeline.
/mode activates a persistent behavior profile (conservative, exploratory, executive). /preset expands a short name into a predefined chain.
They’re the highest-level controls: they change the “how” of everything that follows.
global profile · macro expansion · consistency across sessions.
Every command can have acceptance criteria and a fallback; the runtime logs inputs, outputs, tools used, confidence, and failures.
It’s what makes the language auditable — you know what ran, how confident it was, and what to do when it fails.
operational log · stated confidence · safe fallback.
🔍 Truth and verification
The 10 commands in family 01: from /truth to /freshness — the antidote to hallucinations and false certainty.
Separates verified facts, inferences, hypotheses, and unknowns; reports confidence levels.
It’s the language’s most commonly used command — the basic hygiene for any response.
fact vs. inference · verified assumption · stated confidence.
/verify looks for enough evidence to confirm, correct, or reject a specific claim; /evidence shows which proof supports each conclusion.
Turn "trust me" into "check for yourself".
target claim · evidentiary basis · data gap.
Organizes sources by authority, recency, proximity to the facts, and independence.
Three independent sources are worth more than ten citing the same origin.
source hierarchy · dependency between citations.
Breaks content into verifiable claims and classifies each as confirmed, inaccurate, false, inconclusive, or opinion.
It’s the standard audit before publishing or making a decision based on a text.
verifiable claim · verdict taxonomy.
Assigns a confidence level to conclusions and explains which factors would increase or decrease that confidence.
Good decisions use calibrated confidence, not a confident tone.
calibration · conclusion sensitivity.
/uncertainty surfaces missing data and plausible interpretations; /assumptions lists premises marked as verified, reasonable, or weak.
An unstated weak assumption is the #1 cause of polished but incorrect analysis.
uncertainty inventory · assumption status.
Show where each critical piece of information came from: primary data, secondary data, inference, or user-provided.
Distinguishes “I measured it” from “someone told me”—and that changes how much weight the evidence carries.
chain of custody for information.
/freshness identifies what may have changed and requires an update before using time-sensitive data. It closes the module with the complete audit workflow.
Presenting an outdated statistic as current is the easiest mistake to make—and to catch with the right command.
time validity · update trigger · layered auditing.