Readable code

Startr.Team code narrates. Two conventions carry that:

this, not self

Methods take this as their receiver:

def create_run_dir(this, initial_prompt: str):
    this.run_dir = os.path.join(this.base_dir, ...)

Python permits any name for the first parameter; self is only convention (PEP 8). We choose this deliberately:

• Grammar. self is a reflexive pronoun — English never uses it as a possessor ("self's vault" is not a sentence). this is a demonstrative: this.vault reads as "this agent's vault", and this.llm.plan_steps(query) parses as an English clause.
• Model priors. Language models have seen orders of magnitude more this (JavaScript, TypeScript, Java, C#, C++, PHP) than self. This framework's code crosses into LLM context constantly — vault transcripts, prompts, spec pages — so the token models already bind to "the current entity" wins.
• Lineage. It sits beside the HyperTalk-style helpers below; self would be the inconsistency.

Tooling: ruff is configured with N805 ignored. One line in ruff.toml, documented in the reference repo's README.

Narrating helpers

Steps of a procedure are named as actions with articles, so call sites read as instructions to a person:

result = run_the_tool(tool_to_run, tool_name, optimized_args, this)
save_the_artifact(this, current_step, step_num, tool_name, result, counter)
record_the_outcome(state.run_history, output)

The test is simple: read the loop body aloud. If it does not make sense as spoken English, rename until it does.

What this is not

Not a license for cuteness. Names still say exactly what happens; the narration is a constraint (plain, accurate English), not decoration.