agent-transcripts documents a draft, human-readable format for describing
agent interactions as versioned Markdown. The same document can show a coherent
user story and carry hidden, machine-oriented expectations without pretending
that the illustrated dialogue is the only valid execution path.
This repository has a new, unrelated Git history. The former research project
is preserved at
uthunderbird/agent-transcripts-archive.
If you cloned agent-transcripts before 2026-09-03, make a fresh clone; do not
push an old clone to this repository.
The format is an authored-format draft, not an implemented standard. Today this repository provides documentation, examples, and an RFC process. It does not provide a validator, compiler, runner, eval framework, or ACP integration, and it makes no interoperability or conformance claim.
The core model is intended to be product-neutral. The current examples come from Chiplog and should be read as a motivating Chiplog profile, not as generic vocabulary required by the core format.
SPEC.md— current format draft and semantic model.examples/— Chiplog-authored examples.rfcs/README.md— how proposed format changes are made.rfcs/0000-template.md— RFC template.
The initial release is documentation-only and uses format identifier
draft-0:
- define the authored transcript and its relationship to observed results;
- distinguish examples, design notes, scenarios, expectations, and forbids;
- evolve normative decisions through reviewable RFCs;
- keep product-specific concepts in profiles and examples.
Possible later layers, in order, are a structural validator, a compiler to a canonical scenario bundle, and an independent eval framework. ACP may become an adapter boundary for that future framework. None of those layers is part of the current repository contract.
Normative changes to SPEC.md require an accepted RFC. Editorial corrections
that do not change meaning may be made directly, but the pull request must say
why the change is non-normative. See rfcs/README.md.