atif-sql

atif-sql reads Claude Code session transcripts and Codex CLI rollouts, converts each session to ATIF (Harbor’s Agent Trajectory Interchange Format), materializes the results as a corpus of ATIF documents on disk, and layers DuckDB views and macros over that corpus. Converting once at the boundary, with an explicit and tested fidelity policy for what the upstream adapter drops, replaces re-deriving trajectory semantics inside every SQL view.

Pick the agent with --agent claude-code|codex on convert, materialize, status and query. One corpus holds one agent, the two default to separate roots, and sessions.agent names which one a row came from.

It is a uv Python workspace, Apache-2.0, and it replaces claude-sql.

The conversion pipeline: a Claude Code session JSONL or a Codex CLI rollout JSONL is converted once to an ATIF corpus on disk, and every DuckDB view, analytics pipeline and embedding run reads that corpus.

The packages

Package	What it owns
atif-converter	The Claude Code and Codex converters, ported from Harbor’s adapters onto its vendored ATIF models, and the per-agent fidelity policy that accounts for every known upstream conversion gap per session
atif-corpus	Corpus materialization: source discovery, watermarks, quiescence, atomic artifact writes
atif-duck	The DuckDB views and macros over the materialized corpus, declared in a drift-tested static catalog
atif-models	The model alias registry and the structured-output LLM clients for GPT-6 and Claude 5.5. No other package names a Bedrock LLM id (the embedding model ids are atif-embed settings)
atif-analytics	The LLM pipelines: classify, conflicts, friction, perceived
atif-embed	Cohere Embed v4 on Bedrock or EmbeddingGemma 2 on this machine, a Lance vector store written through DuckDB’s lance extension, and the embedding backfill
atif-cli	The cyclopts CLI that composes the rest

Five of the packages carry an import-linter layers contract, and the independence contract is enforced the same way rather than by convention: converter, corpus, duck, models and embed may never import one another, atif-analytics imports only atif-models, and atif-cli is the only composition root.

The discovery loop

atif-sql schema     # every view with its columns, and every macro signature, from the static catalog
atif-sql examples   # runnable example queries, derived from that catalog and executed by the tests
atif-sql query 'SELECT * FROM tool_rank(30) LIMIT 10'

atif-sql examples derives its output from the catalog rather than carrying hardcoded strings, and every example is executed against a fixture corpus by atif-duck’s own test suite, so the listing cannot drift from the schema it documents.

For an agent. Start at For agents, which names which surface to fetch for which question and what not to assume about this repository. Prefer atif-sql schema and atif-sql examples to anything on this site: both are generated from the catalog, so they are right and a page can be stale.