Agent Session Bridge v0.4.0

Agent Session Bridge v0.4.0

This backward-compatible release makes the merged Claude Code evidence work available on PyPI.

Highlights

• agent-session explain renders imported ATIF trajectories as a deterministic Markdown account of steps, tool calls, results, provenance, and conversion limits.
• Claude Code imports provide finer fidelity evidence for unsupported bookkeeping, potentially content-bearing and unrecognized records, ignored fields, and duplicate usage records.
• The project adds a controlled real Claude Code fixture with synthetic data, pseudonymized identifiers, explicit provenance, and regression coverage.
• A private local-only sanitizer and review procedure supports preparing controlled real-session candidates without publishing raw logs.

Correctness and boundaries

• Observed model names and token usage are carried to compatible ATIF step fields. Cost is unavailable and is not inferred.
• The fixture preserves its source SHA-256 on Windows checkouts and verifies tool-result correlation.
• It covers one controlled Windows session from Claude Code 2.1.266 using Read, Bash, Edit, and Write. It does not establish general compatibility across versions, operating systems, session shapes, or tools.

Verification

• Built both wheel and source distribution; both passed twine check.
• Installed the wheel in a fresh environment and confirmed version 0.4.0 and agent-session explain --help.
• Ruff, Mypy, all 213 pytest tests, and the semantic interoperability corpus passed locally.
• Protected pull request checks and post-merge CI and CodeQL passed on the tagged release commit.

Full details are in CHANGELOG.md.