mneme-code 3.5.0

Deterministic Python code-failure memory layer for mneme: redact-before-store, provenance-labelled FailureMemory, vault-ready markdown output.

mneme-code

Deterministic Python code-failure memory for mneme. Parses CPython tracebacks into structured failures, renders them as vault-ground-truth markdown memories, and resolves stack frames to mneme-graph function nodes.

Part of the mneme memory engine.

What it does (v1)

• Traceback parsing. parse_traceback turns a standard CPython traceback into a ParsedTraceback (exception type, message, frames). It never raises; unrecognised input returns None.
• Failure memories. failure_from_traceback + failure_to_markdown produce a markdown note the user owns, with provenance (content_hash, observed_at, trust) and a confidence label.
• Frame resolution. resolve_frames pairs each frame with a mneme-graph function node when one matches, with a clean fallback to None.

Redaction (mneme_core.privacy.redact) is applied to every user-derived string (exception message, code context, file path) before it is stored or rendered. No LLM, no network; library functions are pure (the caller injects timestamps).

Scope and deferrals (v1)

The following are not implemented yet and are documented here for honesty:

• Live test-runner integration.
• Branch-aware failure tracking.
• AGENTS.md procedural-memory parsing and repo runbook generation.
• Fix-trajectory wiring: a fix is modelled as a mneme-temporal claim that supersedes the failure memory. That lifecycle lives in mneme-temporal and is reused, not reinvented here.
• Non-Python tracebacks.