| name | parse-thread |
| description | Parse .eml files, a directory of them, or an .mbox into normalised thread JSON — headers, participants, bodies, attachment manifest, and reconstructed reply chains. Use this first; every other eml-toolkit skill consumes its output. Triggers on "parse these emails", "what's in this .eml", "reconstruct this email thread", "read this mbox". |
Parse an email export into thread JSON
Foundation skill. Produces the JSON contract that thread-to-pdf,
thread-to-markdown, redact-thread and bundle-evidence all read.
Run it
cd ${CLAUDE_PLUGIN_ROOT} && source .venv/bin/activate
PYTHONPATH=scripts python3 -m eml_toolkit.cli parse <source> --out thread.json
<source> is a single .eml, a directory searched recursively for *.eml, or
an .mbox/.mbx.
Add --strip-bodies when you only need to see the structure — it replaces
bodies with a boolean and a 400-char excerpt, which keeps a 200-message export
readable in context.
What it does
- Decodes RFC 2047 headers, so Hebrew display names come out as text, not
=?UTF-8?B?….
- Extracts both
text/html and text/plain alternatives; downstream renderers
prefer HTML and fall back to text.
- SHA-256s each source file and each attachment payload.
- Groups messages into threads by
References/In-Reply-To, then falls back to
normalised subject for messages that carry neither. The subject normaliser
strips Re:/Fwd:/AW:/SV: and the Hebrew תגובה:/הועבר: and Arabic رد:.
- Sorts each thread chronologically; messages with an unparseable
Date sort
last rather than being dropped.
Threads come back largest-first, so --thread 0 in the other skills is usually
the one you want.
Reading the result
Report to the user: how many threads, how many messages in each, the date span,
and the participant list. If one source file produced its own singleton thread,
say so — that usually means its headers were stripped by whatever exported it,
and the subject fallback did not match either.
