|
A synthetic fixture only ever holds the shapes someone thought of, and
the shapes nobody thought of are the ones that leak. Sweeping the user's
own tag:spam is the only source of those, and it earned its place: the
first message found a header the parser did not read, and the corpus
exercised a spoofed Reply-To display name no fixture had.
AGENTS.md records it under Testing, beside the fixture rule it is the
deliberate exception to: ask first, work from the scratchpad, compare
every address in the raw source against the whole ioc output, and let
only counts and stripped domains out. A finding becomes a synthetic
fixture; the real message never enters the repository.
BACKLOG.md opens with the one thing the sweep found that was not worth
fixing blind: boilerplate namespace urls (w3.org doctypes) reported as
indicators. Noise rather than a defect, and the argument for leaving it
alone is real, so it is written down with both sides rather than built.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019NHaqA1Rz5ybed7wFUeQbK
|
|
Same arrangement as qtmaildir: one source of truth, and a thin pointer
beside it so every agent tool reads the same file.
It leads with the three properties that are not negotiable, because each
has a concrete victim and each is a thing a later change could quietly
break. Recipient identifiers must never reach a report, and the guarantee
is structural rather than a step someone remembers. Nothing is fetched or
resolved, which the suite verifies by running with sockets disabled. The
trust boundary is configured rather than guessed, and its test is
mutation-checked because walking one hop too far reports an innocent party
the attacker named.
It also records the traps that were found the hard way rather than
reasoned about: ipaddress.ip_network(42) returning 0.0.0.42/32 instead of
raising, a string trusted_relays iterating characters, goog.json looking
like the right SPF source while listing all of Google, and the provider
table having been wrong in every entry when it was first written from
memory.
The hand-testing rule is stated with its evidence: the prompts are the
user's to test, and one pass over them found five defects, four being the
same mistake of validating an answer somewhere other than where it was
given.
Every factual claim in it was checked against the code before committing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KphFXTc2QajxXsHWyvGJ4R
|
