fix(imessage): the SMS pre-route branch skipped the outbound audit log

[dshakes]

Found by the first real end-to-end test after the mute was lifted.

A contact texted "Hi", the bot replied "hey what's up" — the path works. But the audit log added in #224 to make reply quality measurable had no record of it.

Cause

The reply went out via the SMS pre-route branch (the contact is on RCS, so iMessage is skipped in favour of Text Message Forwarding). That branch returns at its own return { ok: true }, before the audit line at the end of send().

Every RCS/SMS-only contact's replies were invisible to the audit.

Why this matters more than a missing log line

The whole point of the audit was that a week of traffic left only ~33 verbatim samples. The contacts most likely to be missing from the corpus are exactly the ones on non-iMessage handsets — a silent, systematic sampling bias in the data meant to answer "does the bot sound like me?"

Measuring voice against a corpus that structurally excludes a class of contacts would have produced confident, wrong conclusions.

Fix

Both early returns now audit, same field shape as the main site plus via: "sms-preroute" so the paths stay distinguishable.

The test asserts every send path reaches an audit line — an invariant, not a value. That's the bug class: a new return in send() silently drops a contact out of the corpus.

57/57 imessage-bridge, 1498/1498 bridge-core, typecheck clean.

[github-actions]

🔎 Codex cross-audit (agent:audit)

No Blocking findings.

I did not find correctness, security, or missed-edge regressions in the PR diff. The runtime change only adds the missing "outbound sent" audit log on the successful SMS pre-route early return, using the same bounded textPreview pattern as the normal send path.

Verification note: I attempted the targeted Vitest file, but dependencies are not installed in this workspace:

sh: 1: vitest: not found

[github-actions]

🔷 Gemini cross-audit (agent:audit-gemini)

Audit Report: unified diff in pr-diff.patch

🔴 Blocking Issues

• None found. The proposed changes are extremely clean, highly targeted, and correct.

---

🟢 Correctness & Edge Cases

1. Resolved Gap: The patch correctly identifies and plugs a critical logging gap. When replies are sent to RCS/SMS-only contacts via the native SMS service (Text Message Forwarding), they return early (return { ok: true }) before reaching the bottom of the send() method. By adding the audit block in this branch, all successful SMS pre-route sends are now captured.
2. Execution Safety: All properties and parameters in the added logging call are safe to invoke:
   • text is typed as a string on the method signature, ensuring text.slice(0, 160) and text.length are safe and null-tolerant.
   • isBotSelfMessage(text) is correctly imported/in-scope and operates safely.
3. No Duplicate Logging: Since the pre-route branch terminates immediately after with return { ok: true }, there is no risk of dual-logging the same send event.

---

🔒 Security & Privacy

• PII Protection: The log statement conforms with standard privacy practices in the repository by capping the verbatim preview at 160 characters (text.slice(0, 160)), ensuring full PII is not leaked into the persistence layer.
• Traceability: Including { via: "sms-preroute" } as a property provides valuable tracing context while keeping the core message "outbound sent" consistent and easily searchable across all pathways.

---

🧪 Test Suite Evaluation

• Assertion Validity: The new tests in silent-drop-audit.test.ts use source introspection (readFileSync and regex/indexOf slicing) to statically assert the logic boundaries in session.ts. This ensures no future developer can introduce a regression or bypass auditing.
• Test Bounds: Slicing the file backwards by 4000 characters is safe given the 11,000+ line size of session.ts, and correctly scopes the entire send() function to prevent false positives from neighboring methods.