Repository navigation
research(claude): evaluate a privacy-preserving local receipt probe #73
Description
Current roadmap — 2026-10-13 KST
- Target milestone: 0.6.0 — Authenticated receipts and Claude research.
- Planning priority: P2 (
priority:P2). - Dependencies/order: design(handoff): adopt the Python #181 delivery state model and JSON contract #69 receipt semantics; evaluate feat(handoff): add explicit receipt producer and authenticated collector bootstrap #132 explicit receipts as an alternative. This research is not a prerequisite for feat(handoff): add explicit receipt producer and authenticated collector bootstrap #132.
- Remaining scope: Claude privacy go/no-go only. A documented no-go is acceptable; body-reading implementation is not promised.
This section supersedes older milestone/version/priority statements below. Existing acceptance criteria, evidence limits and historical records remain applicable. Versioned milestones are provisional planning targets, not committed dates or implementation/live-test/merge/release authorization. P1 means a milestone prerequisite/current-release item; P2 a planned extension/research track; P3 later dependent work.
The Claude inbox socket accepts one JSON line and defines no response, ACK or status query; the receiver's Delivered/Held/Refused decision is not observable by a raw-socket sender, and notify_when_idle only covers a local main conversation. The receiver's local transcript does record peer-originated messages, but that format is undocumented and internal and requires same-user access to the full conversation history (Python #181, study).
Default remains unsupported. This issue is research only.
Scope and completion criteria
- Evaluate whether a local, opt-in probe can report only a correlation-token match and peer-origin markers, never message text, without depending on unstable internal fields.
- Assess privacy, same-user access, format drift and failure modes; decide go/no-go and record the decision.
- Any implementation needs a separate issue and explicit authorization for live tests.
Activity
abruption commentedon Nov 2, 2026
Independent privacy investigation — 2026-11-03
Recommendation: no-go for a passive private-transcript receipt probe under the current privacy and evidence constraints. Keep the route unsupported. The voluntary, explicitly delegated receipt design in #132 remains a separate path.
Evidence baseline: TypeScript source
920317896fa838e2edc2d48f0f23ea126801a72b.Reconsider only if an independently qualified interface exposes correlation separately from body text, documented peer-origin/status semantics and a validated original session/receipt-context binding. Not finding such a contract in the reviewed pages does not prove that no private or future API exists.
Independent tool run: Antigravity CLI 1.4.12, explicitly selected
gemini-4-argon-medium, fresh noninteractive plan+sandbox, 336.125 seconds, success, one attempt, no model fallback. The runner reported the selected model; separate provider canonical model metadata was not exposed. Input was seven sanitized source/contract files plus issue #73 (eight evidence files) and the research prompt. Gemini's web tool returned search summaries, so Codex separately reopened the official primary documents before accepting external claims.No user transcript/profile/authentication material, live inbox, native/model turn, delivery or ACK test was accessed or performed. No runtime change or support expansion follows from this report. Explicit evidence files were kept in the task's private external-volume directory. The runner internally created and auto-deleted a static JSON schema tempfile in the default macOS temp location because TMPDIR was omitted; no secret was included. This path deviation is recorded rather than treated as compliant temp placement.
The issue remains open with this research result recorded; no implementation or live-test claim is made.
abruption commentedon Nov 3, 2026
Decision: NO-GO — 2026-11-03
The owner approved the NO-GO disposition after the independent
gemini-4-argon-mediuminvestigation and primary-source cross-check recorded above. Close this research issue without implementing a transcript-derived probe.The current privacy/evidence constraints are not met by reading private conversation bodies and returning only a correlation flag. The reviewed official hooks/SDK surfaces do not establish an authenticated receipt interface for arbitrary existing Claude inbox sessions. This is a decision about the reviewed route, not a claim that no future API can exist.
The explicit, voluntary receipt producer/collector in #132 remains the supported development direction and is independent of this closed probe. Claude native generation remains unproven; any delegated receipt-context extension must preserve its separate versioned meaning. Reopen/re-scope only with a documented body-independent interface and separately qualified binding.
No runtime probe, live session, model turn, delivery/ACK experiment or publication is added by this closure.