Fix duplicate Discord DMs + phone transcript dedup by liususan091219 · Pull Request #43 · sonichi/sutando

liususan091219 · 2026-03-30T13:58:49Z

Summary

Discord bridge: deduplicate messages using message.id — gateway replays on reconnect were creating duplicate task files and sending duplicate DM replies
Phone server: fix duplicate transcript lines across Gemini reconnects (track processed index + dedup against last entry)
Phone server: cache task results per call session so repeated requests get instant replay instead of new round-trip

Test plan

Send several Discord DMs in quick succession — should get 1 reply each, not duplicates
Make a phone call, trigger a Gemini reconnect — transcript should not have doubled lines
On a phone call, ask the same question twice — second answer should be instant (cached)

🤖 Generated with Claude Code

On Gemini reconnect, conversationContext.items could reset or re-populate history, causing turn.end to re-process already-transcribed items. Now tracks a lastProcessedIdx that resets on context shrink, plus deduplicates against the last recorded transcript entry. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

If the caller repeats a request (e.g. Gemini reconnected and they ask again), return the cached result instantly instead of creating a new task and making them wait. Cache is per-call session. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Discord gateway can replay events on reconnect, creating duplicate task files and duplicate DM replies. Now tracks seen message IDs to skip already-processed messages. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

sonichi

LGTM — task dedup via cache for phone calls, message ID dedup for Discord DMs, transcript dedup on reconnect. All needed fixes.

The same PR #43 commit also dropped {attachment_note} from the task text, so Discord messages with file attachments lost the attachment reference. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Fix watcher reliability: replace fswatch with persistent polling fswatch -1 exits after one event and dies silently, causing tasks to pile up for hours. Replace with a 2s polling loop that never exits — catches tasks within 2 seconds regardless of fswatch state. Fixes 8-minute phone task delays and missed Discord messages. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Fix Discord bridge NameError: msg_text → text The bridge received DMs but silently crashed when writing task files because line 273 referenced `msg_text` (undefined) instead of `text`. Every Discord DM task was lost with no error logged. Also improved health check to detect stale bridge logs (>5min). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Restore attachment_note in Discord task file The same PR #43 commit also dropped {attachment_note} from the task text, so Discord messages with file attachments lost the attachment reference. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Revert watcher to fswatch + add mkdir -p safety The polling watcher broke task-notifications (never exits = never notifies). Restore main's fswatch version which exits on detection. Add mkdir -p for tasks/ and results/calls/ at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Add tests/ to .gitignore for local test creation Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Remove health check change — covered by PR #71 PR #71 has a more comprehensive health check fix (heartbeat files, multi-process detection, stale log check). Remove overlap from this PR. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ion) (#489) CodeQL alert #43 (js/resource-exhaustion, warning): the `ttl_ms` query parameter on /mute-state?state=seeing flowed directly into setTimeout's duration. Existing guard only checked `isFinite && > 0`, not an upper bound — an unbounded ttl_ms would schedule a long-lived timer per request (cheap-to-allocate DoS). Fix: clamp with Math.min(ttl, 60000). The "seeing" overlay is UI flash feedback (1.5-3s typical from `screen-capture-server.py`); nothing legitimate needs a minute-plus timer. Default still 3000ms. Closes the CodeQL queue: 15 alerts across 4 PRs (#485, #486, #487, #488, this PR #489). Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…-logging) (#493) ## #43 js/resource-exhaustion (src/web-client.ts:2549) #489's `Math.min(ttl, MAX_TTL_MS)` is an `isNumericFlowStep` (numeric passthrough) in `codeql/javascript-all@2.6.27` — it narrows the value but doesn't register as a sanitizer. The `js/resource-exhaustion` config only blocks taint via `UpperBoundsCheckSanitizerGuard`, which matches `RelationalComparison` AST nodes (`<`, `<=`, `>`, `>=`). Rewrote the clamp as `ttl <= MAX_TTL_MS ? ttl : MAX_TTL_MS` — semantically identical, but CodeQL now sees `ttl` as the lesser operand in a `<=` comparison and blocks taint in the true branch. ## #44 js/clear-text-logging (src/voice-agent.ts:71) The `${value.length}` and `${prefixOk}` interpolations in `console.error` both read properties of `value`. The env var `GEMINI_API_KEY` triggers `HeuristicNames` → tainted source; `allowImplicitRead` in the cleartext config pulls any property of a tainted value into the logger sink. #486 stopped echoing substrings but the length + ternary-prefix diagnostic still flow the taint. Dropped those two interpolations from the log; kept the name, expected format ("AIza..." 35-60 chars), and remediation URL. Diagnostic loss is minor — a malformed key always needs to be regenerated anyway. ## Verification Local CodeQL CLI 2.25.2 + codeql/javascript-queries@2.3.7: - main (dc7bfbe): 2 alerts (resource-exhaustion @ web-client.ts:2549, clear-text-logging @ voice-agent.ts:71) - this branch: 0 alerts `npx tsc --noEmit` clean. Closes #43, #44. Together with #492 (merged earlier this session), this zeroes the CodeQL security-code-scanning queue. Co-authored-by: Xueqing Liu <xueqingliu@Xueqings-Mac-Studio.local> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Phone server: fix duplicate transcript messages across reconnects On Gemini reconnect, conversationContext.items could reset or re-populate history, causing turn.end to re-process already-transcribed items. Now tracks a lastProcessedIdx that resets on context shrink, plus deduplicates against the last recorded transcript entry. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Phone server: cache task results to avoid duplicate work If the caller repeats a request (e.g. Gemini reconnected and they ask again), return the cached result instantly instead of creating a new task and making them wait. Cache is per-call session. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Discord bridge: deduplicate messages using message.id Discord gateway can replay events on reconnect, creating duplicate task files and duplicate DM replies. Now tracks seen message IDs to skip already-processed messages. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ichi#72) * Fix watcher reliability: replace fswatch with persistent polling fswatch -1 exits after one event and dies silently, causing tasks to pile up for hours. Replace with a 2s polling loop that never exits — catches tasks within 2 seconds regardless of fswatch state. Fixes 8-minute phone task delays and missed Discord messages. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Fix Discord bridge NameError: msg_text → text The bridge received DMs but silently crashed when writing task files because line 273 referenced `msg_text` (undefined) instead of `text`. Every Discord DM task was lost with no error logged. Also improved health check to detect stale bridge logs (>5min). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Restore attachment_note in Discord task file The same PR sonichi#43 commit also dropped {attachment_note} from the task text, so Discord messages with file attachments lost the attachment reference. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Revert watcher to fswatch + add mkdir -p safety The polling watcher broke task-notifications (never exits = never notifies). Restore main's fswatch version which exits on detection. Add mkdir -p for tasks/ and results/calls/ at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Add tests/ to .gitignore for local test creation Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Remove health check change — covered by PR sonichi#71 PR sonichi#71 has a more comprehensive health check fix (heartbeat files, multi-process detection, stale log check). Remove overlap from this PR. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…exhaustion) (sonichi#489) CodeQL alert sonichi#43 (js/resource-exhaustion, warning): the `ttl_ms` query parameter on /mute-state?state=seeing flowed directly into setTimeout's duration. Existing guard only checked `isFinite && > 0`, not an upper bound — an unbounded ttl_ms would schedule a long-lived timer per request (cheap-to-allocate DoS). Fix: clamp with Math.min(ttl, 60000). The "seeing" overlay is UI flash feedback (1.5-3s typical from `screen-capture-server.py`); nothing legitimate needs a minute-plus timer. Default still 3000ms. Closes the CodeQL queue: 15 alerts across 4 PRs (sonichi#485, sonichi#486, sonichi#487, sonichi#488, this PR sonichi#489). Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… (clear-text-logging) (sonichi#493) ## sonichi#43 js/resource-exhaustion (src/web-client.ts:2549) sonichi#489's `Math.min(ttl, MAX_TTL_MS)` is an `isNumericFlowStep` (numeric passthrough) in `codeql/javascript-all@2.6.27` — it narrows the value but doesn't register as a sanitizer. The `js/resource-exhaustion` config only blocks taint via `UpperBoundsCheckSanitizerGuard`, which matches `RelationalComparison` AST nodes (`<`, `<=`, `>`, `>=`). Rewrote the clamp as `ttl <= MAX_TTL_MS ? ttl : MAX_TTL_MS` — semantically identical, but CodeQL now sees `ttl` as the lesser operand in a `<=` comparison and blocks taint in the true branch. ## sonichi#44 js/clear-text-logging (src/voice-agent.ts:71) The `${value.length}` and `${prefixOk}` interpolations in `console.error` both read properties of `value`. The env var `GEMINI_API_KEY` triggers `HeuristicNames` → tainted source; `allowImplicitRead` in the cleartext config pulls any property of a tainted value into the logger sink. sonichi#486 stopped echoing substrings but the length + ternary-prefix diagnostic still flow the taint. Dropped those two interpolations from the log; kept the name, expected format ("AIza..." 35-60 chars), and remediation URL. Diagnostic loss is minor — a malformed key always needs to be regenerated anyway. ## Verification Local CodeQL CLI 2.25.2 + codeql/javascript-queries@2.3.7: - main (dc7bfbe): 2 alerts (resource-exhaustion @ web-client.ts:2549, clear-text-logging @ voice-agent.ts:71) - this branch: 0 alerts `npx tsc --noEmit` clean. Closes sonichi#43, sonichi#44. Together with sonichi#492 (merged earlier this session), this zeroes the CodeQL security-code-scanning queue. Co-authored-by: Xueqing Liu <xueqingliu@Xueqings-Mac-Studio.local> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

liususan091219 and others added 3 commits March 30, 2026 09:51

sonichi approved these changes Mar 30, 2026

View reviewed changes

sonichi merged commit 94e7cd1 into main Mar 30, 2026

sonichi deleted the fix/dedup-discord-and-phone branch March 30, 2026 23:02

liususan091219 mentioned this pull request Apr 1, 2026

Health check doesn't detect stuck Discord bridge #70

Closed

liususan091219 mentioned this pull request Apr 1, 2026

Fix Discord bridge NameError + watcher safety + gitignore tests/ #72

Merged

2 of 3 tasks

This was referenced Apr 20, 2026

fix(agent-api): break CodeQL taint via Path.name (path-injection #7, #16-23) #488

Merged

fix(web-client): clamp seeing-TTL to 60s (CodeQL #43 resource-exhaustion) #489

Merged

liususan091219 mentioned this pull request Apr 21, 2026

fix(codeql): close js alerts #43 (resource-exhaustion) + #44 (clear-text-logging) #493

Merged

3 of 4 tasks

Navigation Menu

Repository navigation

Fix duplicate Discord DMs + phone transcript dedup - #43

Fix duplicate Discord DMs + phone transcript dedup#43
sonichi merged 3 commits into
mainfrom
fix/dedup-discord-and-phone

liususan091219 commented Mar 30, 2026

Uh oh!

sonichi left a comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

liususan091219 commented Mar 30, 2026

Summary

Test plan

Uh oh!

sonichi left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants