mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-07 16:11:46 +02:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - People also ask agents for work in their existing chat tools. > - Each external conversation needs one task and a current authorized source. > - Retries, Stop, and provider failures must not duplicate work or expose private data. > - The first chat PR establishes the opt-in provider and data contracts. > - This PR adds experimental channel integration and its durable control plane. > - Users can request work from connected channels and inspect delivery in Paperclip. ## Linked Issues or Issue Description Refs #13100 and #13092. This is the second of exactly two chat PRs. Foundation #13100 is merged and changed 143 files. Runner prerequisite #13092 is also merged. This PR changes 400 files against master, below the 500-file review limit. It contains no wireframe images or HTML galleries. ## What Changed - Add native Slack, GitHub, Microsoft Teams, Telegram, and Discord chat connections. Keep chat disabled unless the operator enables experimental chat connectors. Preserve the production GitHub tool connection and its normal setup path. - Bind each provider bot identity to one immutable Paperclip agent. Bind each admitted external conversation to one task. Paperclip owns tasks, runs, permissions, and audit records. - Add durable admission, per-conversation queues, questions, task controls, progress, final replies, images, files, and delivery receipts. Board comments remain internal unless explicitly sent to the channel. - Check current identity, provider reach, resource access, credentials, runtime generation, and exact source before provider effects. Keep private responses private. Never send raw reasoning, private logs, credentials, or tool arguments. - Hold uncertain sends for explicit audited resolution. Make Board Send-to-channel atomic and idempotent. Keep reconnect and setup credentials in Paperclip secret storage. - Preserve current native-runner authority across retries, lost acknowledgements, and recovery. Keep immutable input and completion contracts separate from newer user input. Receipt reconciliation cannot launch a provider. - Reconcile chat close/new ordering and provider-effect lock order. Audit resource access changes in the same transaction. Submit only the selected resource from each UI toggle so stale pages cannot undo unrelated access changes. - Drain Codex stdout before certifying process exit. Bound the drain with the existing shutdown grace. Preserve observed terminal authority without treating an undrained process as successful or reusable. - Incorporate master `018ca5da` with its ACP Stop, mobile task layout, runner packaging, and official lock changes. Preserve dedicated chat-answer continuations in both directions when ordinary queued comments are adopted after Stop. - Fence late adapter readiness behind an earlier Stop for the same run. Preserve verified cleanup for registered adapters. Handle single Stop, agent pause, duplicate Stops, and failure release without creating a false cancellation receipt. - Incorporate master's `6dd48cad4` wake-queue extraction. Preserve exact failed-chat retry authorization and lineage, retired question-source suppression, and the block on generic recovery that would discard the admitted source. Fresh deferred input retains its separate promotion path. - Incorporate master `2a05b5ed3` and its queue-admission extraction, simplified transaction ports, and separate runner CI job. Preserve exact durable receipts, actor separation, and dedicated-answer isolation through the new module. A failed receipt insert rolls back the accompanying deferred-wake merge. ## Verification Current head: `afe19299d06253cb628eb398e91d1200ea9f412a`, incorporating master `2a05b5ed3457ea33efd6895520447d1d97fe98d8`. The conflicts are resolved. This successor fixes two test-harness boundaries exposed by CI: per-case route-module preparation and actual durable-save completion before intentional runner termination. Production code and all existing test/turn deadlines are unchanged. [Exact-head Greptile review](https://github.com/paperclipai/paperclip/pull/13038#issuecomment-5587250594) is **5/5**, completed September 10 at 13:20:55 UTC, with no actionable findings or open review threads. [Fresh exact-head CI](https://github.com/paperclipai/paperclip/actions/runs/34481724341) passes **all 24 jobs**, including Build and both required aggregates. Normal exact-head guarded merge was attempted and rejected by the remaining branch approval policy: CODEOWNER review is required and no human approval is present. Normal **squash auto-merge is enabled** as of September 10 at 13:36:26 UTC. Requested CODEOWNERS have been notified; no approval bypass or self-approval was used. Earlier-head results below remain historical evidence, not qualification of this successor. - Final exact-head Linux evidence: 995/995 chat integration cases; 36/36 agent-skills routes; 35/35 runner live-session cases, including real process kill/resume; 1948 runner Vitest cases with three existing benchmark/platform guards; 870/870 API-authority cases; and 104 browser cases with four existing optional skips. Rust, conformance/replay, full repository build, typecheck, canary, all server/workspace shards, and both required aggregates pass with normal CI concurrency. Earlier failed attempts remain recorded below. - Latest test-only qualification: 141/141 route/permissions/authentication cases pass in separate cold forks, with plain server types and independent review clear. The real-runner suite passes 35/35, with plain runner types and independent review clear. A controlled premature-save acknowledgement fails as expected; matching ownership/effect/process evidence, rejected saves, real turn outcome, test abort, and pre-kill liveness are covered. No local reproduction of the original CI scheduling failure is claimed. The preceding [CI run](https://github.com/paperclipai/paperclip/actions/runs/34479680858) passes 21/24 jobs, including all 995 Linux chat cases and browser aggregate (104 passed, four existing optional skips); only Build, the skills serialized shard, and the required verification aggregate fail. Its exact-head Greptile review was 5/5. Both failed job logs are retained. - Final fixture qualification: all eight focused Discord cases and all 995 chat integration cases pass. The exact modal statement/PID is observed before taking the real connection lock; the test then proves its actual blocking relationship before mutation. Original SQL execution, provider behavior, negative assertions, and 1s/15s timeouts remain unchanged. Independent review is clear and test/production hashes remain frozen. The preceding [CI attempt](https://github.com/paperclipai/paperclip/actions/runs/34477184777) passed 22 jobs, including Build/runner, typecheck, canary, all other test shards, and browser aggregate (104 passed, four existing optional skips); the two fixture failures and failed verification aggregate remain recorded, not relabeled as a pass. - Current queue-module composition: 308/308 recovery/batching/queue/Stop tests; 995/995 full chat integration; 89/89 module tests, including real PostgreSQL receipt-insert rollback; 24/24 workflow/module-boundary tests; plain server and UI types. All four actual local process/ACP browser paths pass in 1.4 minutes. Fresh databases, no skips or retries, stable reviewed source hashes. The initial boundary failure is retained; its no-op service wrapper was removed without changing recovery context or weakening the check. An exploratory standalone test-directory typecheck fails because its new upstream transformation config is not a standalone typechecking project; standard CI/build does not invoke it, and no configuration was weakened to suppress those diagnostics. - The preceding head `e02a63d462ce5d47433b0aeb632bb6fd20aab1ba` passed [all 24 CI jobs](https://github.com/paperclipai/paperclip/actions/runs/34436462958) and exact-head Greptile review at 5/5. Required CODEOWNER review prevented its normal merge before master advanced again. - Final extracted-module composition: 307/307 recovery, batching, queue and Stop-control tests; 995/995 full chat integration; 49/49 module tests including eight PostgreSQL adapter cases; and 19/19 issue-update tests. Plain server types pass. All four actual local process/ACP browser paths pass in 1.3 minutes. Fresh databases, no skips or retries in these cohorts, frozen source hashes, and independent review clear. - The preceding head `3e4e1c1c` passes [all PR CI jobs](https://github.com/paperclipai/paperclip/actions/runs/34415826820), including Build and required `ci / verify` and `ci / e2e`. Both the original Rust failure and the previously load-sensitive lineage fixture pass with unchanged Linux concurrency. Master advanced afterward and required this reconciliation. - Final master composition: 448/448 focused UI tests, 186/186 adapter tests, 24/24 queue/control tests, and 11/11 packaging tests. Plain UI, server, shared, and adapter types pass. Token gates and diff checks pass. Independent server and UI reviews are clear. - Stop-registration regression: both real-service cases fail against exact `a95` source and pass with the fix. The full corrected recovery/control suite passes 265/265. Duplicate-owner and failed-Stop controls also pass. Plain server types pass. The readiness barrier prevents provider startup without adding an acknowledgment to an already terminal run. - Final qualification strengthens terminal-field equality and repeats both affected cases successfully on a fresh database. All four actual local process/ACP browser paths pass again in 1.3 minutes, without skips or retries. The final screenshot shows Cancelled, a paused subtree, retained input, and no error toast. - Two new actual-service regressions fail before the merge fix. They prove that queued-comment adoption could consume a dedicated chat answer or add unrelated input to that answer. The fixed four-case cohort passes, including ordinary upstream continuation and adapter Stop controls. Full recovery passes 257/257. All four actual local process/ACP Stop browser flows pass in 1.4 minutes, without skips or retries, on a fresh database. - The unchanged runner artifact was qualified with 171/171 transport tests, 870/870 API-authority tests, conformance 1/1, and replay 11/11. Six controlled reader tests prove the exit/drain repair. Its local serial Rust workspace passed 546 top-level cases plus two invoked helpers; the later passing Linux CI supplies default-concurrency evidence. - Prior exact-source full chat integration passes 995/995. Settings regressions cover concurrent stale pages, 501 destinations, pending state, rejected updates, and explicit retry. These deterministic tests do not prove live provider behavior. - Retained failed attempts and their causes are in the [qualification log](https://github.com/paperclipai/paperclip/blob/afe19299d06253cb628eb398e91d1200ea9f412a/doc/plans/chat-adapters/2026-09-08-chat-queue-and-webhook-repair.md). The first merge adapter run timed out while macOS slept for 290 seconds. Its unchanged repeat passed with a temporary sleep guard. No assertion, deadline, or CI gate was weakened. Review commands include `pnpm --filter @paperclipai/server exec vitest run src/__tests__/heartbeat-process-recovery.test.ts src/__tests__/issue-queued-comments-routes.test.ts` and `pnpm exec playwright test --config tests/e2e/playwright.config.ts tests/e2e/acp-stop-continuation.spec.ts`. Database suites require fresh disposable databases. See the [browser runbook](https://github.com/paperclipai/paperclip/blob/afe19299d06253cb628eb398e91d1200ea9f412a/doc/plans/chat-adapters/2026-09-04-chat-adapters-browser-e2e-runbook.md) for provider setup and separate live acceptance steps. ## Risks - This remains experimental. Deterministic tests and bounded live evidence do not establish every provider feature, tenant, permission layout, or media shape. Teams work-tenant qualification is still open. - Failed and uncertain provider effects remain visible and can require operator action. A transport receipt does not prove recipient visibility. - Native controller and runner artifacts must remain compatible. Preserve lease ownership, terminal authority, source binding, and quarantine during future changes. - Access and audit rows commit together, but activity notifications remain best-effort. This is not a new durable event outbox. - The PR operation does not deploy a live server, replace its runner, or change provider permissions. Remaining live qualification is documented in the [temporary handoff](https://github.com/paperclipai/paperclip/blob/afe19299d06253cb628eb398e91d1200ea9f412a/doc/plans/chat-adapters/2026-09-08-open-qualification-followups.md). ## Model Used OpenAI Codex assisted with implementation, tool execution, testing, and review. The work records `gpt-6-astra` assistance. The environment does not report a context-window size. No private reasoning traces are included. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
25 KiB
25 KiB
Platform-specific chat adapter research source
Internal synthesis source — do not publish as the reviewer-facing artifact.
Date: 2026-09-04
Paperclip base: origin/master at 8430bd897f01dd4b91e0970efffb71b97e5a2685
Chat SDK snapshot: 51322dde8f4aafd8a7fc7a20cbfd7ae45cafaa5c
This is an implementation research snapshot. The current executable contract is the browser runbook and platform-surfaces companion; upstream capabilities described here are not automatically shipped Paperclip capabilities.
Research question
What provider-specific setup, configuration, and interaction behavior must Paperclip expose for Slack, GitHub, Discord, Microsoft Teams, and Telegram while preserving the shared rule that Paperclip owns agents, issues, runs, permissions, publications, and audit history?
Claim-gap matrix
| Claim needed for the product plan | Primary evidence | Confidence | Product consequence |
|---|---|---|---|
| Slack can use verified HTTP webhooks, OAuth installations, or Socket Mode. | Chat SDK Slack adapter at the pinned revision, Slack request verification, Slack Socket Mode | High | Paperclip selects direct verified callback or its relay from instance reachability. Socket Mode is an instance-admin escape hatch, never an endpoint-wizard choice. |
| Slack events and interactive payloads must be acknowledged quickly. | Slack Events API, Slack interactivity | High | Persist first, acknowledge immediately, process asynchronously; modal-opening callbacks require a fast path because trigger IDs expire. |
| Slack gives the fullest Chat SDK interaction surface. | Pinned Slack adapter feature table, Slack Agent Sessions | High | Native threads, streaming, Block Kit, actions, modals, slash commands, files, DMs, reactions, and ephemeral responses are used automatically whenever the installation, conversation, and Paperclip authorization permit them. |
| A GitHub App is the least-privilege production credential. | GitHub App registration, choosing GitHub App permissions, pinned GitHub adapter setup | High | Recommend a GitHub App with Issues/PR write and Metadata read, installed only on selected repositories. PAT is a development-only fallback. |
| GitHub chat does not imply repository-code tool access. | GitHub permissions are independently selectable and installation repositories are scoped in GitHub App permissions. | High | The chat connector does not request Contents permission. An agent that must read/write code receives a separate tool-purpose connection and grant. |
| GitHub's native conversation unit is the existing issue, PR conversation, or review-comment thread. | Pinned GitHub adapter thread model, issue comments API | High | A mention binds an existing GitHub object/thread to one Paperclip issue. Paperclip does not create an extra chat thread. PR-level and inline review threads remain distinct. |
| Teams setup includes app/bot registration, a reachable messaging endpoint, a customer-owned Teams app, and tenant installation policy. | Teams app registration quickstart, pinned Teams adapter setup | High | Provide the exact endpoint and manifest settings; require the customer-owned Entra/Azure Bot/Teams app path; detect disabled sideloading without promising a Paperclip-generated package. |
| Teams normally receives mentions; subscribed channel/group replies require the shipped RSC entries. | Teams all-message/RSC guidance, resource-specific permissions | High | The customer-owned Teams app includes ChannelMessage.Read.Group and ChatMessage.Read.Chat for fixed addressed-thread behavior. They are not endpoint toggles and do not grant tenant-wide Graph history/directory access. |
| Some Teams Graph operations require broader Entra permissions and admin consent. | Graph chat message permissions, pinned Teams adapter history/user lookup notes | High | User.Read.All and DM history remain optional, visibly privileged add-ons. The basic mention/reply experience must not depend on them. |
| Discord bots receive messages and interactions through a long-lived Gateway client. | Discord Gateway documentation, pinned Discord adapter | High | Use a direct outbound Gateway runtime. Do not request a webhook URL, interactions public key, slash-command setup, or endpoint delivery choice for the current feature set. |
| Discord message content and channel actions require explicit intent and effective bot permissions. | Discord Gateway intents, Discord OAuth2 | High | Verify Message Content Intent, server installation, and per-channel permissions. Generate only the bot-scope install URL with the reviewed permission integer; provider availability remains the ceiling on Paperclip access. |
| Telegram webhooks and long polling are mutually exclusive, and webhooks support a secret-token header. | Telegram Bot API getUpdates and setWebhook, pinned Telegram adapter modes |
High | Cloud/public instances use verified webhooks; local long-running development may poll; private production uses the Paperclip relay rather than an unverified webhook. |
| Telegram privacy mode changes which group replies the bot receives. | Telegram bot privacy FAQ, Telegram bot features | High | Keep privacy mode on. A group starts work with @bot; continuing turns must reply to the bot's message or mention it. Forum topics can use a stable topic boundary. |
| Telegram's Chat SDK surface is narrower and rate-sensitive. | Pinned Telegram feature/streaming table, Telegram bot FAQ | High | Default to throttled post-and-edit, use native drafts only in private chats when enabled, use inline buttons instead of modals/selects, and never promise ephemeral messages. |
| Telegram supports commands as a platform, but the pinned Chat SDK adapter does not expose the shared slash-command feature. | Telegram commands, pinned adapter feature table | High | Parse a small Paperclip command vocabulary (/new, /status, /close) as normal messages in Telegram glue until the SDK adapter exposes command events. |
Provider synthesis
Slack
- External setup: create or select a Slack app; use a generated manifest for the exact bot scopes and events; set the Paperclip webhook URL for Events, Interactivity, and optional slash commands; install to a workspace; store bot/OAuth credentials through Paperclip secret references; verify bot identity, signature, scopes, events, and workspace. Managed Add to Slack remains optional. For Enterprise Grid, installation identity may be the enterprise rather than a single team.
- Default behavior: a root
@botmention activates the endpoint. The first bot reply under that root establishes the Slack thread and the Paperclip issue binding. Human replies in the bound thread continue without another mention. A DM conversation is a stable issue boundary. - Automatic capability behavior: Agent Sessions/native streaming and stop, Block Kit actions, modals, slash commands, files, and ephemeral denials are used whenever supported and authorized. App Home/suggested prompts are installation concerns. Socket Mode is an instance-admin delivery escape hatch, not an endpoint setting or response-feature switch.
- Operational caveats: acknowledge events and actions within Slack's deadline; persist delivery before asynchronous processing; dedupe by event ID; ignore the bot's own messages; surface missing scopes and bot-not-in-channel distinctly; token rotation and OAuth reinstallation must not change the endpoint identity.
GitHub
- External setup: create a GitHub App with webhook URL/secret,
application/json, Issues read/write, Pull requests read/write, and Metadata read; subscribe to Issue comment and Pull request review comment; generate/store the App private key; install to selected repositories. GitHub Enterprise Server adds an API base URL. PAT is a visibly non-production fallback. - Credential separation: the chat-purpose app does not request Contents, Actions, Administration, or other repository tool permissions. If Maya needs code access during a Paperclip run,
/appscreates a separate GitHub tool connection with its own credential audience and agent grant. - Default behavior:
@botin an issue comment or PR conversation binds that existing issue/PR to one Paperclip issue. An inline review-comment thread has a different stable key and binds separately. Further human comments in the bound context continue the task; self-authored bot comments and redeliveries are suppressed. - Rendering: GitHub-Flavored Markdown, reactions, comment edits, and links. No native streaming, DMs, ephemeral messages, modals, or Chat SDK buttons/selects. Progress should update one bot comment at a coarse cadence; governed actions link to authenticated Paperclip pages. The current adapter treats URLs in inbound comments as ordinary text and does not ingest them as files. Outbound artifacts become Paperclip URLs because the adapter has no file-upload surface.
Microsoft Teams
- External setup: Paperclip provides the messaging endpoint and required manifest settings. The operator creates the single-tenant Entra app, Azure Bot, and customer-owned Teams app in Microsoft, then enters client ID, tenant ID, and client secret in Paperclip. Custom app upload/install must be allowed. No provisioning helper is shipped; Paperclip does not currently generate a package or install link. Paperclip verifies Entra/bot identity, endpoint reachability, tenant mode, and installation activity.
- Authentication: the portable customer-owned path uses a client ID, tenant ID, and client secret for a single-tenant registration. Federated workload identity and sovereign/GCC deployment remain future instance-level concerns, not endpoint choices.
- Default behavior: personal, team, and group-chat scopes are enabled as installed; direct mentions are the least-privilege trigger. In a Teams channel, a new post and its replies are one native thread and one Paperclip issue. A group chat or DM uses its stable conversation identity.
- Permissions: the shipped customer-owned app manifest includes the resource-specific entries required for subscribed channel/group replies. Broader Entra permissions such as
User.Read.AllandChat.Read.Allare not required by the basic connector and are not exposed as endpoint controls. - Rendering: Adaptive Cards, buttons, task-module/modal interactions, reactions, typing, and targeted-message/DM fallback. Paperclip's durable webhook pipeline sets native Teams streaming to false and uses bounded post/edit output in DMs, channels, and groups. Native file receipt/upload is personal-chat-only; channel/group files require a separate Graph connection and otherwise fall back to a safe Paperclip link without ingestion. Select menus and slash commands fall back to text or cards.
Discord
- External setup: create one customer-owned application bot per immutable Paperclip agent, enable Message Content Intent, and enter Application ID, Server ID, and bot token. Paperclip generates a server-pinned install URL with only the
botscope and the reviewed permission integer, then verifies application identity, intent, membership, and effective text-channel permissions. - Transport: a long-lived outbound Gateway client receives messages, reactions, interactions, edits, and deletes. Discord therefore needs no public Paperclip callback, interactions public key, or endpoint delivery selector. Reconnect, resume, and provider
retry_aftertiming are runtime behavior. - Default behavior: a root
@botmention creates a Discord public thread and exactly one Paperclip task; replies in that thread continue it without another mention. Direct messages use separate linear task generations. A globally unique Discord Application ID prevents one provider bot identity from representing multiple Paperclip agents. - Rendering: bounded post/edit output, embeds, supported buttons, reactions, and native files are automatic when permitted. The current connector does not advertise slash commands, modals, true ephemeral responses, or proactive DMs.
- Qualification boundary: Paperclip preflights endpoint, resource, principal, and root-message admission before provider-thread creation, so a denied root creates no Discord thread or Paperclip work. An allowed root persists a provisional receipt before the provider effect, and recovery idempotently creates or reuses the thread, including existing-thread error
160004. Those guarantees have deterministic and fresh-database evidence but still require real-provider fault-path proof.
Telegram
- External setup: create one bot per Paperclip agent with BotFather; set name and an available username; paste the bot token into Paperclip. Paperclip calls
getMe, registers suggested commands, and selects verified webhook, relay, or local-development polling from the deployment. The operator adds the bot to intended chats and keeps privacy mode on. Forum-topic creation requires additional admin rights and is optional. - Default privacy: privacy mode stays on. In a group, an
@botmessage activates a task; follow-ups must reply to a bot message or mention it. Unaddressed group traffic is invisible/ignored. A forum topic is a stable task boundary when present. - Linear conversation boundary: the first DM creates the active Paperclip issue.
/newor the New task inline button closes/shelves the current binding and starts another;/statusand/closeare parsed from normal Telegram messages. Ordinary groups without topics use the activation message/reply chain as the visible conversation, while Paperclip stores the explicit active binding. - Rendering: typing/reaction acknowledgement, throttled post-and-edit by default, native draft preview in DMs when available, MarkdownV2/rich-message fallback, inline buttons and URL buttons, files/media. No ephemeral response, modal, select, or list-threads support. Permission denials use a concise normal reply or DM.
- Operational caveats: dedupe by
update_id; configureallowed_updates; show pending webhook updates and last error; respect per-chat/group flood limits; callback payloads contain only opaque short IDs; rotate a leaked bot token through BotFather and update the secret reference.
Cross-provider decisions resulting from the research
- The shared two-decision onboarding remains intact. Provider complexity belongs in the provider handoff and post-connect settings, not in a universal wizard.
- Each setup surface follows one vertical sequence while clearly labeling what Paperclip provides and what must be completed at the provider. Paperclip can generate manifests, URLs, secrets, and copyable commands, but it cannot pretend that tenant/workspace/repository installation policy is under Paperclip control.
- Capability status is rendered from the adapter registry on Overview. Response features are not endpoint controls: the runtime uses the maximum safe supported behavior and explains the exact fallback when provider support, installation permissions, conversation type, or Paperclip authorization prevent it.
- Provider permission escalation is incremental. Basic mention/reply operation uses the smallest viable permission set; history, all-message visibility, user-directory lookup, and code/tool access are separate grants.
- Conversation boundaries are provider-native and explicit: Slack thread, GitHub object/review thread, Teams post thread or conversation, Discord public thread or DM generation, Telegram DM/group active binding or forum topic.
- Every provider-specific interaction still enters the same durable delivery → principal/permission → issue binding → Paperclip wakeup → safe publication flow.
Residual validation before implementation
- Confirm the exact Slack app manifest against the Chat SDK version finally pinned for implementation, including any optional Agent Sessions scopes/events.
- Test Teams channel reply events in mention-only mode and document precisely when subscribed replies require RSC versus ordinary bot conversation delivery.
- Test Telegram privacy-on reply delivery, forum-topic identifiers, native draft support, and effective file-download ceilings against the production Bot API version.
- Confirm whether the GitHub adapter should include Discussion comments in the first release; the pinned adapter documents issues, PR conversations, and review comments, so Discussions should remain out of the launch promise unless implemented and tested.
- Run provider sandbox fixtures for redelivery, edits/deletes, bot self-messages, credential revocation, permission drift, and uninstall/reinstall identity stability.