Summary

• Problem: Telegram direct-message prompts can re-ingest messages that already belong to the active session history, causing duplicated transcript context and lower quality responses.
• Why it matters: direct chats are the most sensitive path for repeated assistant/user turns; duplicate context wastes prompt budget and can cause the model to answer stale turns.
• What changed: Telegram cached messages can now remember session keys, outbound Telegram replies are marked after successful dispatch, and direct-chat prompt context skips ordinary same-session messages while preserving reply targets.
• What did NOT change (scope boundary): group/supergroup ambient context is preserved, reply-chain targets remain selectable, core SDK/plugin contracts were not changed, and no changelog entry was edited.
• Contributor context: while working on an Oracle zip workflow, prompt/runtime overhead stood out as a practical performance issue; this PR is a small Telegram-side cleanup intended to help OpenClaw stay lean upstream.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes [Bug] Telegram conversation context duplicates session history, causing context pollution and degraded response quality #82040
• Related [Bug] Telegram conversation context duplicates session history, causing context pollution and degraded response quality #82040
• This PR fixes a bug or regression

Real behavior proof (required for external PRs)

• Behavior or issue addressed: a same-session Telegram message is recorded once in durable cache metadata and excluded from later direct-chat ambient context, unless it is needed as a reply target.
• Real environment tested: macOS local workstation, Node v24.14.1, pnpm v11.1.0, OpenClaw built from source with the Telegram extension runtime/cache harness.
• Exact steps or command run after this patch:
  • node scripts/run-vitest.mjs extensions/telegram/src/message-cache.test.ts extensions/telegram/src/bot-message-context.dm-session.test.ts extensions/telegram/src/bot-message-context.thread-binding.test.ts extensions/telegram/src/bot-handlers.runtime.test.ts extensions/telegram/src/bot.test.ts
  • pnpm build
  • git diff --check origin/main..HEAD
• Evidence after fix (screenshot, recording, terminal capture, console output, redacted runtime log, linked artifact, or copied live output):
  • Copied live terminal output from the local Telegram runtime/cache harness command:

$ node scripts/run-vitest.mjs extensions/telegram/src/message-cache.test.ts extensions/telegram/src/bot-message-context.dm-session.test.ts extensions/telegram/src/bot-message-context.thread-binding.test.ts extensions/telegram/src/bot-handlers.runtime.test.ts extensions/telegram/src/bot.test.ts
RUN  v4.1.6 /Users/jun/Developer/new/700_projects/cli-jaw/devlog/_plan/260515_codex_rs_openclaw_hermes_runtime/openclaw-pr-b-82040

Test Files  5 passed (5)
Tests  100 passed (100)
Duration  13.98s

• The persisted-cache regression scenario records a message, calls markSessionBound, resets/reloads the persisted cache bucket, and verifies the reloaded same-session message is deduped.
• The runtime private-chat scenario sends two private-chat messages in the same chat and verifies the second prompt no longer includes duplicate UntrustedStructuredContext.
• Build after final rebase: pnpm build completed successfully through write-cli-compat.
• Whitespace check after final rebase: git diff --check origin/main..HEAD produced no output.
• Observed result after fix: direct-chat prompt context excludes ordinary same-session cached messages after reload, while reply-target selection remains intact; the copied runtime/cache harness output shows 5 Telegram files and 100 scenarios passing with the new session-bound dedupe behavior.
• What was not tested: a live credentialed Telegram bot run against Telegram servers.
• Before evidence (optional but encouraged): [Bug] Telegram conversation context duplicates session history, causing context pollution and degraded response quality #82040 describes duplicated Telegram session history causing context pollution and degraded response quality.

Root Cause (if applicable)

• Root cause: Telegram cached messages did not carry session-bound metadata, so prompt context selection could not tell whether a cached Telegram message had already been represented in the active session transcript.
• Missing detection / guardrail: existing tests covered reply-chain and context selection, but not persisted same-session dedupe after outbound Telegram dispatch.
• Contributing context (if known): direct-message chats reuse the same chat window, so previous assistant/user messages are easy to pick up again as ambient context unless the active session boundary is explicit.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file:
  • extensions/telegram/src/message-cache.test.ts
  • extensions/telegram/src/bot.test.ts
• Scenario the test should lock in:
  • persisted sessionKeys survive reload and dedupe direct-chat ambient context
  • same-session reply targets are preserved when selected through a reply chain
  • a second private-chat prompt does not include duplicate same-session context
• Why this is the smallest reliable guardrail: the bug is inside Telegram-local cache/context selection, so focused cache and runtime handler tests cover the contract without requiring live Telegram credentials.
• Existing test that already covers this (if any): existing reply-chain tests covered target selection but not session-bound dedupe or persisted session keys.
• If no new test is added, why not: N/A, tests were added.

User-visible / Behavior Changes

Telegram direct chats should include less duplicate conversation context. Group/supergroup context and explicit reply targets should keep their previous behavior.

Diagram (if applicable)

Before:
outbound Telegram message -> cached with no session marker
next direct prompt -> ambient context can include same-session cached message again

After:
outbound Telegram message -> cached with sessionKeys[]
next direct prompt -> ordinary same-session cached message skipped
reply target -> still included when explicitly selected

Security Impact (required)

• New permissions/capabilities? (Yes/No): No
• Secrets/tokens handling changed? (Yes/No): No
• New/changed network calls? (Yes/No): No
• Command/tool execution surface changed? (Yes/No): No
• Data access scope changed? (Yes/No): No
• If any Yes, explain risk + mitigation: N/A

Repro + Verification

Environment

• OS: macOS
• Runtime/container: local Node v24.14.1, pnpm v11.1.0
• Model/provider: N/A
• Integration/channel (if any): Telegram extension
• Relevant config (redacted): no live Telegram secrets used for local verification

Steps

1. Build OpenClaw from source with pnpm build.
2. Run the focused Telegram cache/context/runtime test suite.
3. Confirm the diff is whitespace-clean with git diff --check origin/main..HEAD.

Expected

• Same-session direct-chat cached messages are deduped.
• Reply targets and reply-chain context are preserved.
• Group/supergroup context behavior remains unchanged.
• Tests and build pass.

Actual

• Telegram focused suite passed: 5 files / 100 tests.
• pnpm build passed.
• git diff --check origin/main..HEAD produced no output.

Evidence

Attach at least one:

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

• Verified scenarios:
  • persisted session keys survive cache reload
  • direct-chat same-session ambient context is deduped
  • reply targets are not dropped by dedupe
  • group/supergroup context keeps existing reply/ambient behavior
• Edge cases checked:
  • channel_post/supergroup path remains outside direct-chat dedupe
  • Telegram-local callback wiring stays confined to extensions/telegram/src
  • no src/plugin-sdk or core SDK files changed
• What you did not verify:
  • live Telegram bot behavior against Telegram servers

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

• Backward compatible? (Yes/No): Yes
• Config/env changes? (Yes/No): No
• Migration needed? (Yes/No): No
• If yes, exact upgrade steps: N/A

Risks and Mitigations

• Risk: overly broad dedupe could hide a message the user explicitly replied to.
  • Mitigation: same-session nodes are skipped only for ordinary ambient context; reply-target nodes are preserved and covered by tests.
• Risk: persisted cache files may contain older entries without sessionKeys.
  • Mitigation: parser treats the field as optional and defaults to normal existing behavior for older entries.