Fix terminal interactivity by hjc-puro · Pull Request #4 · NousResearch/hermes-agent

hjc-puro · 2025-11-02T00:52:20Z

No description provided.

Fixes NousResearch#633 Problem: - Sequential numbering gaps (e.g., NousResearch#1, NousResearch#2, NousResearch#5, NousResearch#8) confuse users - 200 char truncation too aggressive - Tool messages completely hidden with no indication Fix: 1. Use separate counter for displayed messages only 2. Skip tool messages but show count at end 3. Skip system messages 4. Increase truncation to 300 chars 5. Display 'N tool messages hidden' summary Impact: - Consistent numbering: NousResearch#1, NousResearch#2, NousResearch#3, NousResearch#4 - Users know when tool calls occurred - More context visible per message

…ting fix(api): resolve model provider from config to prevent misrouting

- Task NousResearch#2: Remove Docker dependency (Critical priority) - Hermes runs natively via LocalEnvironment, no containers needed - Make Docker backend optional with graceful fallback to local - Update setup wizard to skip Docker on FreeBSD - Renumbered tasks: Setup Documentation → NousResearch#3, Voice Error Handling → NousResearch#4 - Added detailed acceptance criteria for each task - Programming assistant can now proceed with Docker removal

…agent loop reliability ## Summary Found 5 critical bugs in async error handling, context compression, and cron scheduling: **CRITICAL (2):** 1. Role violation after context compression (context_compressor.py:694-728) - Tool message validation missing when merging summary - Causes API crash and data loss after compression 2. Double-execution race condition in cron scheduler (scheduler.py:843-892) - File lock released before job execution completes - Allows duplicate jobs to be executed (DoS, duplicate messages) **HIGH (1):** 3. Unhandled context compression exceptions in main loop (run_agent.py:8204,8262,8338) - Silent crash when summarizer fails during API loop - No graceful degradation **MEDIUM (2):** 4. Error swallowing in auxiliary_client (auxiliary_client.py:2074-2106) - Original error overwritten on retry failure - Lost error context, unreachable fallback logic 5. Session ID change without exception recovery (run_agent.py:6041-6071) - Session state corruption on DB failures - Broken session lineage ## Details Full analysis with code snippets, scenarios, and fixes in: - AUDIT_ITERATION_2.md (400 lines, detailed technical analysis) - AUDIT_ITERATION_2_SUMMARY.txt (visual summary, testing recommendations) ## Recommended Priority 1. Bug NousResearch#1 (Role Violation) - FIX IMMEDIATELY 2. Bug NousResearch#2 (Double Execution) - FIX IMMEDIATELY 3. Bug NousResearch#3 (Unhandled Exceptions) - FIX SOON 4. Bug NousResearch#4 & NousResearch#5 - FIX AFTER critical bugs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

feat(ux): Phase 3.1 - navigation hotkeys

- Add fetchAgentFullTranscript() — fetches up to 200 messages per session - Enrichment step now pulls all assistant messages, not just the last one - Reports include full agent output with tool call annotations - Transcript summary generated from last 3 assistant messages - MissionAgentSummary type extended with transcript + transcriptSummary

#3 Cost Analytics Dashboard - New Analytics tab with summary cards (total tokens, cost, avg/mission, today, week) - CSS bar charts: cost by agent, cost by model, daily timeline (7d) - No external chart libraries — pure Tailwind #4 Export Mission as Markdown - Download .md file with full mission report (goal, team, transcript, artifacts) - Copy to clipboard button with visual feedback - Wired into Mission Detail Overlay #5 Word-by-word Streaming in Agent Chat - Replaced polling with SSE EventSource in AgentChatPanel - Real-time chunk streaming with fallback to polling on error - Streaming assistant message updates in-place NousResearch#6 Remote Agents Panel - Fetches external sessions from gateway /api/sessions - Filters out local agent sessions — shows only remote/external - Auto-polls every 15s, card layout with status, model, tokens, cost - Open Chat links to ClawSuite chat tab NousResearch#7 Real-time Collaboration (Presence) - BroadcastChannel-based cross-tab presence detection - Shows colored avatars of other users viewing Agent Hub - Heartbeat every 5s, stale cleanup at 30s - Shows which tab each peer is viewing

NousResearch#23) The inspector panel already uses relative /api/memory paths (fixed in main), but the parent GET route was missing — requests fell through to the client router and returned HTML instead of JSON. This adds a proper TanStack createFileRoute handler that proxies memory requests through the workspace server with auth, so the inspector works correctly over Tailscale, LAN, and remote access. Rebased from PR #4 onto current main.

…re method to SessionDB

…t, stop/undo honesty, json_error crash, codex validation, deep-link race Bug #1: ChatPage loadSession reads res.items (not res.transcript) to match backend Bug NousResearch#2: Add GET /api/gui/session-search backed by SessionDB.search_messages (FTS5) Bug NousResearch#3: Stop button now checks res.supported before claiming run was stopped Bug NousResearch#4: Undo button now checks res.supported before removing messages locally Bug NousResearch#5: Fix _json_error positional calls in handle_chat_compress (was crashing 500) Bug NousResearch#6: Codex provider validation now also guards switching TO openai-codex Bug NousResearch#7: Deep-link hash check runs before health callback to prevent race condition

- connection.py: cap header read at 8KB to prevent DoS from malicious handler - handler.py: use .find() instead of `in` + .index() to eliminate race in patch - handler.py: add truncated field to execute response when output exceeds 50KB - server.py: include error data field in formatted error messages - test: add timeout to test client recv, handle TimeoutExpired in close Fixes issues NousResearch#1, NousResearch#4, NousResearch#5, NousResearch#6, NousResearch#8, NousResearch#10 from Qwen 3.5 peer review on PR NousResearch#19. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The adapter's outbound send_reaction() (patch NousResearch#3) was reachable from gateway code but invisible to the agent — no tool surface, and the inbound message_handle (already on MessageEvent / SessionSource) was never injected into the model-visible session-context block. - gateway/session.py: Platform.SENDBLUE branch with iMessage notes plus a triggering-message-handle IDs block (current turn only). - tools/send_message_tool.py: send_message gains action='react' with message_handle / reaction params, _handle_react validation arms, and a _react_sendblue dispatcher that mirrors _send_sendblue's throwaway- adapter pattern. _REACTION_PLATFORMS gates platform support. - agent/prompt_builder.py: sharpen the sendblue blurb to point at the now-real send_message(action='react', ...) mechanism and the handle source (the original tapback line from patch NousResearch#4 was aspirational). - tests/tools/test_sendblue_reaction.py: 15 tests covering dispatch, validation arms, schema, and reaction normalization. - LOCAL_PATCHES.md: new patch NousResearch#11 entry + file-index updates.

New aeyeops/googlechat/streaming-spec.md covering three post-DEMO-1 UX gaps: (1) no acknowledgement before first LLM token, (2) no progressive delivery, (3) GFM markdown rendering as literal characters. Spec pins adapter-local wiring onto existing gateway/base-class hooks (send_typing, send, edit_message with finalize=, REQUIRES_EDIT_FINALIZE) so the work stays within ADR-003's I/O-only scope. Verified API surface 2026-04-20: spaces.messages.patch supports updateMask=text OR cardsV2, 1 write/sec per space cap, cardsV2.textParagraph HTML subset (b/i/u/s/font color/a/br/code/pre/ul/ol/li/time). Design: three-phase one-message lifecycle — placeholder create on send_typing, progressive patch updateMask=text during streaming, content-aware finalize routing to either Chat markdown (simple) or cardsV2 HTML (nested lists, language-tagged code). Stream consumer cadence (1s / 40 chars) already matches Chat's rate cap. Also: M6 section appended to build-plan.md (C29-C34 commit sequence + DEMO NousResearch#4), C25's no-op disposition marked superseded, roadmap R1 stale 'probably C15' reference updated to reflect the new sequence, and README gains a docs index.

Codex round-10 adversarial review finding NousResearch#4 (HIGH): a pre_memory_write guard plugin that raises during validation was silently logged by the generic ``invoke_hook`` (per-callback try/except) and the helper saw zero results — interpreted as "no block", so the memory write proceeded. Contradicts the D-013 fail-closed intent (which only handled ImportError at the call sites). Implementation: * PluginManager.invoke_hook now accepts a keyword-only ``strict`` flag. When True, a raised callback surfaces a synthetic ``{"action": "block", "message": ...}`` result instead of being silently logged. Other callbacks still run so independent guards retain their vote. * get_pre_memory_write_block_message invokes with ``strict=True``. Hook exceptions become a conservative block message that names the callback, exception type, and exception message. Existing block-flow tests that mock ``invoke_hook`` are unaffected (the wrapper still goes through invoke_hook). Tests (red → green): * test_helper_blocks_when_hook_callback_raises — single broken guard fails closed. * test_helper_blocks_when_one_callback_raises_alongside_others — a raised callback is not masked by a well-behaved observer. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`hermes gateway install --system` previously bailed out with "Service installation not supported on this platform" on Windows because the install path was systemd/launchd only and depended on pwd/grp/geteuid that don't exist there. Add a Windows Service Control Manager backend so the gateway runs unattended, survives reboots, and is administered with the same install/uninstall/start/stop/restart/status verbs as on Linux/macOS. gateway/_windows_service.py defines a HermesGatewayService subclass of win32serviceutil.ServiceFramework. SvcDoRun spawns `hermes gateway run` as a child process in CREATE_NEW_PROCESS_GROUP and supervises it; SvcStop sends CTRL_BREAK_EVENT so the gateway's existing SIGBREAK handler (added in c5a31ba) drives the same graceful shutdown sequence systemd's SIGTERM hits on Linux. The child is wrapped in the kill-tree Job Object (9e173f6) so a hung gateway can't leave orphan subprocesses. The CLI helpers `windows_service_{install,uninstall,start,stop, restart,status}` in hermes_cli/gateway.py replace the systemd_*/launchd_* counterparts on Windows, and the dispatch in `_gateway_command_inner` now branches on `is_windows()` alongside the existing platform checks. `_is_service_installed()` queries SCM via OpenService when on Windows. User identity replacements for NousResearch#4: * is_admin() — ctypes.windll.shell32.IsUserAnAdmin (replaces os.geteuid()==0) * get_current_username() — win32api.GetUserName (replaces pwd.getpwuid) `gateway install --system` accepts `--run-as-user`, `--password`, and `--startup` on Windows. Without `--run-as-user`, the service runs as LocalSystem (the SCM default). POSIX behavior is unchanged: --password and --startup are ignored on those platforms per their help text. Service environment (HERMES_BIN, HERMES_PROFILE, PYTHONUNBUFFERED) is persisted under HKLM\SYSTEM\CurrentControlSet\Services\HermesGatewayEnvironment as REG_MULTI_SZ — SCM merges this into the spawned process's env automatically, so SvcDoRun finds the right binary after reboot without a config file lookup race. Tests added: 17 unit tests with mocked pywin32 calls (never touch the real SCM). Covers install with default LocalSystem, run-as-user requires password, run-as-user passthrough, invalid startup rejected, uninstall stops running service first, query_service_status SCM-code translation, lifecycle dispatch, and HermesGatewayService class metadata + command/env builders. Also: agent/copilot_acp_client.py last-resort HOME fallback now uses tempfile.gettempdir() instead of hardcoded "/tmp" so it works on Windows where /tmp doesn't exist. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* docs: add downstream fork maintenance workflow * docs: remove mixed-version registry delta --------- Co-authored-by: asdigitos <274687674+asdigitos@users.noreply.github.com>

- Task NousResearch#2: Remove Docker dependency (Critical priority) - Hermes runs natively via LocalEnvironment, no containers needed - Make Docker backend optional with graceful fallback to local - Update setup wizard to skip Docker on FreeBSD - Renumbered tasks: Setup Documentation → NousResearch#3, Voice Error Handling → NousResearch#4 - Added detailed acceptance criteria for each task - Programming assistant can now proceed with Docker removal

Anthropic's Messages API accepts a first-class `document` content block so Claude can read PDFs (text + images + layout) natively instead of requiring caller-side text extraction. Available on direct Anthropic, Vertex AI, and AWS Bedrock via the AnthropicBedrock SDK. Hermes's `anthropic_messages` api_mode routes Opus 4.7 / Sonnet 4.6 / Haiku 4.5 through this path (on both native and Bedrock), so document support just needed wiring on the message converter. Before: `_convert_content_part_to_anthropic` only recognized text and image parts. A content part like `{"type": "pdf", "document_url": "https://arxiv.org/pdf/..."}` would pass through as an unrecognized block and either break the request or be silently dropped by the SDK. After: two new entry points. 1. `_document_source_from_part` — a source-shape normalizer that accepts every practical input form: * Native Anthropic `source` dict passthrough (base64/url/text/content) * Shorthand string: URL, file path, or `data:application/pdf;base64,...` * Shorthand dict: `{url: ...}`, `{base64: ...}`, `{file_path: ...}`, `{text: ...}`, `{data, media_type}` combos File-path reading is intentionally fail-loud — a silent drop of a PDF attachment would be worse than a clear IOError. 2. `_convert_content_part_to_anthropic` recognizes `type="document"` (native passthrough, preserving `title`, `context`, `citations`, `cache_control`) and four convenience aliases (`pdf`, `file`, `input_file`, `document_url`) that map to the same normalized `{type: document, source: ..., ...}` envelope. Composes correctly with existing features: * `cache_control` markers on document blocks preserve through the converter — large static PDFs can be cached with prompt caching. * Document + image + text blocks in the same user message all convert (verified end-to-end via `convert_messages_to_anthropic`). * Works on AnthropicBedrock since the SDK shares the same content block types (verified: anthropic 0.96.0 ships `DocumentBlockParam`, `Base64PDFSourceParam`, `URLPDFSourceParam`, `PlainTextSourceParam`, `ContentBlockSourceParam`). 29 new tests in tests/agent/test_anthropic_document_blocks.py covering every input shape, metadata preservation (title/context/citations), cache_control pass-through, malformed-part rejection, the four alias types, and end-to-end message conversion with mixed document/image/text content. Closes gap NousResearch#4 identified in the Phase 2 re-verification (PraxVault/Hermes/Reference/Decisions/bedrock-phase2-audit/04-current-architecture).

- Task NousResearch#2: Remove Docker dependency (Critical priority) - Hermes runs natively via LocalEnvironment, no containers needed - Make Docker backend optional with graceful fallback to local - Update setup wizard to skip Docker on FreeBSD - Renumbered tasks: Setup Documentation → NousResearch#3, Voice Error Handling → NousResearch#4 - Added detailed acceptance criteria for each task - Programming assistant can now proceed with Docker removal

hjc-puro added 2 commits November 2, 2025 08:52

fix terminal interactivity

faecbdd

terminal tool

a6ec797

teknium1 merged commit d4544f0 into main Nov 2, 2025

teknium1 mentioned this pull request Mar 5, 2026

Feature: Enhanced Extension System with Tool Interception & Lifecycle Events (inspired by Pi) #359

Open

erosika mentioned this pull request Mar 11, 2026

feat(honcho): async writes, memory modes, session title integration, setup CLI #736

Merged

13 tasks

heyalchang mentioned this pull request Mar 12, 2026

feat: payload visualization dashboard heyalchang/hermes-agent#1

Merged

6 tasks

This was referenced Mar 14, 2026

fix(cli): non-blocking startup update check, version consistency, and banner deduplication #1257

Closed

fix(cli): non-blocking startup update check, version consistency, and banner deduplication #1266

Closed

SHL0MS mentioned this pull request Mar 31, 2026

[UX] Quick commands not shown in /help output #4090

Open

iRonin mentioned this pull request Apr 3, 2026

feat: Alt+Enter queues follow-up messages without interrupting agent #4788

Open

intertwine mentioned this pull request Apr 3, 2026

[codex] add observational memory provider #4787

Closed

This was referenced Apr 4, 2026

fix: DingTalk platform adapter has multiple bugs preventing message processing #5037

Closed

fix: DingTalk platform adapter has multiple bugs preventing message processing #5038

Closed

lucaspirola mentioned this pull request Apr 5, 2026

feat: Rich Markdown rendering with skin-aware themes and /markdown toggle #5150

Closed

4 tasks

sudo-yf pushed a commit to sudo-yf/hermes-agent that referenced this pull request Apr 5, 2026

Merge pull request NousResearch#4 from deboste/fix/model-provider-rou…

b0da4db

…ting fix(api): resolve model provider from config to prevent misrouting

aaronlab mentioned this pull request Apr 9, 2026

fix: error context preservation, WAL checkpoint, hook timeout #6622

Open

5 tasks

h4x3rotab referenced this pull request in Clawdi-AI/hermes-agent Apr 10, 2026

Merge pull request #4 from outsourc-e/phase3.1-hotkeys

8e4223d

feat(ux): Phase 3.1 - navigation hotkeys

Vex-Dravex added a commit to Vex-Dravex/hermes-agent that referenced this pull request Apr 10, 2026

ralph: story NousResearch#4 — Add session_checkpoints table and captu…

7211fa1

…re method to SessionDB

konsisumer mentioned this pull request Apr 11, 2026

fix(gateway): break stuck session resume loops on restart #7747

Closed

3 tasks

Vex-Dravex added a commit to Vex-Dravex/hermes-agent that referenced this pull request Apr 11, 2026

ralph: story NousResearch#4 — Add session_checkpoints table and captu…

54efc2c

…re method to SessionDB

maiixu mentioned this pull request Apr 12, 2026

fix(gateway): show 🎙 transcript prefix in streaming mode #8644

Closed

trevorgordon981 mentioned this pull request Apr 13, 2026

fix(bluebubbles): repair same-host webhook flow and iMessage sanitization #9272

Open

5 tasks

liyoungc mentioned this pull request May 2, 2026

feat(memory): Hermes V3 long-term memory — W1 + W2 (sqlite_vec + import + read + memdebug) liyoungc/hermes-agent#2

Merged

pty819 mentioned this pull request May 3, 2026

[continuing #19101] AGENTS.md gaps round 2: missing auxiliary config, unverified --now flag, stale PR refs, more #19107

Open

arsitekberotot mentioned this pull request May 4, 2026

[Bug]: OpenAI-Codex credential pool can drop newly added credential after stale auth.json rewrite during rotation #19566

Open

1 task

jbellsolutions mentioned this pull request May 6, 2026

Dashboard Chat tab unusable in Docker image — four compounding bugs (ui-tui chown, /opt/data file chown, HOME inheritance, ink-bundle staleness loop) #20739

Open

ygd58 mentioned this pull request May 6, 2026

[Bug/Architecture] Severe context loss, truncation-overwrites, and memory limitations during complex coding workflow #20849

Open

ZK-Snarky mentioned this pull request May 9, 2026

fix(telegram): fix 4 streaming truncation bugs — safe_limit, cursor cleanup, metadata forwarding, overflow chunking #5270

Closed

teknium1 mentioned this pull request May 10, 2026

Skill auto-creation learns from transient failures, causing persistent tool avoidance (learned helplessness) #6051

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix terminal interactivity#4

Fix terminal interactivity#4
teknium1 merged 2 commits into
mainfrom
fix-terminal

hjc-puro commented Nov 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

hjc-puro commented Nov 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants