Skip to content

CHANGELOG.md

Latest commit

 

History

History
114 lines (100 loc) · 5.57 KB

CHANGELOG.md

File metadata and controls

114 lines (100 loc) · 5.57 KB

Changelog

All notable changes to cli-agents are documented in this file.

[0.1.1] - 2026-03-18

Fixed

  • claude-agent/hooks/forward_to_session.py
    • now resolves the active tmux pane instead of assuming :0.0
    • now waits for a ready Claude prompt before sending input, reducing dropped first-message cases during startup
    • now recognizes the current ❯ Claude prompt in addition to older prompt variants
  • claude-agent/hooks/start_claude.sh and codex-agent/hooks/start_codex.sh
    • now send one extra empty Enter shortly after startup by default
    • this helps unblock first-run prompts such as trusting a newly entered workdir
  • claude-agent/hooks/pane_monitor.sh
    • now treats the current Claude ❯ prompt as an idle-state indicator during monitor state detection
  • codex-agent/hooks/forward_to_session.py
    • now mirrors the same active-pane and ready-state forwarding behavior for Codex sessions
  • claude-agent/hooks/on_complete.py and codex-agent/hooks/on_complete.py
    • now extract outbound messageId from real openclaw message send --json output instead of assuming a top-level field
  • claude-agent/hooks/pane_monitor.sh and codex-agent/hooks/pane_monitor.sh
    • now use the shared message-id extractor so monitor-generated notifications also write reply-route mappings reliably

Added

  • bridge/extract_message_id.py
    • shared helper that recursively extracts messageId / message_id from nested OpenClaw plugin JSON output

Fixed

  • codex-agent/hooks/pane_monitor.sh
    • now resets HAS_PROMPT=0 on every monitor loop iteration so prompt state does not leak across checks and misclassify working -> idle transitions

Verified

  • End-to-end Discord thread routing now works for the claude-agent demo flow:
    • thread message -> tmux session
    • Claude reply -> Discord thread
    • reply message id -> reply route map
    • Discord reply-to -> original tmux session via reply-target-message-id
  • claude-agent/hooks/pane_monitor.sh was checked for the same state-reset issue and already resets HAS_PROMPT correctly each loop.

[0.1.0] - 2026-03-16

Initial consolidated cli-agents repository release.

Added

  • Consolidated repository layout:
    • codex-agent/
    • claude-agent/
    • bridge/
  • Shared bridge utilities:
    • bridge/agent_session_router.py
    • bridge/reply_route_map.py
    • bridge/check_agent_hooks.py
    • bridge/check_route_conflicts.py
  • Repository-level README.md
  • Repository-level .gitignore
  • Repository-level acknowledgements to upstream projects

Changed vs upstream dztabel-happy/codex-agent

Source: https://github.com/dztabel-happy/codex-agent

  • Added session route context support via per-session route files
  • Added reply-target mapping so outbound message IDs can route future thread replies back to the correct tmux session
  • Added thread-aware forwarding helper: hooks/forward_to_session.py
  • Added start_codex_openclaw.sh for explicit OpenClaw source-context startup
  • Added doctor.sh and bridge-level hook preflight / route conflict checks
  • start_codex.sh
    • now writes route metadata (chat_id, channel, account, agent_name, workdir, trace_id)
    • now performs hook preflight and cross-agent route conflict checks before launch
    • now supports managed source routing through environment variables
    • now defaults to yolo/auto mode unless --approval is specified
  • on_complete.py
    • now ignores unmanaged sessions
    • now resolves route context dynamically
    • now stores reply-route mappings for completion notifications
    • now supports channel/account-aware delivery instead of a single hardcoded target model
  • pane_monitor.sh
    • now routes approval notifications through per-session context
    • now exits early for unmanaged sessions
    • now supports account-aware outbound messaging
  • stop_codex.sh
    • now sends session-close notifications using route context
    • now removes per-session route files on cleanup
  • Documentation updated for skills/... installation paths and managed thread routing workflow

Changed vs upstream N1nEmAn/claude-agent

Source: https://github.com/N1nEmAn/claude-agent

  • Added shared bridge integration with codex-agent-compatible routing model
  • Added thread-aware forwarding helper: hooks/forward_to_session.py
  • Added per-session route file support via hooks/route_context.py
  • Added bridge-level hook preflight / route conflict checks
  • Added doctor.sh
  • start_claude.sh
    • now supports explicit per-session flags for --chat-id, --channel, --account, --agent
    • now writes per-session route metadata
    • now performs preflight and conflict detection before launch
    • now defaults to auto mode unless --approval is specified
  • on_complete.py
    • now resolves route context dynamically instead of relying only on global env
    • now stores reply-route mappings for completion notifications
    • now supports account-aware delivery and cleaner thread-targeted wake behavior
  • pane_monitor.sh
    • now detects work-start transitions and can proactively notify the bound thread
    • now stores reply-route mappings for monitor-generated messages
    • now uses bridge-local reply_route_map.py
  • stop_claude.sh
    • now sends session-close notifications using route context
    • now removes per-session route files on cleanup
  • Documentation updated for the consolidated repository layout and multi-session thread routing workflow

Repository maintenance

  • Removed nested .git directories from the old split repos
  • Created a single standalone Git repository for cli-agents
  • Cleaned __pycache__ / *.pyc from version control and added ignore rules