Switches the interactive chat input from inquire::Text to rustyline with
bracket paste mode enabled, so pasted multi-line text stays as a single
message instead of being split on each newline. Adds slash-command tab
completion via rustyline's Completer trait and a placeholder hint via
Hinter. History is now managed natively by rustyline with arrow-key
navigation.

https://claude.ai/code/session_017YeCUYbAKZqsGV4599Rrso

…5r' into fix/distri

…sion

- Replace "assistant:" label with "◆ agent_name:" using the current agent's
  display name instead of a generic role label
- Add /tools slash command to toggle tool call output on/off, letting users
  hide noisy tool blocks and focus on agent responses
- Suppress internal probe calls (load_skill("?"), discovery GET requests)
  from non-verbose output since they're agent internals
- Show thread ID in welcome header and on exit so users can easily resume
  sessions with /resume last or /resume <id>
- Show short thread ID when context is cleared via /clear

https://claude.ai/code/session_017YeCUYbAKZqsGV4599Rrso

…5r' into fix/distri

Use rustyline's ConditionalEventHandler to bind Ctrl+O as a toggle for
tool output visibility. The hint text updates to show "[tools hidden ·
Ctrl+O]" when tools are hidden. Shared state via Arc<AtomicBool> between
the helper (for hints) and the handler (for toggling). Help message now
has a dedicated KEYBOARD SHORTCUTS section.

https://claude.ai/code/session_017YeCUYbAKZqsGV4599Rrso

- Add --resume <id|last> flag to both `tui` and `run` commands so users
  can resume a thread directly from the CLI without entering interactive
  mode first
- Implement double Ctrl+C to exit: first press shows a hint, second
  press within 2 seconds exits. Ctrl+D still exits immediately
- Print thread message history when resuming (via --resume flag or
  /resume command), showing user prompts and assistant responses in a
  compact gray format so the user has context
- Add get_thread_messages() to the Distri client for fetching thread
  message history from the server

https://claude.ai/code/session_017YeCUYbAKZqsGV4599Rrso

…5r' into fix/distri

… gpt-5, o-series)

These models reject the legacy max_tokens parameter. Now dynamically selects
which parameter to use based on model name. Only one is ever sent per request.

* feat(types): add SkillRelation, path, and graph traversal methods to SkillStore trait

* feat(types): add path field to NewSkill and SkillListItem for hierarchy support

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(stores): add path and relations fields to SkillRecord in diesel_store

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(tools): add tag and path matching to tool_search

- tool_search now matches on tags and path fields (present on skill-backed tools)
- Include tags/path in search result JSON when available
- Add path field to AvailableSkill type for prompt rendering
- Orchestrator includes path in skill list when present: (id: `...`, path: ...)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(tools): add dry-run simulator — safe tools execute, unsafe tools get LLM-simulated responses

- simulator.rs: calls gpt-5-nano (Azure) to generate plausible tool responses
- Whitelist of safe tools (tool_search, load_skill, write_todos, search, final)
- Unsafe tools (api_request, connection_request, execute_shell, etc.) are simulated
- ExecutorContext gets dry_run: bool flag
- execute_tool_calls_with_timeout checks flag before each tool execution
- Simulated responses marked with _simulated: true

* feat(api): propagate dry_run flag from message metadata to ExecutorContext

Add dry_run field to ExecutorContextMetadata so clients can pass
{"dry_run": true} in the JSON-RPC message metadata. The A2A handler
reads this field and sets ExecutorContext.dry_run, enabling the eval
simulator to trigger simulated tool execution without side effects.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(simulator): intercept external tools (api_request) in dry-run mode before handle_external_tool_inline

External tools were bypassing dry-run check and going straight to
handle_external_tool_inline which waits for a client response that never
comes in eval mode. Now dry-run check happens BEFORE the external tool check.

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

…ill-hierarchy branch for testing

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…OAuth proxy

api_request now supports two modes:
- path: Distri platform API (existing)
- url + connection_id: external API with auto-injected OAuth token (new)

Agent calls: api_request({method: 'GET', url: 'https://googleapis.com/...', connection_id: '<id>'})
Tool proxies via POST /connections/{id}/request, token auto-injected.